Px Full Res x_full_px\",\n \"locations on the section -> semi-quantiative, local, analysis def parse_cellcounter_to_dict(fpath): '''Parse Cell-Counter Xml\",\n \"float(tree.find('./Image_Properties/Y_Calibration').text) rt = tree.find('Marker_Data') #re-root the tree for type_ in rt.iter('Marker_Type'): cells =\",\n \"os.listdir(base_) if f[-3:] == 'xml' and f[0] != '.'] #Work through each xml\",\n \"#Loop through Cells and Annotate. for cell_type, cell_label in types_of_interest.iteritems(): print 'Working on\",\n \"-> Px Full Res x_full_px = vals['counter'][cell_type][i][0] * vals['counter']['x_cal'] #in um y_full_px =\",\n \"file: ' + os.path.split(xml_)[1] + '...' + str(e+1) + '/' + str(len(xml_files)) #Find\",\n \"xml.etree.ElementTree as ET import csv import os import re from ij import IJ\",\n \"f[0] != '.'] #Work through each xml file f_out_path = '/path/to/annotation/out.tsv' with open(f_out_path,'w')\",\n \"= '/'.join(os.path.split(xml_)[0].split('/')[:-1]) files_found = [os.path.join(root, f) for (root, dirs, files) in os.walk(search_dir) for\",\n \"of extracting cell -> small image. anim, vals = input_dict #Loop through Cells\",\n \"image. anim, vals = input_dict #Loop through Cells and Annotate. for cell_type, cell_label\",\n \"Full Res x_full_px = vals['counter'][cell_type][i][0] * vals['counter']['x_cal'] #in um y_full_px = vals['counter'][cell_type][i][1] *\",\n \"import xml.etree.ElementTree as ET import csv import os import re from ij import\",\n \"be used in subsequent analysis to cluster cells from similar locations on the\",\n \"were passed to extract_cells input_item = (re.search('(?<=_).*',orig_f_name[:-4]).group(), {'fullres':fullres_image, 'counter':parse_cellcounter_to_dict(xml_)}) input_dict = input_item types_of_interest={'7':'tdtom','8':'gfp'}\",\n \"cell -> small image. anim, vals = input_dict #Loop through Cells and Annotate.\",\n \"= [os.path.join(root, f) for (root, dirs, files) in os.walk(search_dir) for f in files\",\n \"int(marker_[1].text))) # cells_dict[type_.find('Type').text] = cells return cells_dict #Load Xml Files xml_locs = ['/path/to/xml/files']\",\n \"cells = [] for marker_ in type_.iter('Marker'): cells.append((int(marker_[0].text), int(marker_[1].text))) # cells_dict[type_.find('Type').text] = cells\",\n \"xml_files = [os.path.join(base_, f) for base_ in xml_locs for f in os.listdir(base_) if\",\n \"the orig .nd2 file, copied from find_extract_cells.py, see that code for more details.\",\n \"cells.append((int(marker_[0].text), int(marker_[1].text))) # cells_dict[type_.find('Type').text] = cells return cells_dict #Load Xml Files xml_locs =\",\n \"#Work through each xml file f_out_path = '/path/to/annotation/out.tsv' with open(f_out_path,'w') as fout: fout.write('\\\\t'.join(['cell','x_um','y_um']))\",\n \"that were passed to extract_cells input_item = (re.search('(?<=_).*',orig_f_name[:-4]).group(), {'fullres':fullres_image, 'counter':parse_cellcounter_to_dict(xml_)}) input_dict = input_item\",\n \"[os.path.join(root, f) for (root, dirs, files) in os.walk(search_dir) for f in files if\",\n \"ij import IJ from loci.plugins.in import ImporterOptions from loci.plugins import BF from ij.plugin\",\n \"vals = input_dict #Loop through Cells and Annotate. for cell_type, cell_label in types_of_interest.iteritems():\",\n \"in rt.iter('Marker_Type'): cells = [] for marker_ in type_.iter('Marker'): cells.append((int(marker_[0].text), int(marker_[1].text))) # cells_dict[type_.find('Type').text]\",\n \"import BF from ij.plugin import ImagesToStack from ij import io #Records metadata (x,y\",\n \"semi-quantiative, local, analysis def parse_cellcounter_to_dict(fpath): '''Parse Cell-Counter Xml file to Dictionary Inputs: fpath\",\n \"path to xml file on disk Values: (dict). Keys 'x_cal', 'y_cal' = (float)\",\n \"with open(f_out_path,'w') as fout: fout.write('\\\\t'.join(['cell','x_um','y_um'])) for e,xml_ in enumerate(xml_files): print 'Working on file:\",\n \"= (re.search('(?<=_).*',orig_f_name[:-4]).group(), {'fullres':fullres_image, 'counter':parse_cellcounter_to_dict(xml_)}) input_dict = input_item types_of_interest={'7':'tdtom','8':'gfp'} #Copied from the \\\"Extract Cells\\\",\",\n \"the tree for type_ in rt.iter('Marker_Type'): cells = [] for marker_ in type_.iter('Marker'):\",\n \"\\\"Could not find fullres image.\\\" raise ValueError('Found 0 or >1 matching file') #Generate\",\n \"for type_ in rt.iter('Marker_Type'): cells = [] for marker_ in type_.iter('Marker'): cells.append((int(marker_[0].text), int(marker_[1].text)))\",\n \"'/'.join(os.path.split(xml_)[0].split('/')[:-1]) files_found = [os.path.join(root, f) for (root, dirs, files) in os.walk(search_dir) for f\",\n \"(float) calibrations in each axis. Keys '1'-'8' = (lists) of tuples containing cell\",\n \"== orig_f_name] if len(files_found) == 1: fullres_image = files_found[0] else: print \\\"Could not\",\n \"as used in find_extract_cells xml_files = [os.path.join(base_, f) for base_ in xml_locs for\",\n \"'y_cal' = (float) calibrations in each axis. Keys '1'-'8' = (lists) of tuples\",\n \"out_title = '_'.join([anim, cell_label, str(i)]) fout.write('\\\\n' + '\\\\t'.join([out_title, str(x_full_px), str(y_full_px)])) #Final tsv of\",\n \"tree = ET.parse(fpath) cells_dict = {} cells_dict['x_cal'] = float(tree.find('./Image_Properties/X_Calibration').text) cells_dict['y_cal'] = float(tree.find('./Image_Properties/Y_Calibration').text) rt\",\n \"and f[0] != '.'] #Work through each xml file f_out_path = '/path/to/annotation/out.tsv' with\",\n \"files) in os.walk(search_dir) for f in files if f == orig_f_name] if len(files_found)\",\n \"#in um y_full_px = vals['counter'][cell_type][i][1] * vals['counter']['y_cal'] #in um #Write Information out_title =\",\n \"cells that were extracted with 1_find_extract_cells.py #metadata will be used in subsequent analysis\",\n \"i in range(len(vals['counter'][cell_type])): print 'Iteration ' + str(i+1) + '/' + str(len(vals['counter'][cell_type])) #Convert\",\n \"# cells_dict[type_.find('Type').text] = cells return cells_dict #Load Xml Files xml_locs = ['/path/to/xml/files'] #same\",\n \"on cell_type ' + cell_label for i in range(len(vals['counter'][cell_type])): print 'Iteration ' +\",\n \"file on disk Values: (dict). Keys 'x_cal', 'y_cal' = (float) calibrations in each\",\n \"in the form (x,y) ''' tree = ET.parse(fpath) cells_dict = {} cells_dict['x_cal'] =\",\n \"import csv import os import re from ij import IJ from loci.plugins.in import\",\n \"that were extracted with 1_find_extract_cells.py #metadata will be used in subsequent analysis to\",\n \"+ cell_label for i in range(len(vals['counter'][cell_type])): print 'Iteration ' + str(i+1) + '/'\",\n \"fullres_image = files_found[0] else: print \\\"Could not find fullres image.\\\" raise ValueError('Found 0\",\n \"+ str(len(xml_files)) #Find the orig .nd2 file, copied from find_extract_cells.py, see that code\",\n \"writing to disk instead of extracting cell -> small image. anim, vals =\",\n \"f) for (root, dirs, files) in os.walk(search_dir) for f in files if f\",\n \"in type_.iter('Marker'): cells.append((int(marker_[0].text), int(marker_[1].text))) # cells_dict[type_.find('Type').text] = cells return cells_dict #Load Xml Files\",\n \"in os.walk(search_dir) for f in files if f == orig_f_name] if len(files_found) ==\",\n \"= vals['counter'][cell_type][i][0] * vals['counter']['x_cal'] #in um y_full_px = vals['counter'][cell_type][i][1] * vals['counter']['y_cal'] #in um\",\n \"file, copied from find_extract_cells.py, see that code for more details. orig_f_name = re.search('(?<=CellCounter_).*(?=\\\\\\\\-Downsampled)',\",\n \"+ '...' + str(e+1) + '/' + str(len(xml_files)) #Find the orig .nd2 file,\",\n \"for base_ in xml_locs for f in os.listdir(base_) if f[-3:] == 'xml' and\",\n \"in find_extract_cells xml_files = [os.path.join(base_, f) for base_ in xml_locs for f in\",\n \"enumerate(xml_files): print 'Working on file: ' + os.path.split(xml_)[1] + '...' + str(e+1) +\",\n \"dirs, files) in os.walk(search_dir) for f in files if f == orig_f_name] if\",\n \"== 1: fullres_image = files_found[0] else: print \\\"Could not find fullres image.\\\" raise\",\n \"'Working on file: ' + os.path.split(xml_)[1] + '...' + str(e+1) + '/' +\",\n \"in os.listdir(base_) if f[-3:] == 'xml' and f[0] != '.'] #Work through each\",\n \"IJ from loci.plugins.in import ImporterOptions from loci.plugins import BF from ij.plugin import ImagesToStack\",\n \"file f_out_path = '/path/to/annotation/out.tsv' with open(f_out_path,'w') as fout: fout.write('\\\\t'.join(['cell','x_um','y_um'])) for e,xml_ in enumerate(xml_files):\",\n \"1: fullres_image = files_found[0] else: print \\\"Could not find fullres image.\\\" raise ValueError('Found\",\n \"cells from similar locations on the section -> semi-quantiative, local, analysis def parse_cellcounter_to_dict(fpath):\",\n \"'''Parse Cell-Counter Xml file to Dictionary Inputs: fpath (str) path to xml file\",\n \"+ '/' + str(len(vals['counter'][cell_type])) #Convert Px Downsampled -> Px Full Res x_full_px =\",\n \"+ str(i+1) + '/' + str(len(vals['counter'][cell_type])) #Convert Px Downsampled -> Px Full Res\",\n \"not find fullres image.\\\" raise ValueError('Found 0 or >1 matching file') #Generate the\",\n \"axis. Keys '1'-'8' = (lists) of tuples containing cell positions in the form\",\n \"Dictionary Inputs: fpath (str) path to xml file on disk Values: (dict). Keys\",\n \"disk Values: (dict). Keys 'x_cal', 'y_cal' = (float) calibrations in each axis. Keys\",\n \"for i in range(len(vals['counter'][cell_type])): print 'Iteration ' + str(i+1) + '/' + str(len(vals['counter'][cell_type]))\",\n \"marker_ in type_.iter('Marker'): cells.append((int(marker_[0].text), int(marker_[1].text))) # cells_dict[type_.find('Type').text] = cells return cells_dict #Load Xml\",\n \"from loci.plugins import BF from ij.plugin import ImagesToStack from ij import io #Records\",\n \"<filename>imageproc_OE_IF_quant/2_annotate_extracted_cells.py import xml.etree.ElementTree as ET import csv import os import re from ij\",\n \"f_out_path = '/path/to/annotation/out.tsv' with open(f_out_path,'w') as fout: fout.write('\\\\t'.join(['cell','x_um','y_um'])) for e,xml_ in enumerate(xml_files): print\",\n \"Px Full Res x_full_px = vals['counter'][cell_type][i][0] * vals['counter']['x_cal'] #in um y_full_px = vals['counter'][cell_type][i][1]\",\n \"ET.parse(fpath) cells_dict = {} cells_dict['x_cal'] = float(tree.find('./Image_Properties/X_Calibration').text) cells_dict['y_cal'] = float(tree.find('./Image_Properties/Y_Calibration').text) rt = tree.find('Marker_Data')\",\n \"'/' + str(len(xml_files)) #Find the orig .nd2 file, copied from find_extract_cells.py, see that\",\n \"= ['/path/to/xml/files'] #same as used in find_extract_cells xml_files = [os.path.join(base_, f) for base_\",\n \"xml_locs = ['/path/to/xml/files'] #same as used in find_extract_cells xml_files = [os.path.join(base_, f) for\",\n \"to extract_cells input_item = (re.search('(?<=_).*',orig_f_name[:-4]).group(), {'fullres':fullres_image, 'counter':parse_cellcounter_to_dict(xml_)}) input_dict = input_item types_of_interest={'7':'tdtom','8':'gfp'} #Copied from\",\n \"[os.path.join(base_, f) for base_ in xml_locs for f in os.listdir(base_) if f[-3:] ==\",\n \"on disk Values: (dict). Keys 'x_cal', 'y_cal' = (float) calibrations in each axis.\",\n \"inputs that were passed to extract_cells input_item = (re.search('(?<=_).*',orig_f_name[:-4]).group(), {'fullres':fullres_image, 'counter':parse_cellcounter_to_dict(xml_)}) input_dict =\",\n \"import ImporterOptions from loci.plugins import BF from ij.plugin import ImagesToStack from ij import\",\n \"len(files_found) == 1: fullres_image = files_found[0] else: print \\\"Could not find fullres image.\\\"\",\n \"#Write Information out_title = '_'.join([anim, cell_label, str(i)]) fout.write('\\\\n' + '\\\\t'.join([out_title, str(x_full_px), str(y_full_px)])) #Final\",\n \"original inputs that were passed to extract_cells input_item = (re.search('(?<=_).*',orig_f_name[:-4]).group(), {'fullres':fullres_image, 'counter':parse_cellcounter_to_dict(xml_)}) input_dict\",\n \"import IJ from loci.plugins.in import ImporterOptions from loci.plugins import BF from ij.plugin import\",\n \"= '_'.join([anim, cell_label, str(i)]) fout.write('\\\\n' + '\\\\t'.join([out_title, str(x_full_px), str(y_full_px)])) #Final tsv of form\",\n \"#Records metadata (x,y location) for cells that were extracted with 1_find_extract_cells.py #metadata will\",\n \"io #Records metadata (x,y location) for cells that were extracted with 1_find_extract_cells.py #metadata\",\n \"+ str(e+1) + '/' + str(len(xml_files)) #Find the orig .nd2 file, copied from\",\n \"in range(len(vals['counter'][cell_type])): print 'Iteration ' + str(i+1) + '/' + str(len(vals['counter'][cell_type])) #Convert Px\",\n \"from find_extract_cells.py, see that code for more details. orig_f_name = re.search('(?<=CellCounter_).*(?=\\\\\\\\-Downsampled)', os.path.split(xml_)[1]).group() +\",\n \"'1'-'8' = (lists) of tuples containing cell positions in the form (x,y) '''\",\n \"to cluster cells from similar locations on the section -> semi-quantiative, local, analysis\",\n \"(str) path to xml file on disk Values: (dict). Keys 'x_cal', 'y_cal' =\",\n \"to xml file on disk Values: (dict). Keys 'x_cal', 'y_cal' = (float) calibrations\",\n \"in files if f == orig_f_name] if len(files_found) == 1: fullres_image = files_found[0]\",\n \"open(f_out_path,'w') as fout: fout.write('\\\\t'.join(['cell','x_um','y_um'])) for e,xml_ in enumerate(xml_files): print 'Working on file: '\",\n \"details. orig_f_name = re.search('(?<=CellCounter_).*(?=\\\\\\\\-Downsampled)', os.path.split(xml_)[1]).group() + '.nd2' search_dir = '/'.join(os.path.split(xml_)[0].split('/')[:-1]) files_found = [os.path.join(root,\"\n]","isConversational":false}}},{"rowIdx":3590,"cells":{"ngram":{"kind":"list like","value":["max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter):","for i in range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0, 4) print('action","i in range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0, 4) print('action is',","<reponame>Abluceli/Multi-agent-Reinforcement-Learning-Algorithms from env_SingleCatchPigs import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000","env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch the pig', reward,","random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0)","0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\"iter= \", i) env.render() action","action = random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward', reward,","4) print('action is', action) reward, done = env.step(action) print('reward', reward, 'done', done) if","\", i) env.render() action = random.randint(0, 4) print('action is', action) reward, done =","range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0, 4) print('action is', action) reward,","i) env.render() action = random.randint(0, 4) print('action is', action) reward, done = env.step(action)","env.render() action = random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward',","reward, done = env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch","action) reward, done = env.step(action) print('reward', reward, 'done', done) if reward > 0:","print('reward', reward, 'done', done) if reward > 0: print('catch the pig', reward, done)","random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward', reward, 'done', done)","10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\"iter= \",","4], 0) for i in range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0,","print('action is', action) reward, done = env.step(action) print('reward', reward, 'done', done) if reward","= random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward', reward, 'done',","0) for i in range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0, 4)","in range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0, 4) print('action is', action)","env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for","print(\"iter= \", i) env.render() action = random.randint(0, 4) print('action is', action) reward, done","from env_SingleCatchPigs import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2,","EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in","2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\"iter= \", i) env.render()","= 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\"iter=","= env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch the pig',","env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\"iter= \", i) env.render() action =","done = env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch the","is', action) reward, done = env.step(action) print('reward', reward, 'done', done) if reward >","EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4,","env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\"iter= \", i)","env_SingleCatchPigs import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2],","import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4],","import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0)","= EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i"],"string":"[\n \"max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter):\",\n \"for i in range(max_iter): print(\\\"iter= \\\", i) env.render() action = random.randint(0, 4) print('action\",\n \"i in range(max_iter): print(\\\"iter= \\\", i) env.render() action = random.randint(0, 4) print('action is',\",\n \"<reponame>Abluceli/Multi-agent-Reinforcement-Learning-Algorithms from env_SingleCatchPigs import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000\",\n \"env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch the pig', reward,\",\n \"random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0)\",\n \"0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\\\"iter= \\\", i) env.render() action\",\n \"action = random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward', reward,\",\n \"4) print('action is', action) reward, done = env.step(action) print('reward', reward, 'done', done) if\",\n \"\\\", i) env.render() action = random.randint(0, 4) print('action is', action) reward, done =\",\n \"range(max_iter): print(\\\"iter= \\\", i) env.render() action = random.randint(0, 4) print('action is', action) reward,\",\n \"i) env.render() action = random.randint(0, 4) print('action is', action) reward, done = env.step(action)\",\n \"env.render() action = random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward',\",\n \"reward, done = env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch\",\n \"action) reward, done = env.step(action) print('reward', reward, 'done', done) if reward > 0:\",\n \"print('reward', reward, 'done', done) if reward > 0: print('catch the pig', reward, done)\",\n \"random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward', reward, 'done', done)\",\n \"10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\\\"iter= \\\",\",\n \"4], 0) for i in range(max_iter): print(\\\"iter= \\\", i) env.render() action = random.randint(0,\",\n \"print('action is', action) reward, done = env.step(action) print('reward', reward, 'done', done) if reward\",\n \"= random.randint(0, 4) print('action is', action) reward, done = env.step(action) print('reward', reward, 'done',\",\n \"0) for i in range(max_iter): print(\\\"iter= \\\", i) env.render() action = random.randint(0, 4)\",\n \"in range(max_iter): print(\\\"iter= \\\", i) env.render() action = random.randint(0, 4) print('action is', action)\",\n \"env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for\",\n \"print(\\\"iter= \\\", i) env.render() action = random.randint(0, 4) print('action is', action) reward, done\",\n \"from env_SingleCatchPigs import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2,\",\n \"EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in\",\n \"2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\\\"iter= \\\", i) env.render()\",\n \"= 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\\\"iter=\",\n \"= env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch the pig',\",\n \"env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\\\"iter= \\\", i) env.render() action =\",\n \"done = env.step(action) print('reward', reward, 'done', done) if reward > 0: print('catch the\",\n \"is', action) reward, done = env.step(action) print('reward', reward, 'done', done) if reward >\",\n \"EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4,\",\n \"env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter): print(\\\"iter= \\\", i)\",\n \"env_SingleCatchPigs import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2],\",\n \"import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4],\",\n \"import EnvSingleCatchPigs import random env = EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0)\",\n \"= EnvSingleCatchPigs(7) max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i\"\n]","isConversational":false}}},{"rowIdx":3591,"cells":{"ngram":{"kind":"list like","value":["== {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level","data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty flags","if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index() return d _TSV_GZ_FILENAME","conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool: return","pd.DataFrame: split = series.str.split(\" \") df = pd.DataFrame( { \"value\": split.apply(lambda l: l[0]","utf-8 -*- import re import gzip import pandas as pd import numpy as","= header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, )","import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE","split.apply(lambda l: l[0] if l else None), \"flag\": split.apply(lambda l: l[1] if l","pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], )","= d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell] del","types # that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\" data[\"flag\"]","s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels","a string, for performance reasons. # This is a pandas PerformanceWarning: # \"your","d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell]","performance may suffer as PyTables will pickle object types # that it cannot","= index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle","import gzip import pandas as pd import numpy as np from eust.core import","pickle object types # that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values]","split.apply(lambda l: l[1] if l and len(l) > 1 else None), } )","a pandas PerformanceWarning: # \"your performance may suffer as PyTables will pickle object","for performance reasons. # This is a pandas PerformanceWarning: # \"your performance may","= _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty flags by","if l and len(l) > 1 else None), } ) return df def","_download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool:","s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column","1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)),","split = series.str.split(\" \") df = pd.DataFrame( { \"value\": split.apply(lambda l: l[0] if","\"flag\": split.apply(lambda l: l[1] if l and len(l) > 1 else None), }","file as a string, for performance reasons. # This is a pandas PerformanceWarning:","pd.Series) -> pd.DataFrame: split = series.str.split(\" \") df = pd.DataFrame( { \"value\": split.apply(lambda","saved in the HDF # file as a string, for performance reasons. #","l: l[1] if l and len(l) > 1 else None), } ) return","pandas as pd import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE","def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0]","return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\" \") df =","<filename>eust/tables/data.py # -*- coding: utf-8 -*- import re import gzip import pandas as","PyTables will pickle object types # that it cannot map directly to c-types","_split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\" \") df = pd.DataFrame( { \"value\":","all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d)","index_data = index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot","level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index","gzip import pandas as pd import numpy as np from eust.core import _download_file,","sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\")","= np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings = d.index.unique(\"time\") matches_year","it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\" data[\"flag\"] = data[\"flag\"].replace({\"\": None})","def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index","= d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names))",") d = _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float)","\"time\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\"","it at this point so that the null flag is saved in the","string, for performance reasons. # This is a pandas PerformanceWarning: # \"your performance","= re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) ->","_set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return","path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME","isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True","= the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path,","np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def","this point so that the null flag is saved in the HDF #","assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels],","= d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index =","at this point so that the null flag is saved in the HDF","re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series:","(issue #3) # # Doing it at this point so that the null","import re import gzip import pandas as pd import numpy as np from","= index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0,","l: l[0] if l else None), \"flag\": split.apply(lambda l: l[1] if l and","_TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path,","in d.index.names: time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if","that the null flag is saved in the HDF # file as a","for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int) d =","will pickle object types # that it cannot map directly to c-types #","> 1 else None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df","_HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError:","re import gzip import pandas as pd import numpy as np from eust.core","_DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s))","else None), \"flag\": split.apply(lambda l: l[1] if l and len(l) > 1 else","d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name","= pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"],","index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str)","names=row_dimension_names, ) # cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type))","(_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int) d","as a string, for performance reasons. # This is a pandas PerformanceWarning: #","\"rb\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME","_split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\" in","pd.DataFrame( { \"value\": split.apply(lambda l: l[0] if l else None), \"flag\": split.apply(lambda l:","_read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty flags by None","tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data","_HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) #",") # cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type)) ==","pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name =","that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\" data[\"flag\"] = data[\"flag\"].replace({\"\":","new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\",","matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\",","np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings = d.index.unique(\"time\") matches_year =","= _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\"","d[\"value\"] = d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s)","1 else None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df =","index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame:","def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path","d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d","assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip()","row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) #","index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert","= pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d =","path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir /","return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def","in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index() return","as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url,","gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir /","# Doing it at this point so that the null flag is saved","= series.str.split(\" \") df = pd.DataFrame( { \"value\": split.apply(lambda l: l[0] if l","-*- import re import gzip import pandas as pd import numpy as np","_read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir):","= index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) ->","time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index() return d","_read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names,","dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data =","= _set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME","d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] = np.nan d[\"value\"]","from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s:","by None (issue #3) # # Doing it at this point so that","data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty flags by None (issue",") return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_)","d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def","handle multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d,","index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer,","as PyTables will pickle object types # that it cannot map directly to","duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\"))","= (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int)","len(l) > 1 else None), } ) return df def _set_multiindex_dtype(index, level, type_):","hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty flags by None (issue #3)","\"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer,","l and len(l) > 1 else None), } ) return df def _set_multiindex_dtype(index,","None), \"flag\": split.apply(lambda l: l[1] if l and len(l) > 1 else None),","d = d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH =","\"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings = d.index.unique(\"time\")","cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str} assert","s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index()","index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d","pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True )","# \"your performance may suffer as PyTables will pickle object types # that","set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for","empty flags by None (issue #3) # # Doing it at this point","This is a pandas PerformanceWarning: # \"your performance may suffer as PyTables will","_HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty flags by None (issue #3) #","# that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\" data[\"flag\"] =","inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] = np.nan d[\"value\"] =","header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) #","PerformanceWarning: # \"your performance may suffer as PyTables will pickle object types #","the null flag is saved in the HDF # file as a string,","# file as a string, for performance reasons. # This is a pandas","_set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME =","def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\" \") df = pd.DataFrame( {","d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"]","object types # that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\"","= pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names","df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index =","_HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f: return","-> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\" \")","pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path =","_TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir","suffer as PyTables will pickle object types # that it cannot map directly","for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"]","# no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda","else None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame()","with gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir","= the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data =","Replace empty flags by None (issue #3) # # Doing it at this","pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d = d.stack()","d = _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float) if","_download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir /","so that the null flag is saved in the HDF # file as","null flag is saved in the HDF # file as a string, for","reasons. # This is a pandas PerformanceWarning: # \"your performance may suffer as","dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir","\"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f) def","may suffer as PyTables will pickle object types # that it cannot map","#3) # # Doing it at this point so that the null flag","= d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names)","flag is saved in the HDF # file as a string, for performance",">= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays(","except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace","HDF # file as a string, for performance reasons. # This is a","d.index.names: time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year):","d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"] ==","str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\"","/ _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except","int) d = d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH","d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d","header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data))","l[1] if l and len(l) > 1 else None), } ) return df","assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\")) d.index","d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index,","multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series),","_YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series)","d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name,","/ _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf(","\"your performance may suffer as PyTables will pickle object types # that it","Doing it at this point so that the null flag is saved in","pandas PerformanceWarning: # \"your performance may suffer as PyTables will pickle object types","d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s","index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer)","= \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as","return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer)","point so that the null flag is saved in the HDF # file","\"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f:","# This is a pandas PerformanceWarning: # \"your performance may suffer as PyTables","level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] =","and len(l) > 1 else None), } ) return df def _set_multiindex_dtype(index, level,","in the HDF # file as a string, for performance reasons. # This","header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data","all(matches_year): d.index = _set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index() return d _TSV_GZ_FILENAME =","coding: utf-8 -*- import re import gzip import pandas as pd import numpy","re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame:","= top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) ==","_download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path =","return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index","flags by None (issue #3) # # Doing it at this point so","eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str)","assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d =","complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty flags by None (issue #3) # #","map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\" data[\"flag\"] = data[\"flag\"].replace({\"\": None}) return data","{str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in","data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"],","if l else None), \"flag\": split.apply(lambda l: l[1] if l and len(l) >","index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional","= row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no","assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name =","top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data)","the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path)","# -*- coding: utf-8 -*- import re import gzip import pandas as pd","import pandas as pd import numpy as np from eust.core import _download_file, conf","new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell =","= index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d =","/ _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path =","d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names,","series.str.split(\" \") df = pd.DataFrame( { \"value\": split.apply(lambda l: l[0] if l else","-> pd.DataFrame: split = series.str.split(\" \") df = pd.DataFrame( { \"value\": split.apply(lambda l:","return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell","d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >=","== len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data","the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH)","= \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f)",") # Replace empty flags by None (issue #3) # # Doing it","\"value\": split.apply(lambda l: l[0] if l else None), \"flag\": split.apply(lambda l: l[1] if","bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\" \") df","hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data =","-> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path","l else None), \"flag\": split.apply(lambda l: l[1] if l and len(l) > 1","no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s:","-> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name","complib=conf[\"hdf_complib\"], ) # Replace empty flags by None (issue #3) # # Doing","len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\",\")) d.index =","_read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data","\"time\" in d.index.names: time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s in time_strings)","_HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\")","pd import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\")","= d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for","# cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str}","\":\", \"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings =","= dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME","pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names =","top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell]","= re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def","the HDF # file as a string, for performance reasons. # This is","len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name","type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def","_TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame:","-*- coding: utf-8 -*- import re import gzip import pandas as pd import","_read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir):","column labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns","as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\")","try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH,","= pd.DataFrame( { \"value\": split.apply(lambda l: l[0] if l else None), \"flag\": split.apply(lambda","df = pd.DataFrame( { \"value\": split.apply(lambda l: l[0] if l else None), \"flag\":","row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates","f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path)","as pd import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE =","== \":\", \"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\" in d.index.names: time_strings","def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split","if \"time\" in d.index.names: time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s in","is a pandas PerformanceWarning: # \"your performance may suffer as PyTables will pickle","len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data =","dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path","# Replace empty flags by None (issue #3) # # Doing it at","numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE =","labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert","# # Doing it at this point so that the null flag is","d = pd.read_csv(path_or_buffer, sep=\"\\t\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\")","= d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\"","cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\" data[\"flag\"] = data[\"flag\"].replace({\"\": None}) return","_is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split =","time_strings = d.index.unique(\"time\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index","FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace empty","d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels(","del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1","def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \"rb\") as f: return _read_tsv(f) def _download_tsv_gz(url,","None (issue #3) # # Doing it at this point so that the","in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"] == \":\", \"value\"] = np.nan","d.loc[d[\"value\"] == \":\", \"value\"] = np.nan d[\"value\"] = d[\"value\"].astype(float) if \"time\" in d.index.names:","d _TSV_GZ_FILENAME = \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) ->","list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d = d.stack() assert","l[0] if l else None), \"flag\": split.apply(lambda l: l[1] if l and len(l)","is saved in the HDF # file as a string, for performance reasons.","\") df = pd.DataFrame( { \"value\": split.apply(lambda l: l[0] if l else None),","row_dimension_names, header_dimension_name = top_left_cell.split(\"\\\\\") row_dimension_names = row_dimension_names.split(\",\") index_data = d[top_left_cell] del d[top_left_cell] assert","[level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\"value\"] == \":\",","{ \"value\": split.apply(lambda l: l[0] if l else None), \"flag\": split.apply(lambda l: l[1]","performance reasons. # This is a pandas PerformanceWarning: # \"your performance may suffer","import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\"^[a-z_0-9]+$\") _YEAR_RE = re.compile(r\"^(1|2)[0-9]{3}$\") def _is_valid_dimension_name(s: str) ->","None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level]","= \"data.tsv.gz\" _HDF_FILENAME = \"data.h5\" _HDF_TABLE_PATH = \"eurostat_table\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with","bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\" \") df = pd.DataFrame(","def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try:","} ) return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] =","d.index = _set_multiindex_dtype(d.index, \"time\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \"data.tsv.gz\""],"string":"[\n \"== {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level\",\n \"data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags\",\n \"if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index() return d _TSV_GZ_FILENAME\",\n \"conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool: return\",\n \"pd.DataFrame: split = series.str.split(\\\" \\\") df = pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0]\",\n \"utf-8 -*- import re import gzip import pandas as pd import numpy as\",\n \"= header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, )\",\n \"import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE\",\n \"split.apply(lambda l: l[0] if l else None), \\\"flag\\\": split.apply(lambda l: l[1] if l\",\n \"pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], )\",\n \"= d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del\",\n \"types # that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\\\" data[\\\"flag\\\"]\",\n \"s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels\",\n \"a string, for performance reasons. # This is a pandas PerformanceWarning: # \\\"your\",\n \"d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell]\",\n \"performance may suffer as PyTables will pickle object types # that it cannot\",\n \"= index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle\",\n \"import gzip import pandas as pd import numpy as np from eust.core import\",\n \"pickle object types # that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values]\",\n \"split.apply(lambda l: l[1] if l and len(l) > 1 else None), } )\",\n \"a pandas PerformanceWarning: # \\\"your performance may suffer as PyTables will pickle object\",\n \"for performance reasons. # This is a pandas PerformanceWarning: # \\\"your performance may\",\n \"= _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags by\",\n \"if l and len(l) > 1 else None), } ) return df def\",\n \"_download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool:\",\n \"s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column\",\n \"1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)),\",\n \"split = series.str.split(\\\" \\\") df = pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0] if\",\n \"\\\"flag\\\": split.apply(lambda l: l[1] if l and len(l) > 1 else None), }\",\n \"file as a string, for performance reasons. # This is a pandas PerformanceWarning:\",\n \"pd.Series) -> pd.DataFrame: split = series.str.split(\\\" \\\") df = pd.DataFrame( { \\\"value\\\": split.apply(lambda\",\n \"saved in the HDF # file as a string, for performance reasons. #\",\n \"l: l[1] if l and len(l) > 1 else None), } ) return\",\n \"pandas as pd import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE\",\n \"def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0]\",\n \"return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\\\" \\\") df =\",\n \"<filename>eust/tables/data.py # -*- coding: utf-8 -*- import re import gzip import pandas as\",\n \"PyTables will pickle object types # that it cannot map directly to c-types\",\n \"_split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\\\" \\\") df = pd.DataFrame( { \\\"value\\\":\",\n \"all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d)\",\n \"index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot\",\n \"level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index\",\n \"gzip import pandas as pd import numpy as np from eust.core import _download_file,\",\n \"sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\")\",\n \"= np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year\",\n \"it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\\\" data[\\\"flag\\\"] = data[\\\"flag\\\"].replace({\\\"\\\": None})\",\n \"def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index\",\n \"= d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names))\",\n \") d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float)\",\n \"\\\"time\\\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\"\",\n \"it at this point so that the null flag is saved in the\",\n \"string, for performance reasons. # This is a pandas PerformanceWarning: # \\\"your performance\",\n \"= re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) ->\",\n \"_set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return\",\n \"path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME\",\n \"isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True\",\n \"= the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path,\",\n \"np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def\",\n \"this point so that the null flag is saved in the HDF #\",\n \"assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels],\",\n \"= d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index =\",\n \"at this point so that the null flag is saved in the HDF\",\n \"re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series:\",\n \"(issue #3) # # Doing it at this point so that the null\",\n \"import re import gzip import pandas as pd import numpy as np from\",\n \"= index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0,\",\n \"l: l[0] if l else None), \\\"flag\\\": split.apply(lambda l: l[1] if l and\",\n \"_TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path,\",\n \"in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if\",\n \"that the null flag is saved in the HDF # file as a\",\n \"for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d =\",\n \"will pickle object types # that it cannot map directly to c-types #\",\n \"> 1 else None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df\",\n \"_HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError:\",\n \"re import gzip import pandas as pd import numpy as np from eust.core\",\n \"_DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s))\",\n \"else None), \\\"flag\\\": split.apply(lambda l: l[1] if l and len(l) > 1 else\",\n \"d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name\",\n \"= pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"],\",\n \"index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str)\",\n \"names=row_dimension_names, ) # cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type))\",\n \"(_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d\",\n \"as a string, for performance reasons. # This is a pandas PerformanceWarning: #\",\n \"\\\"rb\\\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME\",\n \"_split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in\",\n \"pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0] if l else None), \\\"flag\\\": split.apply(lambda l:\",\n \"_read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags by None\",\n \"tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data\",\n \"_HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) #\",\n \") # cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type)) ==\",\n \"pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name =\",\n \"that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\\\" data[\\\"flag\\\"] = data[\\\"flag\\\"].replace({\\\"\\\":\",\n \"new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\",\",\n \"matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\",\",\n \"np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year =\",\n \"= _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\"\",\n \"d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s)\",\n \"1 else None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df =\",\n \"index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame:\",\n \"def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path\",\n \"d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d\",\n \"assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip()\",\n \"row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) #\",\n \"index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert\",\n \"= pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d =\",\n \"path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir /\",\n \"return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def\",\n \"in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index() return\",\n \"as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url,\",\n \"gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir /\",\n \"# Doing it at this point so that the null flag is saved\",\n \"= series.str.split(\\\" \\\") df = pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0] if l\",\n \"-*- import re import gzip import pandas as pd import numpy as np\",\n \"_read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir):\",\n \"= index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) ->\",\n \"time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index() return d\",\n \"_read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names,\",\n \"dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data =\",\n \"= _set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME\",\n \"d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"]\",\n \"from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s:\",\n \"by None (issue #3) # # Doing it at this point so that\",\n \"data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags by None (issue\",\n \") return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_)\",\n \"d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def\",\n \"handle multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d,\",\n \"index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer,\",\n \"as PyTables will pickle object types # that it cannot map directly to\",\n \"duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\"))\",\n \"= (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int)\",\n \"len(l) > 1 else None), } ) return df def _set_multiindex_dtype(index, level, type_):\",\n \"hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags by None (issue #3)\",\n \"\\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer,\",\n \"l and len(l) > 1 else None), } ) return df def _set_multiindex_dtype(index,\",\n \"None), \\\"flag\\\": split.apply(lambda l: l[1] if l and len(l) > 1 else None),\",\n \"d = d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH =\",\n \"\\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\")\",\n \"cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str} assert\",\n \"s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index()\",\n \"index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d\",\n \"pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True )\",\n \"# \\\"your performance may suffer as PyTables will pickle object types # that\",\n \"set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for\",\n \"empty flags by None (issue #3) # # Doing it at this point\",\n \"This is a pandas PerformanceWarning: # \\\"your performance may suffer as PyTables will\",\n \"_HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags by None (issue #3) #\",\n \"# that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\\\" data[\\\"flag\\\"] =\",\n \"inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] =\",\n \"header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) #\",\n \"PerformanceWarning: # \\\"your performance may suffer as PyTables will pickle object types #\",\n \"the null flag is saved in the HDF # file as a string,\",\n \"# file as a string, for performance reasons. # This is a pandas\",\n \"_set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME =\",\n \"def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\\\" \\\") df = pd.DataFrame( {\",\n \"d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"]\",\n \"object types # that it cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\\\"\",\n \"= pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names\",\n \"df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index =\",\n \"_HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f: return\",\n \"-> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\\\" \\\")\",\n \"pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path =\",\n \"_TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir\",\n \"suffer as PyTables will pickle object types # that it cannot map directly\",\n \"for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"]\",\n \"# no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda\",\n \"else None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame()\",\n \"with gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir\",\n \"= the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data =\",\n \"Replace empty flags by None (issue #3) # # Doing it at this\",\n \"pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d = d.stack()\",\n \"d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if\",\n \"_download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir /\",\n \"so that the null flag is saved in the HDF # file as\",\n \"null flag is saved in the HDF # file as a string, for\",\n \"reasons. # This is a pandas PerformanceWarning: # \\\"your performance may suffer as\",\n \"dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir\",\n \"\\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f) def\",\n \"may suffer as PyTables will pickle object types # that it cannot map\",\n \"#3) # # Doing it at this point so that the null flag\",\n \"= d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names)\",\n \"flag is saved in the HDF # file as a string, for performance\",\n \">= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays(\",\n \"except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace\",\n \"HDF # file as a string, for performance reasons. # This is a\",\n \"d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year):\",\n \"d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"] ==\",\n \"str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\\\"\",\n \"/ _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except\",\n \"int) d = d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH\",\n \"d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d\",\n \"header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data))\",\n \"l[1] if l and len(l) > 1 else None), } ) return df\",\n \"assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index\",\n \"d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index = _set_multiindex_dtype(d.index,\",\n \"multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series),\",\n \"_YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series)\",\n \"d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name,\",\n \"/ _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf(\",\n \"\\\"your performance may suffer as PyTables will pickle object types # that it\",\n \"Doing it at this point so that the null flag is saved in\",\n \"pandas PerformanceWarning: # \\\"your performance may suffer as PyTables will pickle object types\",\n \"d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s\",\n \"index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer)\",\n \"= \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as\",\n \"return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer)\",\n \"point so that the null flag is saved in the HDF # file\",\n \"\\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f:\",\n \"# This is a pandas PerformanceWarning: # \\\"your performance may suffer as PyTables\",\n \"level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] =\",\n \"and len(l) > 1 else None), } ) return df def _set_multiindex_dtype(index, level,\",\n \"in the HDF # file as a string, for performance reasons. # This\",\n \"header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data\",\n \"all(matches_year): d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index() return d _TSV_GZ_FILENAME =\",\n \"coding: utf-8 -*- import re import gzip import pandas as pd import numpy\",\n \"re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame:\",\n \"= top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) ==\",\n \"_download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path =\",\n \"return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index\",\n \"flags by None (issue #3) # # Doing it at this point so\",\n \"eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str)\",\n \"assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in d.index.levels], inplace=True ) d =\",\n \"complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags by None (issue #3) # #\",\n \"map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\\\" data[\\\"flag\\\"] = data[\\\"flag\\\"].replace({\\\"\\\": None}) return data\",\n \"{str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level in\",\n \"data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"],\",\n \"if l else None), \\\"flag\\\": split.apply(lambda l: l[1] if l and len(l) >\",\n \"index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional\",\n \"= row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no\",\n \"assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name =\",\n \"top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data)\",\n \"the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path)\",\n \"# -*- coding: utf-8 -*- import re import gzip import pandas as pd\",\n \"import pandas as pd import numpy as np from eust.core import _download_file, conf\",\n \"new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell =\",\n \"= index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d =\",\n \"/ _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path =\",\n \"d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index = pd.MultiIndex.from_arrays( list(zip(*index_data)), names=row_dimension_names,\",\n \"series.str.split(\\\" \\\") df = pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0] if l else\",\n \"-> pd.DataFrame: split = series.str.split(\\\" \\\") df = pd.DataFrame( { \\\"value\\\": split.apply(lambda l:\",\n \"return new_index def _read_tsv(path_or_buffer) -> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell\",\n \"d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >=\",\n \"== len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data\",\n \"the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH)\",\n \"= \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f)\",\n \") # Replace empty flags by None (issue #3) # # Doing it\",\n \"\\\"value\\\": split.apply(lambda l: l[0] if l else None), \\\"flag\\\": split.apply(lambda l: l[1] if\",\n \"bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\\\" \\\") df\",\n \"hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data =\",\n \"-> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path\",\n \"l else None), \\\"flag\\\": split.apply(lambda l: l[1] if l and len(l) > 1\",\n \"no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s:\",\n \"-> pd.DataFrame: d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name\",\n \"complib=conf[\\\"hdf_complib\\\"], ) # Replace empty flags by None (issue #3) # # Doing\",\n \"len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data = index_data.apply(lambda s: s.split(\\\",\\\")) d.index =\",\n \"_read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try: data\",\n \"\\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s in time_strings)\",\n \"_HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\")\",\n \"pd import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\")\",\n \"= d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for\",\n \"# cannot handle multidimensional column labels d = d.stack() assert set(d.apply(type)) == {str}\",\n \"\\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings =\",\n \"= dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME\",\n \"pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names =\",\n \"top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell]\",\n \"= re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def\",\n \"the HDF # file as a string, for performance reasons. # This is\",\n \"len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name\",\n \"type_): index_df = index.to_frame() index_df[level] = index_df[level].astype(type_) new_index = index_df.set_index(index.names).index return new_index def\",\n \"_TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame:\",\n \"-*- coding: utf-8 -*- import re import gzip import pandas as pd import\",\n \"_read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir):\",\n \"column labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns\",\n \"as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\")\",\n \"try: data = pd.read_hdf(hdf_path, _HDF_TABLE_PATH) except FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH,\",\n \"= pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0] if l else None), \\\"flag\\\": split.apply(lambda\",\n \"df = pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0] if l else None), \\\"flag\\\":\",\n \"row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates\",\n \"f: return _read_tsv(f) def _download_tsv_gz(url, dst_dir): path = dst_dir / _TSV_GZ_FILENAME _download_file(url, path)\",\n \"as pd import numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE =\",\n \"== \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names: time_strings\",\n \"def _is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split\",\n \"if \\\"time\\\" in d.index.names: time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s in\",\n \"is a pandas PerformanceWarning: # \\\"your performance may suffer as PyTables will pickle\",\n \"len(index_data) # no duplicates assert len(row_dimension_names) >= 1 d.columns.name = header_dimension_name index_data =\",\n \"dst_dir / _TSV_GZ_FILENAME _download_file(url, path) def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path\",\n \"# Replace empty flags by None (issue #3) # # Doing it at\",\n \"numpy as np from eust.core import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE =\",\n \"labels d = d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert\",\n \"# # Doing it at this point so that the null flag is\",\n \"d = pd.read_csv(path_or_buffer, sep=\\\"\\\\t\\\", header=0, dtype=str) top_left_cell = d.columns[0] row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\")\",\n \"= d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\"\",\n \"cannot map directly to c-types # [inferred_type->mixed,key->block0_values] [items->['flag']]\\\" data[\\\"flag\\\"] = data[\\\"flag\\\"].replace({\\\"\\\": None}) return\",\n \"_is_valid_dimension_name(s: str) -> bool: return bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split =\",\n \"time_strings = d.index.unique(\\\"time\\\") matches_year = (_YEAR_RE.match(s) for s in time_strings) if all(matches_year): d.index\",\n \"FileNotFoundError: data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\\\"hdf_complevel\\\"], complib=conf[\\\"hdf_complib\\\"], ) # Replace empty\",\n \"d.stack() assert set(d.apply(type)) == {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels(\",\n \"del d[top_left_cell] assert len(set(index_data)) == len(index_data) # no duplicates assert len(row_dimension_names) >= 1\",\n \"def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with gzip.open(path_or_buffer, \\\"rb\\\") as f: return _read_tsv(f) def _download_tsv_gz(url,\",\n \"None (issue #3) # # Doing it at this point so that the\",\n \"in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan\",\n \"d.loc[d[\\\"value\\\"] == \\\":\\\", \\\"value\\\"] = np.nan d[\\\"value\\\"] = d[\\\"value\\\"].astype(float) if \\\"time\\\" in d.index.names:\",\n \"d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) ->\",\n \"list(zip(*index_data)), names=row_dimension_names, ) # cannot handle multidimensional column labels d = d.stack() assert\",\n \"l[0] if l else None), \\\"flag\\\": split.apply(lambda l: l[1] if l and len(l)\",\n \"is saved in the HDF # file as a string, for performance reasons.\",\n \"\\\") df = pd.DataFrame( { \\\"value\\\": split.apply(lambda l: l[0] if l else None),\",\n \"row_dimension_names, header_dimension_name = top_left_cell.split(\\\"\\\\\\\\\\\") row_dimension_names = row_dimension_names.split(\\\",\\\") index_data = d[top_left_cell] del d[top_left_cell] assert\",\n \"[level.str.strip() for level in d.index.levels], inplace=True ) d = _split_values_flags(d) d.loc[d[\\\"value\\\"] == \\\":\\\",\",\n \"{ \\\"value\\\": split.apply(lambda l: l[0] if l else None), \\\"flag\\\": split.apply(lambda l: l[1]\",\n \"performance reasons. # This is a pandas PerformanceWarning: # \\\"your performance may suffer\",\n \"import _download_file, conf _DIMENSION_NAME_RE = re.compile(r\\\"^[a-z_0-9]+$\\\") _YEAR_RE = re.compile(r\\\"^(1|2)[0-9]{3}$\\\") def _is_valid_dimension_name(s: str) ->\",\n \"None), } ) return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level]\",\n \"= \\\"data.tsv.gz\\\" _HDF_FILENAME = \\\"data.h5\\\" _HDF_TABLE_PATH = \\\"eurostat_table\\\" def _read_tsv_gz(path_or_buffer) -> pd.DataFrame: with\",\n \"bool(_DIMENSION_NAME_RE.match(s)) def _split_values_flags(series: pd.Series) -> pd.DataFrame: split = series.str.split(\\\" \\\") df = pd.DataFrame(\",\n \"def _read(the_dir): hdf_path = the_dir / _HDF_FILENAME tsv_gz_path = the_dir / _TSV_GZ_FILENAME try:\",\n \"} ) return df def _set_multiindex_dtype(index, level, type_): index_df = index.to_frame() index_df[level] =\",\n \"d.index = _set_multiindex_dtype(d.index, \\\"time\\\", int) d = d.sort_index() return d _TSV_GZ_FILENAME = \\\"data.tsv.gz\\\"\"\n]","isConversational":false}}},{"rowIdx":3592,"cells":{"ngram":{"kind":"list like","value":["curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats =","' does not have training argument.') training = kwargs['training'] if training: # Set","stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx +","\"\"\"Returns function that provides current learning rate based on config. NOTE: This returns","a pyplot, save to buffer and return TB compatible image.\"\"\" plt.figure() plt.plot(x, y)","lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE","\"\"\"Intializes learning rate, optimizer and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step =","gru_layer.stateful = False return dataset def gen_plot(x, y): \"\"\"Create a pyplot, save to","status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return","tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if","len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)]","division from __future__ import print_function from config import CONFIG import json import tensorflow","train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be","config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict =","'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step","softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im,","steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)]","v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1)","steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original']","False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper","global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours in embedding space.\"\"\" # Set","tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def","returns a function as in Eager we need to call assign to update","return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range","to see the future. steps = np.arange(step - (num_steps - 1) * stride,","warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int =","[image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros(","as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict)","- (num_steps - 1) * stride, step + stride, stride) return steps def","tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that provides current learning rate based on","called with set_learning_phase decorator which' ' does not have training argument.') training =","elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else: return","emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len *","frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list","opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support","[] n = 0 def cond(n): if max_embs is None: return True else:","to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb","= [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries),","global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step):","y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) #","and learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE ==","' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for","file is not provided or is corrupted\" return config def prepare_gpu(ind=-1): ind =","frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list","'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len","emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames,","n < max_embs # Make Recurrent Layers stateful, set batch size. # We","dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels':","than one batch to be passed and we don't want to automatically #","labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens =","for i in range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps","global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on config and learning rate.\"\"\"","to return the current learning rate based on the provided config. Raises: ValueError:","global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return","seq_lens_list = [] names_list = [] seq_labels_list = [] if keep_data: frames_list =","path).\"\"\" # Instantiate checkpoint and restore from any pre-existing checkpoint. # Since model","= dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else:","depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that provides current learning rate based on config.","in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n)","config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config","1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v,","function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the function f after setting proper learning","to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False","as tf import numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top","\"\"\"Returns learning rate during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int =","axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize","we are embedding the whole sequence and that can take # more than","been passed in the config. \"\"\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if","gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \"\"\"Create a","%s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt","'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if","k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from","seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow:","time.time() def set_learning_phase(f): \"\"\"Sets the correct learning phase before calling function f.\"\"\" def","batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \"\"\"Simple timer for","def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json')","= tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model,","embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow:","names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames =","= np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps)))))","m random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m,","TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image =","= num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride","np.arange(step - (num_steps - 1) * stride, step + stride, stride) return steps","if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step","embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len])","tf.expand_dims(image, 0) return image class Stopwatch(object): \"\"\"Simple timer for measuring elapsed time.\"\"\" def","each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True","axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split,","list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function","iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data","tensorflow as tf import numpy as np import matplotlib.pyplot as plt # pylint:","> -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be the","num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate during","im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix","names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames =","raise ValueError('You might be overwriting a directory that already ' 'has train_logs. Please","== 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x,","PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch","num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in","batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def","optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD(","= [] seq_labels = [] if keep_data: frames = [] if optical_flow: frame_original","keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers for other evaluation","disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS,","2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats =","optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not","= tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch',","def reset(self): self.time = time.time() def set_learning_phase(f): \"\"\"Sets the correct learning phase before","assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if","in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config =","* frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data =","= json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise","global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x","get_lr_fn(optimizer_config): \"\"\"Returns function that provides current learning rate based on config. NOTE: This","Please provide a new logdir name in ' 'config or pass --force_train while","if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config is not","frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size,","int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) * frames_per_batch > seq_len:","\"\"\"Get embeddings from a one epoch iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step","len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data:","= np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch,","labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)]","return TB compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf","return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for","Need to do this as some modalities might not exist. if len(v.shape) >","io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image image =","def elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval","tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries,","* lr + is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow","emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn,","**kwargs): \"\"\"Calls the function f after setting proper learning phase.\"\"\" if 'training' not","= np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames,","= int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) * frames_per_batch >","tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer","optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset","curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy',","/ np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations","image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image, 0)","batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape","if max_embs is None: return True else: return n < max_embs # Make","hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers:","FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames =","absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS","checkpoint and restore from any pre-existing checkpoint. # Since model is a dict","batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours in embedding space.\"\"\" # Set learning_phase","= None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config","exist. Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True:","t=1.0): e = np.exp(np.array(w) / t) dist = e / np.sum(e) return dist","ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind],","config as no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w')","return the current learning rate based on the provided config. Raises: ValueError: in","f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the function f after setting proper learning phase.\"\"\"","= [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) *","= np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len,","new logdir name in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir)","return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c","seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i +","target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \"\"\"Sets","\"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or","gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches","Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \"\"\"Intializes","ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in","data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data through base","training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n,","in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps)","[] steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list = []","# Reset statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE ==","seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label","is not provided or is corrupted\" return config def prepare_gpu(ind=-1): ind = int(ind)","embs = [] labels = [] steps = [] seq_lens = [] names","CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def","train_logs. Please provide a new logdir name in ' 'config or pass --force_train","'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f","class Stopwatch(object): \"\"\"Simple timer for measuring elapsed time.\"\"\" def __init__(self): self.reset() def elapsed(self):","import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps):","to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path,","else: # Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0)","optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we","not in kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does not","def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS","elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval def","easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size,","logging.info( 'Using the existing passed in config as no config.json file exists in","if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len =","config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going","num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps):","warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup * warmup_lr","step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours in","frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.')","== 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len","random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range (0, n) #","frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) ==","emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False,","len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as e: # Memory growth must be set","os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in","== 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported.","in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()])","elapsed time.\"\"\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self,","tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr =","frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert","n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles =","CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]]","names_list = [] seq_labels_list = [] if keep_data: frames_list = [] if optical_flow:","boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns","steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list = [] if","* CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats =","learning phase before calling function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the function f","Args: optimizer_config: EasyDict, contains params required to initialize the learning rate and the","- is_warmup) * lr + is_warmup * warmup_lr return lr # Minimally adapted","num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for k,","Only support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda","lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS,","time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time","of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64)","%s not supported. Only support' 'the following decay types: fixed, exp_decay, manual,' 'and","# np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks:","(i + 1) * frames_per_batch > seq_len: num_steps = seq_len - i *","from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict","j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im):","if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding","one exists on the path).\"\"\" # Instantiate checkpoint and restore from any pre-existing","does not have training argument.') training = kwargs['training'] if training: # Set learning_phase","and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list =","for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries,","isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else: return config","import io import math import os import time from absl import flags from","restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint (if one exists on the path).\"\"\" #","else: raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer","if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to","# NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len):","if (i + 1) * frames_per_batch > seq_len: num_steps = seq_len - i","\"Logical GPUs\"]) except RuntimeError as e: # Memory growth must be set before","data.items(): # Need to do this as some modalities might not exist. if","# reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer","== len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames)","None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES *","CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats,","need to call assign to update the learning rate. Args: optimizer_config: EasyDict, contains","frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for k, v in","on the provided config. Raises: ValueError: in case invalid params have been passed","do this as some modalities might not exist. if len(v.shape) > 1 and","steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name])","global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate","plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF","return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries =","launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs'))","num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats =","import print_function from config import CONFIG import json import tensorflow as tf import","model['emb'] embs_list = [] labels_list = [] steps_list = [] seq_lens_list = []","= model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn,","as e: # Memory growth must be set before GPUs have been initialized","curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d,","softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e / np.sum(e) return","boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] *","proper learning phase.\"\"\" if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase","step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix =","for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful","single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx +","inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO:","the function f after setting proper learning phase.\"\"\" if 'training' not in kwargs:","adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] +","target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \"\"\"Sets the correct learning phase","not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def","CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir,","= tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint)","global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int,","exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def","the path).\"\"\" # Instantiate checkpoint and restore from any pre-existing checkpoint. # Since","get_data(iterator): \"\"\"Return a data dict which contains all the requested sequences.\"\"\" data =","tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be the same across","= np.arange(step - (num_steps - 1) * stride, step + stride, stride) return","tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def","= get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v","seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames","== len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps =","done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f):","lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def","CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = []","max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status,","* num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2","seq_labels_list = [] if keep_data: frames_list = [] if optical_flow: frame_original_list = []","and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate","= [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance =","existing passed in config as no config.json file exists in ' '%s', logdir)","config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4)","num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats","lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay(","seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len *","sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e","return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a","[seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps,","chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in","e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random","setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json')","'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step):","force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path)","the whole sequence and that can take # more than one batch to","= [] labels_list = [] steps_list = [] seq_lens_list = [] names_list =","= tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float","to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using","in the config. \"\"\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE ==","dict([(k, to_dict(v)) for k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False,","tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1","def wrapper(*args, **kwargs): \"\"\"Calls the function f after setting proper learning phase.\"\"\" if","tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't","gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for","v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps)","frames_per_batch curr_data = {} for k, v in data.items(): # Need to do","in config as no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path,","on the path).\"\"\" # Instantiate checkpoint and restore from any pre-existing checkpoint. #","following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr,","directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager,","data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats","CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data dict which contains all the requested sequences.\"\"\"","n = 0 def cond(n): if max_embs is None: return True else: return","We don't want to see the future. steps = np.arange(step - (num_steps -","or overwrite: logging.info( 'Using the existing passed in config as no config.json file","return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0)","> 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float =","step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate,","with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k, v in","logdir name in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def","a function as in Eager we need to call assign to update the","cond(n): try: print(n) embs = [] labels = [] steps = [] seq_lens","plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png')","# Memory growth must be set before GPUs have been initialized logging.info(e) os.environ[\"CUDA_VISIBLE_DEVICES\"]","def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint (if one exists on the path).\"\"\"","overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not","flags from absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS","kwargs['training'] if training: # Set learning_phase to True to use models in training","tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split):","provides current learning rate based on config. NOTE: This returns a function as","= tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step)","= data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs","else: raise ValueError('Learning rate decay type %s not supported. Only support' 'the following","data through base CNN.\"\"\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES","keep_data: frames = [] if optical_flow: frame_original = [] # Reset GRU states","axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list","e: # Memory growth must be set before GPUs have been initialized logging.info(e)","f after setting proper learning phase.\"\"\" if 'training' not in kwargs: raise ValueError('Function","= [] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru':","split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist","cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy())","passed in the config. \"\"\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE","= [] n = 0 def cond(n): if max_embs is None: return True","in case invalid params have been passed in the config. \"\"\" lr_params =","global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries","can insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager =","= [] steps = [] seq_lens = [] names = [] seq_labels =","and we don't want to automatically # reset hidden states after each batch.","embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'),","for config to exist. Going to sleep ' ' %s for secs.', config_timeout_seconds)","dict which contains all the requested sequences.\"\"\" data = iterator.get_next() return data, data['chosen_steps'],","states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states()","nearest neighbours in embedding space.\"\"\" # Set learning_phase to False to use models","for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict =","learning rate based on the provided config. Raises: ValueError: in case invalid params","emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW:","Since model is a dict we can insert multiple modular networks in this","learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers:","checkpoint (if one exists on the path).\"\"\" # Instantiate checkpoint and restore from","if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels']","# We don't want to see the future. steps = np.arange(step - (num_steps","- 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list","lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE","from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries","dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float /","= get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0),","boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step,","dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if","batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate during warm","= softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return","get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number","based on config and learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate)","return config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path =","optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from a one epoch iterator.\"\"\" keep_labels =","dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info(","object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries =","that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file)","training = kwargs['training'] if training: # Set learning_phase to True to use models","np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False,","MomentumOptimizer .') return opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and global step.\"\"\"","import tensorflow as tf import numpy as np import matplotlib.pyplot as plt #","len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps","script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path","__future__ import absolute_import from __future__ import division from __future__ import print_function from config","lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries)","* [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label =","else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = []","in Eager we need to call assign to update the learning rate. Args:","print(n) embs = [] labels = [] steps = [] seq_lens = []","tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) #","= data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len *","= seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data:","a dict we can insert multiple modular networks in this dict. checkpoint =","'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config","return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't","= labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens","config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while","the learning rate decay function. Returns: lr_fn: function, this can be called to","open(config_path) as f: config = json.load(f) assert config is not None, \"config file","ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status,","names = [] seq_labels = [] if keep_data: frames = [] if optical_flow:","np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs),","config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir)","data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0]","a data dict which contains all the requested sequences.\"\"\" data = iterator.get_next() return","from __future__ import print_function from config import CONFIG import json import tensorflow as","as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and","p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence)","int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try:","warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr =","more than one batch to be passed and we don't want to automatically","checkpoint. # Since model is a dict we can insert multiple modular networks","-1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint (if","GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers:","tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory","tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup * warmup_lr return","single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator,","try: print(n) embs = [] labels = [] steps = [] seq_lens =","+ num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx,","# more than one batch to be passed and we don't want to","Memory growth must be set before GPUs have been initialized logging.info(e) os.environ[\"CUDA_VISIBLE_DEVICES\"] =","based on the provided config. Raises: ValueError: in case invalid params have been","axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image","emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch))","= os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting","= emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list =","states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful","--force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir)","reset(self): self.time = time.time() def set_learning_phase(f): \"\"\"Sets the correct learning phase before calling","emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0]","for measuring elapsed time.\"\"\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time","manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index =","num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats,","+ is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow object_detection code.","rate based on the provided config. Raises: ValueError: in case invalid params have","sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats","frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy',","= data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels,","= tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] =","format='png') buf.seek(0) # Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4)","lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0,","frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels)","CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else:","num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames","training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack(","provided or is corrupted\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS =","tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) /","seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if","\"\"\"Generate cycles for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def","return time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self):","'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats","= optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return","k, v in data.items(): # Need to do this as some modalities might","in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v))","= [] names_list = [] seq_labels_list = [] if keep_data: frames_list = []","= time.time() def set_learning_phase(f): \"\"\"Sets the correct learning phase before calling function f.\"\"\"","rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where(","cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats","is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow object_detection code. def","get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def","= tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step","version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles","NOTE: This returns a function as in Eager we need to call assign","if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape =","config = json.load(f) assert config is not None, \"config file is not provided","break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data dict which contains all the requested","CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not supported.","def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind >","[] if optical_flow: frame_original = [] # Reset GRU states for each video.","nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance","num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d',","num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx + num_steps) #","the config. \"\"\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay':","'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only","== 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr,","t) dist = e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): #","get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see","if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers for other","Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn =","= CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps = np.arange(step","num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps","dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range (0,","return opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and global step.\"\"\" optimizer =","dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status =","def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list],","supported. Only support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.') return","keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n","= {} for k, v in data.items(): # Need to do this as","def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours in embedding","GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as e: # Memory growth must be","the future. steps = np.arange(step - (num_steps - 1) * stride, step +","might not exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes =","tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the","in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir","num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn'","split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take float64","- 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1,","v in data.items(): # Need to do this as some modalities might not","np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for","sim_matrix to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' %","'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow:","embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names)","learning phase.\"\"\" if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator","[] seq_lens = [] names = [] seq_labels = [] if keep_data: frames","be passed and we don't want to automatically # reset hidden states after","= [] steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list =","# Make Recurrent Layers stateful, set batch size. # We do this as","[] seq_labels_list = [] if keep_data: frames_list = [] if optical_flow: frame_original_list =","Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG","[num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 *","lr = (1.0 - is_warmup) * lr + is_warmup * warmup_lr return lr","modalities might not exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes","end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not supported. Only","lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on config and learning rate.\"\"\" if","This returns a function as in Eager we need to call assign to","not exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx,","optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs)","import division from __future__ import print_function from config import CONFIG import json import","model is a dict we can insert multiple modular networks in this dict.","raise ValueError('Learning rate decay type %s not supported. Only support' 'the following decay","cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On","frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset =","optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs)","pre-existing checkpoint. # Since model is a dict we can insert multiple modular","Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with","are embedding the whole sequence and that can take # more than one","(lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based","rate based on config. NOTE: This returns a function as in Eager we","boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max(","steps = [] seq_lens = [] names = [] seq_labels = [] if","img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for","for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step,","image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf,","f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name =","gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len]","# pylint: disable=g-import-not-at-top import io import math import os import time from absl","def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def","= optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model","checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config] elif","i in range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps =","GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\",","n, axis=-1): # Generate m random permuations of range (0, n) # NumPy","tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles,","range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps =","num_steps): \"\"\"Visualizes a batch.\"\"\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries","if len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len)","cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps,","code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries)","= json.load(f) assert config is not None, \"config file is not provided or","os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a","math import os import time from absl import flags from absl import logging","np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels,","'r') as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break","sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w)","x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p","== len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels:","time.\"\"\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval):","seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow:","return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step,","data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i","'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory","restore from any pre-existing checkpoint. # Since model is a dict we can","import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a","get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from a one","= global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int <","names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished","in embedding space.\"\"\" # Set learning_phase to False to use models in inference","sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def","if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original,","= get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step","= os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to","(1.0 - is_warmup) * lr + is_warmup * warmup_lr return lr # Minimally","for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info(","phase before calling function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the function f after","config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train:","batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours","training argument.') training = kwargs['training'] if training: # Set learning_phase to True to","case invalid params have been passed in the config. \"\"\" lr_params = optimizer_config.LR","is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data dict which","Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the","< max_embs # Make Recurrent Layers stateful, set batch size. # We do","* num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded","already ' 'has train_logs. Please provide a new logdir name in ' 'config","lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for","global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False,","ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore","logging.info( 'Using config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r')","curr_idx = i * frames_per_batch curr_data = {} for k, v in data.items():","= frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for k, v","function. Returns: lr_fn: function, this can be called to return the current learning","j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats),","# Need to do this as some modalities might not exist. if len(v.shape)","CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might","up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant(","tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if","\"\"\"Passes data through base CNN.\"\"\" if num_steps is None: if training: num_steps =","numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import","'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len =","True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as","tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR","return random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate during warm up phase.\"\"\"","CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps = np.arange(step -","or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for","tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0","file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config =","gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical","random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns","labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n +=","alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params):","\"\"\"Create and restore checkpoint (if one exists on the path).\"\"\" # Instantiate checkpoint","(0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles,","flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames = data['frames'] frames_list","num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours in embedding space.\"\"\" # Set learning_phase to","exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k,","np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math import","'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step -","to_dict(v)) for k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True):","get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats","layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers:","= CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def","not provided or is corrupted\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS","epoch iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn']","kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does not have training","memory growth needs to be the same across GPUs for gpu in GPUS:","sequences.\"\"\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training,","learning_rate): \"\"\"Returns optimizer based on config and learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer':","max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config,","multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint,","in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\"","embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens","set_learning_phase(f): \"\"\"Sets the correct learning phase before calling function f.\"\"\" def wrapper(*args, **kwargs):","batch to be passed and we don't want to automatically # reset hidden","labels = [] steps = [] seq_lens = [] names = [] seq_labels","the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate,","= frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) ==","num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx","= flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames = data['frames']","os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config as no config.json","name in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir,","step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx,","optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer,","get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS >","f(*args, **kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path)","for k, v in data.items(): # Need to do this as some modalities","the provided config. Raises: ValueError: in case invalid params have been passed in","as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3),","plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer","= np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None,","tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats","else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return","= v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats,","emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs","f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step,","training: # Set learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1)","status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config]","= seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i","all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len","Raises: ValueError: in case invalid params have been passed in the config. \"\"\"","CONFIG import json import tensorflow as tf import numpy as np import matplotlib.pyplot","def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to","Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args,","config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not","batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 *","learning rate decay function. Returns: lr_fn: function, this can be called to return","keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the","one epoch iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn =","this as we are embedding the whole sequence and that can take #","keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint","Layers stateful, set batch size. # We do this as we are embedding","models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use","import os import time from absl import flags from absl import logging from","if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth","True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to","absolute_import from __future__ import division from __future__ import print_function from config import CONFIG","seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i *","stateful, set batch size. # We do this as we are embedding the","= tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise","the batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \"\"\"Simple timer","return self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \"\"\"Sets the","current learning rate based on config. NOTE: This returns a function as in","= json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return","= CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps","exists on the path).\"\"\" # Instantiate checkpoint and restore from any pre-existing checkpoint.","tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps':","= checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir,","= [1, ] while cond(n): try: print(n) embs = [] labels = []","len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if","staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f","logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as e: # Memory growth","buffer and return TB compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1)","tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config","num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy())","= int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU')","np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist =","# Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add","raise ValueError('Function called with set_learning_phase decorator which' ' does not have training argument.')","[to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k,","for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0]","= model['cnn'] emb = model['emb'] embs_list = [] labels_list = [] steps_list =","if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator which' '","learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs)","'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'),","rate, optimizer and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate","function, this can be called to return the current learning rate based on","tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x)","can take # more than one batch to be passed and we don't","keep_data: frames_list = [] if optical_flow: frame_original_list = [] n = 0 def","'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list","np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]):","lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float","data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0)","float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0):","# np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy())","* num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps *","ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return","= tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr +","exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict)","= np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels","except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list,","config = None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert","axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list,","return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\"","rate during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32)","tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup","np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i]","checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for","= tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list =","* num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats","requested sequences.\"\"\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data,","num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else:","while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is","def gen_plot(x, y): \"\"\"Create a pyplot, save to buffer and return TB compatible","to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict):","import flags from absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg')","\"\"\"Simple timer for measuring elapsed time.\"\"\" def __init__(self): self.reset() def elapsed(self): return time.time()","invalid params have been passed in the config. \"\"\" lr_params = optimizer_config.LR #","f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) +","passed in config as no config.json file exists in ' '%s', logdir) with","# Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn","seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if","fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params))","that can take # more than one batch to be passed and we","to call assign to update the learning rate. Args: optimizer_config: EasyDict, contains params","visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours in embedding space.\"\"\"","def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE ==","labels_list = [] steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list","steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps])","num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride =","self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time()","f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels =","* mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps),","get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i","overwriting a directory that already ' 'has train_logs. Please provide a new logdir","optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): #","optimizer based on config and learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt =","cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list = [] steps_list","if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness","if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data","num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps)","for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data,","max_embs # Make Recurrent Layers stateful, set batch size. # We do this","from any pre-existing checkpoint. # Since model is a dict we can insert","= tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done","learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since","tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else:","* [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0)","dist = e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate","global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on config and learning","from absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS =","mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config = None if","axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0)","= sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e =","tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type","future. steps = np.arange(step - (num_steps - 1) * stride, step + stride,","# Add the batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object):","across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical","frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers for","global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps,","None, \"config file is not provided or is corrupted\" return config def prepare_gpu(ind=-1):","import CONFIG import json import tensorflow as tf import numpy as np import","range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps = seq_len -","one batch to be passed and we don't want to automatically # reset","@tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data through base CNN.\"\"\" if num_steps","* [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps =","on config and learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif","CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list = []","optimizer and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate =","random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate during warm up phase.\"\"\" if","config is not None, \"config file is not provided or is corrupted\" return","len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates","needs to be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True)","tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that provides current learning rate","ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint (if one exists","axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames =","frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset","np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens,","take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w,","based on config. NOTE: This returns a function as in Eager we need","checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(","num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats =","import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def","assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs)","logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir,","\"\"\"Sets the correct learning phase before calling function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls","time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict","batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps,","axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size,","%s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir =","automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for","False return dataset def gen_plot(x, y): \"\"\"Create a pyplot, save to buffer and","= tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im)","return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if","checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects):","emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames =","config import CONFIG import json import tensorflow as tf import numpy as np","Set learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: #","supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step():","ValueError('You might be overwriting a directory that already ' 'has train_logs. Please provide","prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1:","lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in","+ boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0]","import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math import os","* frames_per_batch curr_data = {} for k, v in data.items(): # Need to","random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)),","[] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for","batch size. # We do this as we are embedding the whole sequence","tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles","'has train_logs. Please provide a new logdir name in ' 'config or pass","= range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps","sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r')","os import time from absl import flags from absl import logging from easydict","num_frames_per_step, split): \"\"\"Visualize nearest neighbours in embedding space.\"\"\" # Set learning_phase to False","growth must be set before GPUs have been initialized logging.info(e) os.environ[\"CUDA_VISIBLE_DEVICES\"] = str(ind)","GPUs\"]) except RuntimeError as e: # Memory growth must be set before GPUs","don't want to automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE","from __future__ import division from __future__ import print_function from config import CONFIG import","power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not supported. Only support'","tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats =","im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for","'GPU') try: # Currently, memory growth needs to be the same across GPUs","emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx","permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis),","'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y):","single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get","< warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup *","ValueError('Learning rate decay type %s not supported. Only support' 'the following decay types:","return f(*args, **kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with","a one epoch iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn","emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \"\"\"Create a pyplot, save","lr_fn: function, this can be called to return the current learning rate based","data, training, num_steps=None): \"\"\"Passes data through base CNN.\"\"\" if num_steps is None: if","(num_steps - 1) * stride, step + stride, stride) return steps def get_indices(curr_idx,","def cond(n): if max_embs is None: return True else: return n < max_embs","tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]]","before calling function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the function f after setting","return True else: return n < max_embs # Make Recurrent Layers stateful, set","type %s not supported. Only support' 'the following decay types: fixed, exp_decay, manual,'","\"\"\"Visualizes a batch.\"\"\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries =","training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using","= [] if keep_data: frames_list = [] if optical_flow: frame_original_list = [] n","= tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else","\"\"\"Return a data dict which contains all the requested sequences.\"\"\" data = iterator.get_next()","single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False,","learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer':","config. NOTE: This returns a function as in Eager we need to call","in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in","'Using config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as","steps = np.arange(step - (num_steps - 1) * stride, step + stride, stride)","tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup =","return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data through","embedding space.\"\"\" # Set learning_phase to False to use models in inference mode.","provided config. Raises: ValueError: in case invalid params have been passed in the","[name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels","cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want","* num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats","data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data through base CNN.\"\"\" if","wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f: config","num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) * frames_per_batch","= iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes","set_learning_phase decorator which' ' does not have training argument.') training = kwargs['training'] if","= model['emb'] embs_list = [] labels_list = [] steps_list = [] seq_lens_list =","evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for","data, global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest neighbours in embedding space.\"\"\" #","plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0)","np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0)","take # more than one batch to be passed and we don't want","range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def","= random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning","import numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io","tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in","this as some modalities might not exist. if len(v.shape) > 1 and v.shape[1]","step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e /","= np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names =","1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to","__init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed()","'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does","for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \"\"\"Create","learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict","frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if","tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done =","= tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames']","warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) *","+ num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def","1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed':","= emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx +","support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \"\"\"Intializes learning","1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance)","[] names = [] seq_labels = [] if keep_data: frames = [] if","[] if keep_data: frames = [] if optical_flow: frame_original = [] # Reset","frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs)","to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split,","have training argument.') training = kwargs['training'] if training: # Set learning_phase to True","= [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i","config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing","import math import os import time from absl import flags from absl import","return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay","= steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs","else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats","from absl import flags from absl import logging from easydict import EasyDict import","names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list","return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that provides current learning","global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif","] while cond(n): try: print(n) embs = [] labels = [] steps =","get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1)","{'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames']","'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir)","emb = model['emb'] embs_list = [] labels_list = [] steps_list = [] seq_lens_list","e = np.exp(np.array(w) / t) dist = e / np.sum(e) return dist def","statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for","= False return dataset def gen_plot(x, y): \"\"\"Create a pyplot, save to buffer","to initialize the learning rate and the learning rate decay function. Returns: lr_fn:","num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames","ValueError: in case invalid params have been passed in the config. \"\"\" lr_params","num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 *","overwrite: logging.info( 'Using the existing passed in config as no config.json file exists","callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name","the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU')","space.\"\"\" # Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0)","[] names_list = [] seq_labels_list = [] if keep_data: frames_list = [] if","to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image","= tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently,","data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0]","return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps","= 0 def cond(n): if max_embs is None: return True else: return n","= np.exp(np.array(w) / t) dist = e / np.sum(e) return dist def random_choice_noreplace(m,","RuntimeError as e: # Memory growth must be set before GPUs have been","wrapper(*args, **kwargs): \"\"\"Calls the function f after setting proper learning phase.\"\"\" if 'training'","= tf.expand_dims(image, 0) return image class Stopwatch(object): \"\"\"Simple timer for measuring elapsed time.\"\"\"","try: # Currently, memory growth needs to be the same across GPUs for","__future__ import print_function from config import CONFIG import json import tensorflow as tf","rate decay function. Returns: lr_fn: function, this can be called to return the","= [] names = [] seq_labels = [] if keep_data: frames = []","if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt =","while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir,","* stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps","= len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return","/ warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32)","True else: return n < max_embs # Make Recurrent Layers stateful, set batch","axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames']","learning rate, optimizer and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations","= keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb']","0 def cond(n): if max_embs is None: return True else: return n <","len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels)","optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is","tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path =","logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data,","cycle=False) else: raise ValueError('Learning rate decay type %s not supported. Only support' 'the","**ckpt_objects): \"\"\"Create and restore checkpoint (if one exists on the path).\"\"\" # Instantiate","lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 -","= np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels =","growth needs to be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu,","batch.\"\"\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2)","plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert","as we are embedding the whole sequence and that can take # more","= data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn'","stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps =","'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist.","self.time = time.time() def set_learning_phase(f): \"\"\"Sets the correct learning phase before calling function","float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix,","= True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs = []","to automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru':","lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on","num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the","except RuntimeError as e: # Memory growth must be set before GPUs have","checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint (if one exists on the","after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful =","np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of","logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list,","data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries,","pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\"","for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n):","for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for","axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step)","num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that provides current","dataset def gen_plot(x, y): \"\"\"Create a pyplot, save to buffer and return TB","use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path):","frame_original_list = [] n = 0 def cond(n): if max_embs is None: return","correct learning phase before calling function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the function","np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names,","os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep","curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in","= tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the","optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to","[] labels_list = [] steps_list = [] seq_lens_list = [] names_list = []","manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR","learning rate based on config. NOTE: This returns a function as in Eager","for k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups","max_embs is None: return True else: return n < max_embs # Make Recurrent","provide a new logdir name in ' 'config or pass --force_train while launching","= [] seq_lens_list = [] names_list = [] seq_labels_list = [] if keep_data:","[0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that","steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except","config. \"\"\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def","cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps,","raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .')","'the following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step:","= emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2],","for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting","buf.seek(0) # Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) #","models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config","\"config file is not provided or is corrupted\" return config def prepare_gpu(ind=-1): ind","passed and we don't want to automatically # reset hidden states after each","+ stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx","vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0),","== 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False)","in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats =","cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS","if training: # Set learning_phase to True to use models in training mode.","warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int","single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True,","{} for k, v in data.items(): # Need to do this as some","cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2","config and learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE","global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float","+ 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE ==","\"\"\"Create a pyplot, save to buffer and return TB compatible image.\"\"\" plt.figure() plt.plot(x,","ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be","for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config","np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original","= tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return","'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence =","def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step)","learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn']","len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names)","np.exp(np.array(w) / t) dist = e / np.sum(e) return dist def random_choice_noreplace(m, n,","self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \"\"\"Sets the correct","else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32)","with set_learning_phase decorator which' ' does not have training argument.') training = kwargs['training']","and return TB compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout()","summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix","secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file)","if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert","[(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return","cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats(","function as in Eager we need to call assign to update the learning","lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)()","config. Raises: ValueError: in case invalid params have been passed in the config.","'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep '","labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0)","= {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data:","y): \"\"\"Create a pyplot, save to buffer and return TB compatible image.\"\"\" plt.figure()","/ t) dist = e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1):","keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list","video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len","curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps","lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step):","keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb =","get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for","in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1)","isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k,","config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else:","1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps),","can be called to return the current learning rate based on the provided","def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time() def","that already ' 'has train_logs. Please provide a new logdir name in '","tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to","step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0,","data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label])","curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data,","i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data","= tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint","in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i],","frame_original = [] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE ==","to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb']","+ 1) * frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch","' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file:","as plt # pylint: disable=g-import-not-at-top import io import math import os import time","* warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup)","current learning rate based on the provided config. Raises: ValueError: in case invalid","training, num_steps=None): \"\"\"Passes data through base CNN.\"\"\" if num_steps is None: if training:","might be overwriting a directory that already ' 'has train_logs. Please provide a","emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps,","= frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list #","f: config = json.load(f) assert config is not None, \"config file is not","seq_lens = [] names = [] seq_labels = [] if keep_data: frames =","= frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers","k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory","[] seq_labels = [] if keep_data: frames = [] if optical_flow: frame_original =","We do this as we are embedding the whole sequence and that can","tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list,","and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes,","= tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step -","= [] if optical_flow: frame_original = [] # Reset GRU states for each","as config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file,","assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels)","neighbours in embedding space.\"\"\" # Set learning_phase to False to use models in","def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from a","pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step,","= tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as e: #","is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES","Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix'","i * frames_per_batch curr_data = {} for k, v in data.items(): # Need","keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names =","in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to","tf import numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import","networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10,","create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we can insert multiple","1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list =","be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus =","modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir,","emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1)",":cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate during warm up","num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps =","- 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps,","to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase","io import math import os import time from absl import flags from absl","else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for","im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix","0) return image class Stopwatch(object): \"\"\"Simple timer for measuring elapsed time.\"\"\" def __init__(self):","seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original =","in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches =","'%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k,","axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step]","gen_plot(x, y): \"\"\"Create a pyplot, save to buffer and return TB compatible image.\"\"\"","seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels =","steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs =","Currently, memory growth needs to be the same across GPUs for gpu in","json.load(f) assert config is not None, \"config file is not provided or is","frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if","[seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0]","# Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return","seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels:","as no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as","tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image, 0) return image","the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list,","in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step -","GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except","= emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if","opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else:","v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for","or is corrupted\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU')","= tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img","lr_params): \"\"\"Returns learning rate during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int","phase.\"\"\" if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator which'","False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb =","lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def","return image class Stopwatch(object): \"\"\"Simple timer for measuring elapsed time.\"\"\" def __init__(self): self.reset()","CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator)","return lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates):","the learning rate. Args: optimizer_config: EasyDict, contains params required to initialize the learning","tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float =","keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0)","not supported. Only support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.')","gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs =","np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0])","initialize the learning rate and the learning rate decay function. Returns: lr_fn: function,","logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as e:","% split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take","# np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len)","tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the following'","axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames","* num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that provides","= frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps)","'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate):","number %d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy())","set batch size. # We do this as we are embedding the whole","axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles = random_choice_noreplace(","json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a","get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data through base CNN.\"\"\" if num_steps is None:","frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) ==","to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer","json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You","def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return","absl import flags from absl import logging from easydict import EasyDict import matplotlib","rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates *","+= 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list,","evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False","restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore","to be passed and we don't want to automatically # reset hidden states","measuring elapsed time.\"\"\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def","cycles for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr,","force_train: raise ValueError('You might be overwriting a directory that already ' 'has train_logs.","[num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1,","'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a directory","seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1","im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in","= [] seq_lens = [] names = [] seq_labels = [] if keep_data:",">= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \"\"\"Sets the correct learning","else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data dict which contains all the","= np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels","for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v","single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps =","config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def","frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels:","* num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats,","else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO:","= labels_list # Reset statefulness to recurrent layers for other evaluation tasks. if","logging.debug('On sequence number %d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy',","cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We","all the requested sequences.\"\"\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def","base CNN.\"\"\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS","image = tf.expand_dims(image, 0) return image class Stopwatch(object): \"\"\"Simple timer for measuring elapsed","axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats =","params required to initialize the learning rate and the learning rate decay function.","Stopwatch(object): \"\"\"Simple timer for measuring elapsed time.\"\"\" def __init__(self): self.reset() def elapsed(self): return","range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2","training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps","= tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done","while cond(n): try: print(n) embs = [] labels = [] steps = []","same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS),","ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create","size. # We do this as we are embedding the whole sequence and","keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] =","%d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) #","v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json","single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from","emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step,","num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step -","True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs = [] labels","from a one epoch iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS","matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math import os import","that provides current learning rate based on config. NOTE: This returns a function","keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list):","if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config as","1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens':","# Currently, memory growth needs to be the same across GPUs for gpu","is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr","\"\"\"Calls the function f after setting proper learning phase.\"\"\" if 'training' not in","data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats =","Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in","not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep ' ' %s","range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE","phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS,","warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR *","is not None, \"config file is not provided or is corrupted\" return config","if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE,","return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings","recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in","# Instantiate checkpoint and restore from any pre-existing checkpoint. # Since model is","None: return True else: return n < max_embs # Make Recurrent Layers stateful,","dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO:","global_step, lr_params): \"\"\"Returns learning rate during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0:","learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr,","def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise","return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f:","gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try:","= tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr","tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup","from __future__ import absolute_import from __future__ import division from __future__ import print_function from","GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: #","tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep ' ' %s for","embeddings from a one epoch iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step =","* warmup_lr return lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step,","for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels","pylint: disable=g-import-not-at-top import io import math import os import time from absl import","image class Stopwatch(object): \"\"\"Simple timer for measuring elapsed time.\"\"\" def __init__(self): self.reset() def","data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data through base CNN.\"\"\"","config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file:","def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e / np.sum(e)","through base CNN.\"\"\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES *","[] if keep_data: frames_list = [] if optical_flow: frame_original_list = [] n =","lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES]","n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size,","seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn,","tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return","each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps,","from config import CONFIG import json import tensorflow as tf import numpy as","= [] labels = [] steps = [] seq_lens = [] names =","if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert","steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps =","= [] if optical_flow: frame_original_list = [] n = 0 def cond(n): if","to be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus","- 1) * stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps,","= checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c)","embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names':","setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if","# Generate m random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis)","time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data dict which contains all","CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats","batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1)","axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original =","buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image","num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps =","= tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j","'Using the existing passed in config as no config.json file exists in '","to update the learning rate. Args: optimizer_config: EasyDict, contains params required to initialize","np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels =","global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we can","argument.') training = kwargs['training'] if training: # Set learning_phase to True to use","setting proper learning phase.\"\"\" if 'training' not in kwargs: raise ValueError('Function called with","0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k]","TB compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf =","return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return","checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return","# Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32)","Returns: lr_fn: function, this can be called to return the current learning rate","embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps)","if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)]","logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k, v","0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int,","cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles","warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32)","mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats","for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps =","and the learning rate decay function. Returns: lr_fn: function, this can be called","global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32)","warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr","== 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate,","insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager(","learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE ==","EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes","if keep_data: frames_list = [] if optical_flow: frame_original_list = [] n = 0","embedding the whole sequence and that can take # more than one batch","def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval): return","def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config,","[] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean(","CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames'])","have been passed in the config. \"\"\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda","CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1,","return dict([(k, to_dict(v)) for k, v in config.items()]) else: return config def setup_train_dir(logdir,","visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames = data['frames'] frames_list = tf.unstack(frames,","following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer","training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models in","which contains all the requested sequences.\"\"\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens']","= os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed","if keep_data: frames = [] if optical_flow: frame_original = [] # Reset GRU","iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb","load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f)","doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def","= cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE #","if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps","tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data,","curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps,","= kwargs['training'] if training: # Set learning_phase to True to use models in","use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if","get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step","len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps)","gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs = [] labels =","not have training argument.') training = kwargs['training'] if training: # Set learning_phase to","-1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be the same","Eager we need to call assign to update the learning rate. Args: optimizer_config:","rate and the learning rate decay function. Returns: lr_fn: function, this can be","directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite:","if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break","called to return the current learning rate based on the provided config. Raises:","+ num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'),","plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(),","axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e","# Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer","config to exist. Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds)","logging.info('Waiting for config to exist. Going to sleep ' ' %s for secs.',","steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps","the correct learning phase before calling function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the","- 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps,","axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img)","for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list","elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f =","with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs')","# pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr,","callbacks=[]): \"\"\"Get embeddings from a one epoch iterator.\"\"\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS","steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels","np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps","want to automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE ==","global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not","plt # pylint: disable=g-import-not-at-top import io import math import os import time from","labels_list # Reset statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE","num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' %","compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO()","# Since model is a dict we can insert multiple modular networks in","summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im,","dict we can insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects)","= np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens =","else: logging.info( 'Using config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path,","for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats,","axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32 as","def get_data(iterator): \"\"\"Return a data dict which contains all the requested sequences.\"\"\" data","return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer","if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list =","* tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \"\"\"Returns function that provides current learning rate based","* frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps","in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0]","restore=False, **ckpt_objects): # Since model is a dict we can insert multiple modular","and that can take # more than one batch to be passed and","in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2]","CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists","'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data,","cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE","summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step)","keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from a one epoch iterator.\"\"\" keep_labels = keep_labels","lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR)","with open(config_path) as f: config = json.load(f) assert config is not None, \"config","curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j]","= CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list =","inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config = None","None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data dict which contains","= CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the future.","1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size):","= [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix =","status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def","# np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) #","not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config as no","axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if","== 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ]","learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set","reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in","embs_list = [] labels_list = [] steps_list = [] seq_lens_list = [] names_list","== len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens)","and restore checkpoint (if one exists on the path).\"\"\" # Instantiate checkpoint and","assert config is not None, \"config file is not provided or is corrupted\"","opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER,","while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep ' '","keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from a one epoch iterator.\"\"\" keep_labels","as some modalities might not exist. if len(v.shape) > 1 and v.shape[1] !=","don't want to see the future. steps = np.arange(step - (num_steps - 1)","status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint (if one exists on","AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and global","CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step):","data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None):","__future__ import division from __future__ import print_function from config import CONFIG import json","config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator):","np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0])","config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir,","np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy',","frames = [] if optical_flow: frame_original = [] # Reset GRU states for","num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step","tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models in inference mode.","GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs","see the future. steps = np.arange(step - (num_steps - 1) * stride, step","json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists in","* num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step","EasyDict, contains params required to initialize the learning rate and the learning rate","idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps,","image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image,","random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \"\"\"Returns learning rate","> 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] =","to exist. Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while","[] if optical_flow: frame_original_list = [] n = 0 def cond(n): if max_embs","training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats =","we don't want to automatically # reset hidden states after each batch. if","checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint)","as f: config = json.load(f) assert config is not None, \"config file is","mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1)","lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr,","tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as e: # Memory","indent=4) else: logging.info( 'Using config from config.json that exists in %s.', logdir) with","elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0,","= [] seq_labels_list = [] if keep_data: frames_list = [] if optical_flow: frame_original_list","(batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if","num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries =","tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config):","= np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data:","idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] =","lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning","c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in","manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config,","model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 *","== 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence","get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps)","config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind","if isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return","model['cnn'] emb = model['emb'] embs_list = [] labels_list = [] steps_list = []","buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension","directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if","in range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps = seq_len","lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float","tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \"\"\"Visualize nearest","model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data,","a directory that already ' 'has train_logs. Please provide a new logdir name","tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img =","required to initialize the learning rate and the learning rate decay function. Returns:","lr + is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow object_detection","iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from a one epoch","is_warmup) * lr + is_warmup * warmup_lr return lr # Minimally adapted from","stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps)","in data.items(): # Need to do this as some modalities might not exist.","a new logdir name in ' 'config or pass --force_train while launching script.')","disable=g-import-not-at-top import io import math import os import time from absl import flags","no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file:","seq_labels = [] if keep_data: frames = [] if optical_flow: frame_original = []","n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs':","(if one exists on the path).\"\"\" # Instantiate checkpoint and restore from any","the existing passed in config as no config.json file exists in ' '%s',","tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn'","len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if","NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate","= emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps *","config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True,","sequence number %d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'),","seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps","[] labels = [] steps = [] seq_lens = [] names = []","'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] =","Instantiate checkpoint and restore from any pre-existing checkpoint. # Since model is a","with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds)","tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i],","not None, \"config file is not provided or is corrupted\" return config def","= tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data,","types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step,","num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list","num_steps = seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx =","else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats,","not force_train: raise ValueError('You might be overwriting a directory that already ' 'has","batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step,","Generate m random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return","def get_cnn_feats(cnn, data, training, num_steps=None): \"\"\"Passes data through base CNN.\"\"\" if num_steps is","in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif","frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1)","axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0)",".') return opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and global step.\"\"\" optimizer","optical_flow: frame_original = [] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE","'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE","name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels)","= embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) ==","= seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1)","= lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)]","candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0)","labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names","= lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0","in kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does not have","seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if","decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step),","lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual':","np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps)","corrupted\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS:","mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models in inference","def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range (0, n)","to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return","frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps =","num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries))","def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on config and learning rate.\"\"\" if optimizer_config.TYPE","Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0]","break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list}","max_embs=None, callbacks=[]): \"\"\"Get embeddings from a one epoch iterator.\"\"\" keep_labels = keep_labels and","config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists in %s.',","CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0)","in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \"\"\"Create a pyplot,","= [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step):","tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries","num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames","decay type %s not supported. Only support' 'the following decay types: fixed, exp_decay,","seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in","a batch.\"\"\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list,","if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32)","= tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index,","lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries","matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames","Make Recurrent Layers stateful, set batch size. # We do this as we","get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on config and","== 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif","return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint (if one","[] steps = [] seq_lens = [] names = [] seq_labels = []","warmup_lr return lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries,","function f after setting proper learning phase.\"\"\" if 'training' not in kwargs: raise","np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return","learning rate during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step,","= tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image, 0) return","if optical_flow: frame_original = [] # Reset GRU states for each video. if","from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step,","\"\"\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr,","axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0)","calling function f.\"\"\" def wrapper(*args, **kwargs): \"\"\"Calls the function f after setting proper","Recurrent Layers stateful, set batch size. # We do this as we are","any pre-existing checkpoint. # Since model is a dict we can insert multiple","ValueError('Function called with set_learning_phase decorator which' ' does not have training argument.') training","[0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)),","**ckpt_objects): # Since model is a dict we can insert multiple modular networks","global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr,","frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \"\"\"Get embeddings from a one epoch iterator.\"\"\"","seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx","tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries))","def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and global step.\"\"\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR)","tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path):","chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len","'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9)","tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError","params have been passed in the config. \"\"\" lr_params = optimizer_config.LR # pylint:","directory that already ' 'has train_logs. Please provide a new logdir name in","= (1.0 - is_warmup) * lr + is_warmup * warmup_lr return lr #","as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math","Add the batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \"\"\"Simple","and not force_train: raise ValueError('You might be overwriting a directory that already '","= data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list =","> seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps = frames_per_batch","is a dict we can insert multiple modular networks in this dict. checkpoint","CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps =","os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a directory that already","tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step *","decay function. Returns: lr_fn: function, this can be called to return the current","for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus),","% split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t)","rate. Args: optimizer_config: EasyDict, contains params required to initialize the learning rate and","axis=-1): # Generate m random permuations of range (0, n) # NumPy version:","= optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects):","dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list","json import tensorflow as tf import numpy as np import matplotlib.pyplot as plt","sequence and that can take # more than one batch to be passed","print_function from config import CONFIG import json import tensorflow as tf import numpy","some modalities might not exist. if len(v.shape) > 1 and v.shape[1] != 0:","names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs,","**kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as","- self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time =","1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1])","optimizer_config: EasyDict, contains params required to initialize the learning rate and the learning","during warm up phase.\"\"\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int","import json import tensorflow as tf import numpy as np import matplotlib.pyplot as","def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f: config =","in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1)","[] seq_lens_list = [] names_list = [] seq_labels_list = [] if keep_data: frames_list","1) * stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len):","the current learning rate based on the provided config. Raises: ValueError: in case","in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that","contains all the requested sequences.\"\"\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function","be called to return the current learning rate based on the provided config.","sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists in %s.', logdir)","[image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in","2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else:","def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index","axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1],","exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps,","num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model,","be overwriting a directory that already ' 'has train_logs. Please provide a new","else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \"\"\"Setups directory for training.\"\"\" tf.io.gfile.makedirs(logdir) config_path","split): \"\"\"Visualize nearest neighbours in embedding space.\"\"\" # Set learning_phase to False to","to do this as some modalities might not exist. if len(v.shape) > 1","the learning rate and the learning rate decay function. Returns: lr_fn: function, this","want to see the future. steps = np.arange(step - (num_steps - 1) *","True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is None:","num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step,","assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)]","which' ' does not have training argument.') training = kwargs['training'] if training: #","after setting proper learning phase.\"\"\" if 'training' not in kwargs: raise ValueError('Function called","config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if","if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS","' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as","tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else:","axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)]","= get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if","data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in","# Set learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else:","= data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list =","tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()])","if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a directory that","def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir,","'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \"\"\"Intializes learning rate, optimizer and","in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1):","= np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats =","def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we can insert","'w') as config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config,","frames_list = [] if optical_flow: frame_original_list = [] n = 0 def cond(n):","assign to update the learning rate. Args: optimizer_config: EasyDict, contains params required to","= seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original","num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats =","this can be called to return the current learning rate based on the","= i * frames_per_batch curr_data = {} for k, v in data.items(): #","tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for","return n < max_embs # Make Recurrent Layers stateful, set batch size. #","time from absl import flags from absl import logging from easydict import EasyDict","update the learning rate. Args: optimizer_config: EasyDict, contains params required to initialize the","emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx + num_steps)","CNN.\"\"\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else:","os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config is not None,","return dataset def gen_plot(x, y): \"\"\"Create a pyplot, save to buffer and return","axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)]","matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\"","query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list","# We do this as we are embedding the whole sequence and that","if optical_flow: frame_original_list = [] n = 0 def cond(n): if max_embs is","timer for measuring elapsed time.\"\"\" def __init__(self): self.reset() def elapsed(self): return time.time() -","data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs =","- i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch","curr_data = {} for k, v in data.items(): # Need to do this","rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt","in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models","seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches):","Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries","else: return n < max_embs # Make Recurrent Layers stateful, set batch size.","= get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 *","= get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence","is corrupted\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if","= names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)]","list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v))","' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \"\"\"Setups","if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps)","'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while","tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager,","[1, ] while cond(n): try: print(n) embs = [] labels = [] steps","to buffer and return TB compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0,","if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and","= tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0)","optical_flow: frame_original_list = [] n = 0 def cond(n): if max_embs is None:","keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert","np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels =","len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k]","call assign to update the learning rate. Args: optimizer_config: EasyDict, contains params required","learning rate. Args: optimizer_config: EasyDict, contains params required to initialize the learning rate","pyplot, save to buffer and return TB compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val","' 'has train_logs. Please provide a new logdir name in ' 'config or","mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)])","channels=4) # Add the batch dimension image = tf.expand_dims(image, 0) return image class","1) * frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch else:","this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status","= e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m","dataset['labels'] = labels_list # Reset statefulness to recurrent layers for other evaluation tasks.","%d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"])","and restore from any pre-existing checkpoint. # Since model is a dict we","directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path):","= np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0])","* CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps","dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent","elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly':","axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im","def gen_cycles(num_cycles, batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:,","tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False)","for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert","'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else:","decorator which' ' does not have training argument.') training = kwargs['training'] if training:","!= 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else:","dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \"\"\"Simple timer for measuring","save to buffer and return TB compatible image.\"\"\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy')","lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s","data dict which contains all the requested sequences.\"\"\" data = iterator.get_next() return data,","support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr,","names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if","Reset statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru':","config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \"\"\"Return a data dict","get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on config and learning rate.\"\"\" if optimizer_config.TYPE ==","the requested sequences.\"\"\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn,","np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for","if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names","poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns","[int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p","seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError:","checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config):","models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn'","i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list =","contains params required to initialize the learning rate and the learning rate decay","nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im","axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False)","frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data = {}","self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed() >=","optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay(","function that provides current learning rate based on config. NOTE: This returns a","use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to","stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps =","img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list","as in Eager we need to call assign to update the learning rate.","rate decay type %s not supported. Only support' 'the following decay types: fixed,","elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s","num_steps=None): \"\"\"Passes data through base CNN.\"\"\" if num_steps is None: if training: num_steps","None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config is","CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps","seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) *","whole sequence and that can take # more than one batch to be","im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32","= model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2","# Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries =","* f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping(","= tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step,","restore checkpoint (if one exists on the path).\"\"\" # Instantiate checkpoint and restore","%s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict","cond(n): if max_embs is None: return True else: return n < max_embs #","= io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image image","image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step","n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\"frames\"]) #","if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset","frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens)","we need to call assign to update the learning rate. Args: optimizer_config: EasyDict,","embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] =","tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int,","on config. NOTE: This returns a function as in Eager we need to","do this as we are embedding the whole sequence and that can take","import time from absl import flags from absl import logging from easydict import","to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def","range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1,","if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size,","axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats","def visualize_batch(data, global_step, batch_size, num_steps): \"\"\"Visualizes a batch.\"\"\" frames = data['frames'] frames_list =","in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in","global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \"\"\"Returns optimizer based on config","\"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"]) except RuntimeError as e: # Memory growth must","= [] if keep_data: frames = [] if optical_flow: frame_original = [] #","momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer,","in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \"Physical GPUs,\", len(logical_gpus), \"Logical GPUs\"])","is None: return True else: return n < max_embs # Make Recurrent Layers","other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful =","learning rate and the learning rate decay function. Returns: lr_fn: function, this can","def set_learning_phase(f): \"\"\"Sets the correct learning phase before calling function f.\"\"\" def wrapper(*args,","\"\"\"Setups directory for evaluation.\"\"\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not","import absolute_import from __future__ import division from __future__ import print_function from config import","we can insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager","def get_lr_fn(optimizer_config): \"\"\"Returns function that provides current learning rate based on config. NOTE:","nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im =","\"\"\"Returns optimizer based on config and learning rate.\"\"\" if optimizer_config.TYPE == 'AdamOptimizer': opt","if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original)","\"\"\"Visualize nearest neighbours in embedding space.\"\"\" # Set learning_phase to False to use","EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else: return config def","else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \"\"\"Create and restore checkpoint","candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps","batch_size, cycle_len): \"\"\"Generate cycles for alignment.\"\"\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return"],"string":"[\n \"curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats =\",\n \"' does not have training argument.') training = kwargs['training'] if training: # Set\",\n \"stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx +\",\n \"\\\"\\\"\\\"Returns function that provides current learning rate based on config. NOTE: This returns\",\n \"a pyplot, save to buffer and return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y)\",\n \"lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE\",\n \"\\\"\\\"\\\"Intializes learning rate, optimizer and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step =\",\n \"gru_layer.stateful = False return dataset def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save to\",\n \"status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return\",\n \"tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if\",\n \"len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)]\",\n \"division from __future__ import print_function from config import CONFIG import json import tensorflow\",\n \"train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be\",\n \"config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict =\",\n \"'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step\",\n \"softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im,\",\n \"steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)]\",\n \"v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1)\",\n \"steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original']\",\n \"False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper\",\n \"global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\" # Set\",\n \"tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def\",\n \"returns a function as in Eager we need to call assign to update\",\n \"return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range\",\n \"to see the future. steps = np.arange(step - (num_steps - 1) * stride,\",\n \"warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int =\",\n \"[image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros(\",\n \"as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict)\",\n \"- (num_steps - 1) * stride, step + stride, stride) return steps def\",\n \"tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current learning rate based on\",\n \"called with set_learning_phase decorator which' ' does not have training argument.') training =\",\n \"elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else: return\",\n \"emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len *\",\n \"frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list\",\n \"opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support\",\n \"[] n = 0 def cond(n): if max_embs is None: return True else:\",\n \"to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb\",\n \"= [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries),\",\n \"global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step):\",\n \"y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) #\",\n \"and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE ==\",\n \"' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for\",\n \"file is not provided or is corrupted\\\" return config def prepare_gpu(ind=-1): ind =\",\n \"frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list\",\n \"'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len\",\n \"emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames,\",\n \"n < max_embs # Make Recurrent Layers stateful, set batch size. # We\",\n \"dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels':\",\n \"than one batch to be passed and we don't want to automatically #\",\n \"labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens =\",\n \"for i in range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps\",\n \"global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on config and learning rate.\\\"\\\"\\\"\",\n \"to return the current learning rate based on the provided config. Raises: ValueError:\",\n \"global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return\",\n \"seq_lens_list = [] names_list = [] seq_labels_list = [] if keep_data: frames_list =\",\n \"path).\\\"\\\"\\\" # Instantiate checkpoint and restore from any pre-existing checkpoint. # Since model\",\n \"= dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else:\",\n \"depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current learning rate based on config.\",\n \"in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n)\",\n \"config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config\",\n \"1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v,\",\n \"function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f after setting proper learning\",\n \"to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False\",\n \"as tf import numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top\",\n \"\\\"\\\"\\\"Returns learning rate during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int =\",\n \"axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize\",\n \"we are embedding the whole sequence and that can take # more than\",\n \"been passed in the config. \\\"\\\"\\\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if\",\n \"gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \\\"\\\"\\\"Create a\",\n \"%s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt\",\n \"'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if\",\n \"k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from\",\n \"seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow:\",\n \"time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase before calling function f.\\\"\\\"\\\" def\",\n \"batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \\\"\\\"\\\"Simple timer for\",\n \"def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json')\",\n \"= tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model,\",\n \"embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow:\",\n \"names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames =\",\n \"= np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps)))))\",\n \"m random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m,\",\n \"TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image =\",\n \"= num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride\",\n \"np.arange(step - (num_steps - 1) * stride, step + stride, stride) return steps\",\n \"if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step\",\n \"embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len])\",\n \"tf.expand_dims(image, 0) return image class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\" def\",\n \"each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True\",\n \"axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split,\",\n \"list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function\",\n \"iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data\",\n \"tensorflow as tf import numpy as np import matplotlib.pyplot as plt # pylint:\",\n \"> -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be the\",\n \"num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate during\",\n \"im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix\",\n \"names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames =\",\n \"raise ValueError('You might be overwriting a directory that already ' 'has train_logs. Please\",\n \"== 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x,\",\n \"PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch\",\n \"num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in\",\n \"batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def\",\n \"optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD(\",\n \"= [] seq_labels = [] if keep_data: frames = [] if optical_flow: frame_original\",\n \"keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers for other evaluation\",\n \"disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS,\",\n \"2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats =\",\n \"optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not\",\n \"= tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch',\",\n \"def reset(self): self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase before\",\n \"assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if\",\n \"in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config =\",\n \"* frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data =\",\n \"= json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise\",\n \"global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x\",\n \"get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current learning rate based on config. NOTE: This\",\n \"Please provide a new logdir name in ' 'config or pass --force_train while\",\n \"if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config is not\",\n \"frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size,\",\n \"int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) * frames_per_batch > seq_len:\",\n \"\\\"\\\"\\\"Get embeddings from a one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step\",\n \"len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data:\",\n \"= np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch,\",\n \"labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)]\",\n \"return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf\",\n \"return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for\",\n \"Need to do this as some modalities might not exist. if len(v.shape) >\",\n \"io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image image =\",\n \"def elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval\",\n \"tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries,\",\n \"* lr + is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow\",\n \"emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn,\",\n \"**kwargs): \\\"\\\"\\\"Calls the function f after setting proper learning phase.\\\"\\\"\\\" if 'training' not\",\n \"= np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames,\",\n \"= int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) * frames_per_batch >\",\n \"tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer\",\n \"optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset\",\n \"curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy',\",\n \"/ np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations\",\n \"image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image, 0)\",\n \"batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape\",\n \"if max_embs is None: return True else: return n < max_embs # Make\",\n \"hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers:\",\n \"FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames =\",\n \"absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS\",\n \"checkpoint and restore from any pre-existing checkpoint. # Since model is a dict\",\n \"batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\" # Set learning_phase\",\n \"= None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config\",\n \"exist. Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True:\",\n \"t=1.0): e = np.exp(np.array(w) / t) dist = e / np.sum(e) return dist\",\n \"ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind],\",\n \"config as no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w')\",\n \"return the current learning rate based on the provided config. Raises: ValueError: in\",\n \"f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f after setting proper learning phase.\\\"\\\"\\\"\",\n \"= [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) *\",\n \"= np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len,\",\n \"new logdir name in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir)\",\n \"return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c\",\n \"seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i +\",\n \"target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets\",\n \"\\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or\",\n \"gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches\",\n \"Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes\",\n \"ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in\",\n \"data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data through base\",\n \"training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n,\",\n \"in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps)\",\n \"[] steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list = []\",\n \"# Reset statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE ==\",\n \"seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label\",\n \"is not provided or is corrupted\\\" return config def prepare_gpu(ind=-1): ind = int(ind)\",\n \"embs = [] labels = [] steps = [] seq_lens = [] names\",\n \"CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def\",\n \"train_logs. Please provide a new logdir name in ' 'config or pass --force_train\",\n \"'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f\",\n \"class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self):\",\n \"import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps):\",\n \"to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path,\",\n \"else: # Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0)\",\n \"optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we\",\n \"not in kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does not\",\n \"def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS\",\n \"elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval def\",\n \"easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size,\",\n \"logging.info( 'Using the existing passed in config as no config.json file exists in\",\n \"if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len =\",\n \"config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going\",\n \"num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps):\",\n \"warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup * warmup_lr\",\n \"step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in\",\n \"frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.')\",\n \"== 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len\",\n \"random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range (0, n) #\",\n \"frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) ==\",\n \"emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False,\",\n \"len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as e: # Memory growth must be set\",\n \"os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in\",\n \"== 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported.\",\n \"in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()])\",\n \"elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self,\",\n \"tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr =\",\n \"frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert\",\n \"n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles =\",\n \"CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]]\",\n \"names_list = [] seq_labels_list = [] if keep_data: frames_list = [] if optical_flow:\",\n \"boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns\",\n \"steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list = [] if\",\n \"* CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats =\",\n \"learning phase before calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f\",\n \"Args: optimizer_config: EasyDict, contains params required to initialize the learning rate and the\",\n \"- is_warmup) * lr + is_warmup * warmup_lr return lr # Minimally adapted\",\n \"num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for k,\",\n \"Only support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda\",\n \"lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS,\",\n \"time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time\",\n \"of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64)\",\n \"%s not supported. Only support' 'the following decay types: fixed, exp_decay, manual,' 'and\",\n \"# np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks:\",\n \"(i + 1) * frames_per_batch > seq_len: num_steps = seq_len - i *\",\n \"from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict\",\n \"j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im):\",\n \"if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding\",\n \"one exists on the path).\\\"\\\"\\\" # Instantiate checkpoint and restore from any pre-existing\",\n \"does not have training argument.') training = kwargs['training'] if training: # Set learning_phase\",\n \"and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list =\",\n \"for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries,\",\n \"isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else: return config\",\n \"import io import math import os import time from absl import flags from\",\n \"restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if one exists on the path).\\\"\\\"\\\" #\",\n \"else: raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer\",\n \"if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to\",\n \"# NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len):\",\n \"if (i + 1) * frames_per_batch > seq_len: num_steps = seq_len - i\",\n \"\\\"Logical GPUs\\\"]) except RuntimeError as e: # Memory growth must be set before\",\n \"data.items(): # Need to do this as some modalities might not exist. if\",\n \"# reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer\",\n \"== len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames)\",\n \"None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES *\",\n \"CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats,\",\n \"need to call assign to update the learning rate. Args: optimizer_config: EasyDict, contains\",\n \"frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for k, v in\",\n \"on the provided config. Raises: ValueError: in case invalid params have been passed\",\n \"do this as some modalities might not exist. if len(v.shape) > 1 and\",\n \"steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name])\",\n \"global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate\",\n \"plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF\",\n \"return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries =\",\n \"launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs'))\",\n \"num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats =\",\n \"import print_function from config import CONFIG import json import tensorflow as tf import\",\n \"model['emb'] embs_list = [] labels_list = [] steps_list = [] seq_lens_list = []\",\n \"= model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn,\",\n \"as e: # Memory growth must be set before GPUs have been initialized\",\n \"curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d,\",\n \"softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e / np.sum(e) return\",\n \"boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] *\",\n \"proper learning phase.\\\"\\\"\\\" if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase\",\n \"step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix =\",\n \"for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful\",\n \"single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx +\",\n \"inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO:\",\n \"the function f after setting proper learning phase.\\\"\\\"\\\" if 'training' not in kwargs:\",\n \"adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] +\",\n \"target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase\",\n \"not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def\",\n \"CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir,\",\n \"= tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint)\",\n \"global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int,\",\n \"exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def\",\n \"the path).\\\"\\\"\\\" # Instantiate checkpoint and restore from any pre-existing checkpoint. # Since\",\n \"get_data(iterator): \\\"\\\"\\\"Return a data dict which contains all the requested sequences.\\\"\\\"\\\" data =\",\n \"tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be the same across\",\n \"= np.arange(step - (num_steps - 1) * stride, step + stride, stride) return\",\n \"tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def\",\n \"= get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v\",\n \"seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames\",\n \"== len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps =\",\n \"done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f):\",\n \"lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def\",\n \"CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = []\",\n \"max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status,\",\n \"* num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2\",\n \"seq_labels_list = [] if keep_data: frames_list = [] if optical_flow: frame_original_list = []\",\n \"and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate\",\n \"= [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance =\",\n \"existing passed in config as no config.json file exists in ' '%s', logdir)\",\n \"config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4)\",\n \"num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats\",\n \"lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay(\",\n \"seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len *\",\n \"sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e\",\n \"return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a\",\n \"[seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps,\",\n \"chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in\",\n \"e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random\",\n \"setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json')\",\n \"'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step):\",\n \"force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path)\",\n \"the whole sequence and that can take # more than one batch to\",\n \"= [] labels_list = [] steps_list = [] seq_lens_list = [] names_list =\",\n \"= tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float\",\n \"to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using\",\n \"in the config. \\\"\\\"\\\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE ==\",\n \"dict([(k, to_dict(v)) for k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False,\",\n \"tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1\",\n \"def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f after setting proper learning phase.\\\"\\\"\\\" if\",\n \"tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't\",\n \"gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for\",\n \"v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps)\",\n \"frames_per_batch curr_data = {} for k, v in data.items(): # Need to do\",\n \"in config as no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path,\",\n \"on the path).\\\"\\\"\\\" # Instantiate checkpoint and restore from any pre-existing checkpoint. #\",\n \"following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr,\",\n \"directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager,\",\n \"data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats\",\n \"CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data dict which contains all the requested sequences.\\\"\\\"\\\"\",\n \"n = 0 def cond(n): if max_embs is None: return True else: return\",\n \"We don't want to see the future. steps = np.arange(step - (num_steps -\",\n \"or overwrite: logging.info( 'Using the existing passed in config as no config.json file\",\n \"return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0)\",\n \"> 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float =\",\n \"step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate,\",\n \"with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k, v in\",\n \"logdir name in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def\",\n \"a function as in Eager we need to call assign to update the\",\n \"cond(n): try: print(n) embs = [] labels = [] steps = [] seq_lens\",\n \"plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png')\",\n \"# Memory growth must be set before GPUs have been initialized logging.info(e) os.environ[\\\"CUDA_VISIBLE_DEVICES\\\"]\",\n \"def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if one exists on the path).\\\"\\\"\\\"\",\n \"overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not\",\n \"flags from absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS\",\n \"kwargs['training'] if training: # Set learning_phase to True to use models in training\",\n \"tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split):\",\n \"provides current learning rate based on config. NOTE: This returns a function as\",\n \"= tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step)\",\n \"= data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs\",\n \"else: raise ValueError('Learning rate decay type %s not supported. Only support' 'the following\",\n \"data through base CNN.\\\"\\\"\\\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES\",\n \"keep_data: frames = [] if optical_flow: frame_original = [] # Reset GRU states\",\n \"axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list\",\n \"e: # Memory growth must be set before GPUs have been initialized logging.info(e)\",\n \"f after setting proper learning phase.\\\"\\\"\\\" if 'training' not in kwargs: raise ValueError('Function\",\n \"= [] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru':\",\n \"split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist\",\n \"cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy())\",\n \"passed in the config. \\\"\\\"\\\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE\",\n \"= [] n = 0 def cond(n): if max_embs is None: return True\",\n \"in case invalid params have been passed in the config. \\\"\\\"\\\" lr_params =\",\n \"global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries\",\n \"can insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager =\",\n \"= [] steps = [] seq_lens = [] names = [] seq_labels =\",\n \"and we don't want to automatically # reset hidden states after each batch.\",\n \"embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'),\",\n \"for config to exist. Going to sleep ' ' %s for secs.', config_timeout_seconds)\",\n \"dict which contains all the requested sequences.\\\"\\\"\\\" data = iterator.get_next() return data, data['chosen_steps'],\",\n \"states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states()\",\n \"nearest neighbours in embedding space.\\\"\\\"\\\" # Set learning_phase to False to use models\",\n \"for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict =\",\n \"learning rate based on the provided config. Raises: ValueError: in case invalid params\",\n \"emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW:\",\n \"Since model is a dict we can insert multiple modular networks in this\",\n \"learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers:\",\n \"checkpoint (if one exists on the path).\\\"\\\"\\\" # Instantiate checkpoint and restore from\",\n \"if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels']\",\n \"# We don't want to see the future. steps = np.arange(step - (num_steps\",\n \"- 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list\",\n \"lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE\",\n \"from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries\",\n \"dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float /\",\n \"= get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0),\",\n \"boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step,\",\n \"dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if\",\n \"batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate during warm\",\n \"= softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return\",\n \"get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number\",\n \"based on config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate)\",\n \"return config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path =\",\n \"optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a one epoch iterator.\\\"\\\"\\\" keep_labels =\",\n \"dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info(\",\n \"object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries =\",\n \"that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file)\",\n \"training = kwargs['training'] if training: # Set learning_phase to True to use models\",\n \"np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False,\",\n \"MomentumOptimizer .') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and global step.\\\"\\\"\\\"\",\n \"import tensorflow as tf import numpy as np import matplotlib.pyplot as plt #\",\n \"len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps\",\n \"script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path\",\n \"__future__ import absolute_import from __future__ import division from __future__ import print_function from config\",\n \"lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries)\",\n \"* [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label =\",\n \"else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = []\",\n \"in Eager we need to call assign to update the learning rate. Args:\",\n \"print(n) embs = [] labels = [] steps = [] seq_lens = []\",\n \"tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) #\",\n \"= data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len *\",\n \"= seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data:\",\n \"a dict we can insert multiple modular networks in this dict. checkpoint =\",\n \"'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config\",\n \"return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't\",\n \"= labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens\",\n \"config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while\",\n \"the learning rate decay function. Returns: lr_fn: function, this can be called to\",\n \"open(config_path) as f: config = json.load(f) assert config is not None, \\\"config file\",\n \"ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status,\",\n \"names = [] seq_labels = [] if keep_data: frames = [] if optical_flow:\",\n \"np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs),\",\n \"config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir)\",\n \"data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0]\",\n \"a data dict which contains all the requested sequences.\\\"\\\"\\\" data = iterator.get_next() return\",\n \"from __future__ import print_function from config import CONFIG import json import tensorflow as\",\n \"as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and\",\n \"p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence)\",\n \"int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try:\",\n \"warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr =\",\n \"more than one batch to be passed and we don't want to automatically\",\n \"checkpoint. # Since model is a dict we can insert multiple modular networks\",\n \"-1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if\",\n \"GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers:\",\n \"tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory\",\n \"tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup * warmup_lr return\",\n \"single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator,\",\n \"try: print(n) embs = [] labels = [] steps = [] seq_lens =\",\n \"+ num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx,\",\n \"# more than one batch to be passed and we don't want to\",\n \"Memory growth must be set before GPUs have been initialized logging.info(e) os.environ[\\\"CUDA_VISIBLE_DEVICES\\\"] =\",\n \"based on the provided config. Raises: ValueError: in case invalid params have been\",\n \"axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image\",\n \"emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch))\",\n \"= os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting\",\n \"= emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list =\",\n \"states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful\",\n \"--force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir)\",\n \"reset(self): self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase before calling\",\n \"emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0]\",\n \"for measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time\",\n \"manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index =\",\n \"num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats,\",\n \"+ is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow object_detection code.\",\n \"rate based on the provided config. Raises: ValueError: in case invalid params have\",\n \"sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats\",\n \"frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy',\",\n \"= data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels,\",\n \"= tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] =\",\n \"format='png') buf.seek(0) # Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4)\",\n \"lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0,\",\n \"frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels)\",\n \"CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else:\",\n \"num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames\",\n \"training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack(\",\n \"provided or is corrupted\\\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS =\",\n \"tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) /\",\n \"seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if\",\n \"\\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def\",\n \"return time.time() - self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self):\",\n \"'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats\",\n \"= optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return\",\n \"k, v in data.items(): # Need to do this as some modalities might\",\n \"in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v))\",\n \"= [] names_list = [] seq_labels_list = [] if keep_data: frames_list = []\",\n \"= time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase before calling function f.\\\"\\\"\\\"\",\n \"rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where(\",\n \"cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats\",\n \"is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow object_detection code. def\",\n \"get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def\",\n \"= tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step\",\n \"version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles\",\n \"NOTE: This returns a function as in Eager we need to call assign\",\n \"if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape =\",\n \"config = json.load(f) assert config is not None, \\\"config file is not provided\",\n \"break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data dict which contains all the requested\",\n \"CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not supported.\",\n \"def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind >\",\n \"[] if optical_flow: frame_original = [] # Reset GRU states for each video.\",\n \"nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance\",\n \"num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d',\",\n \"num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx + num_steps) #\",\n \"the config. \\\"\\\"\\\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay':\",\n \"'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only\",\n \"== 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr,\",\n \"t) dist = e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): #\",\n \"get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see\",\n \"if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers for other\",\n \"Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn =\",\n \"= CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps = np.arange(step\",\n \"num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps\",\n \"dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range (0,\",\n \"return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and global step.\\\"\\\"\\\" optimizer =\",\n \"dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status =\",\n \"def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list],\",\n \"supported. Only support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.') return\",\n \"keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n\",\n \"= {} for k, v in data.items(): # Need to do this as\",\n \"def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in embedding\",\n \"GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as e: # Memory growth must be\",\n \"the future. steps = np.arange(step - (num_steps - 1) * stride, step +\",\n \"might not exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes =\",\n \"tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the\",\n \"in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir\",\n \"num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn'\",\n \"split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take float64\",\n \"- 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1,\",\n \"v in data.items(): # Need to do this as some modalities might not\",\n \"np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for\",\n \"sim_matrix to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' %\",\n \"'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow:\",\n \"embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names)\",\n \"learning phase.\\\"\\\"\\\" if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator\",\n \"[] seq_lens = [] names = [] seq_labels = [] if keep_data: frames\",\n \"be passed and we don't want to automatically # reset hidden states after\",\n \"= [] steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list =\",\n \"# Make Recurrent Layers stateful, set batch size. # We do this as\",\n \"[] seq_labels_list = [] if keep_data: frames_list = [] if optical_flow: frame_original_list =\",\n \"Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG\",\n \"[num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 *\",\n \"lr = (1.0 - is_warmup) * lr + is_warmup * warmup_lr return lr\",\n \"modalities might not exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes\",\n \"end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not supported. Only\",\n \"lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on config and learning rate.\\\"\\\"\\\" if\",\n \"This returns a function as in Eager we need to call assign to\",\n \"not exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx,\",\n \"optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs)\",\n \"import division from __future__ import print_function from config import CONFIG import json import\",\n \"model is a dict we can insert multiple modular networks in this dict.\",\n \"raise ValueError('Learning rate decay type %s not supported. Only support' 'the following decay\",\n \"cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On\",\n \"frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset =\",\n \"optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs)\",\n \"pre-existing checkpoint. # Since model is a dict we can insert multiple modular\",\n \"Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with\",\n \"are embedding the whole sequence and that can take # more than one\",\n \"(lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based\",\n \"rate based on config. NOTE: This returns a function as in Eager we\",\n \"boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max(\",\n \"steps = [] seq_lens = [] names = [] seq_labels = [] if\",\n \"img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for\",\n \"for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step,\",\n \"image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf,\",\n \"f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name =\",\n \"gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len]\",\n \"# pylint: disable=g-import-not-at-top import io import math import os import time from absl\",\n \"def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def\",\n \"= optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model\",\n \"checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config] elif\",\n \"i in range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps =\",\n \"GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\",\",\n \"n, axis=-1): # Generate m random permuations of range (0, n) # NumPy\",\n \"tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles,\",\n \"range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps =\",\n \"num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries\",\n \"if len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len)\",\n \"cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps,\",\n \"code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries)\",\n \"= json.load(f) assert config is not None, \\\"config file is not provided or\",\n \"os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a\",\n \"math import os import time from absl import flags from absl import logging\",\n \"np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels,\",\n \"'r') as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break\",\n \"sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w)\",\n \"x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p\",\n \"== len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels:\",\n \"time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval):\",\n \"seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow:\",\n \"return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step,\",\n \"data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i\",\n \"'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory\",\n \"restore from any pre-existing checkpoint. # Since model is a dict we can\",\n \"import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a\",\n \"get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a one\",\n \"= global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int <\",\n \"names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished\",\n \"in embedding space.\\\"\\\"\\\" # Set learning_phase to False to use models in inference\",\n \"sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def\",\n \"if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original,\",\n \"= get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step\",\n \"= os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to\",\n \"(1.0 - is_warmup) * lr + is_warmup * warmup_lr return lr # Minimally\",\n \"for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info(\",\n \"phase before calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f after\",\n \"config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train:\",\n \"batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours\",\n \"training argument.') training = kwargs['training'] if training: # Set learning_phase to True to\",\n \"case invalid params have been passed in the config. \\\"\\\"\\\" lr_params = optimizer_config.LR\",\n \"is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data dict which\",\n \"Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the\",\n \"< max_embs # Make Recurrent Layers stateful, set batch size. # We do\",\n \"* num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded\",\n \"already ' 'has train_logs. Please provide a new logdir name in ' 'config\",\n \"lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for\",\n \"global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False,\",\n \"ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore\",\n \"logging.info( 'Using config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r')\",\n \"curr_idx = i * frames_per_batch curr_data = {} for k, v in data.items():\",\n \"= frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for k, v\",\n \"function. Returns: lr_fn: function, this can be called to return the current learning\",\n \"j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats),\",\n \"# Need to do this as some modalities might not exist. if len(v.shape)\",\n \"CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might\",\n \"up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant(\",\n \"tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if\",\n \"\\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if num_steps is None: if training: num_steps =\",\n \"numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import\",\n \"'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len =\",\n \"True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as\",\n \"tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR\",\n \"return random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate during warm up phase.\\\"\\\"\\\"\",\n \"CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps = np.arange(step -\",\n \"or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for\",\n \"tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0\",\n \"file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config =\",\n \"gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical\",\n \"random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns\",\n \"labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n +=\",\n \"alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params):\",\n \"\\\"\\\"\\\"Create and restore checkpoint (if one exists on the path).\\\"\\\"\\\" # Instantiate checkpoint\",\n \"(0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles,\",\n \"flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames'] frames_list\",\n \"num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\" # Set learning_phase to\",\n \"exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k,\",\n \"np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math import\",\n \"'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step -\",\n \"to_dict(v)) for k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True):\",\n \"get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats\",\n \"layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers:\",\n \"= CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def\",\n \"not provided or is corrupted\\\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS\",\n \"epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn']\",\n \"kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does not have training\",\n \"memory growth needs to be the same across GPUs for gpu in GPUS:\",\n \"sequences.\\\"\\\"\\\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training,\",\n \"learning_rate): \\\"\\\"\\\"Returns optimizer based on config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer':\",\n \"max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config,\",\n \"multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint,\",\n \"in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\"\",\n \"embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens\",\n \"set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase before calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs):\",\n \"batch to be passed and we don't want to automatically # reset hidden\",\n \"labels = [] steps = [] seq_lens = [] names = [] seq_labels\",\n \"the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate,\",\n \"= frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) ==\",\n \"num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx\",\n \"= flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames']\",\n \"os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config as no config.json\",\n \"name in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir,\",\n \"step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx,\",\n \"optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer,\",\n \"get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS >\",\n \"f(*args, **kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path)\",\n \"for k, v in data.items(): # Need to do this as some modalities\",\n \"the provided config. Raises: ValueError: in case invalid params have been passed in\",\n \"as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3),\",\n \"plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer\",\n \"= np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None,\",\n \"tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats\",\n \"else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return\",\n \"= v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats,\",\n \"emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs\",\n \"f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step,\",\n \"training: # Set learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1)\",\n \"status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config]\",\n \"= seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i\",\n \"all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len\",\n \"Raises: ValueError: in case invalid params have been passed in the config. \\\"\\\"\\\"\",\n \"CONFIG import json import tensorflow as tf import numpy as np import matplotlib.pyplot\",\n \"def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to\",\n \"Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args,\",\n \"config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not\",\n \"batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 *\",\n \"learning rate decay function. Returns: lr_fn: function, this can be called to return\",\n \"keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the\",\n \"one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn =\",\n \"this as we are embedding the whole sequence and that can take #\",\n \"keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint\",\n \"Layers stateful, set batch size. # We do this as we are embedding\",\n \"models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use\",\n \"import os import time from absl import flags from absl import logging from\",\n \"if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth\",\n \"True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to\",\n \"absolute_import from __future__ import division from __future__ import print_function from config import CONFIG\",\n \"seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i *\",\n \"stateful, set batch size. # We do this as we are embedding the\",\n \"= tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise\",\n \"the batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \\\"\\\"\\\"Simple timer\",\n \"return self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the\",\n \"current learning rate based on config. NOTE: This returns a function as in\",\n \"= json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return\",\n \"= CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps\",\n \"exists on the path).\\\"\\\"\\\" # Instantiate checkpoint and restore from any pre-existing checkpoint.\",\n \"tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps':\",\n \"= checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir,\",\n \"= [1, ] while cond(n): try: print(n) embs = [] labels = []\",\n \"len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if\",\n \"staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f\",\n \"logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as e: # Memory growth\",\n \"buffer and return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1)\",\n \"tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config\",\n \"num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy())\",\n \"= int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU')\",\n \"np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist =\",\n \"# Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add\",\n \"raise ValueError('Function called with set_learning_phase decorator which' ' does not have training argument.')\",\n \"[to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k,\",\n \"for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0]\",\n \"= model['cnn'] emb = model['emb'] embs_list = [] labels_list = [] steps_list =\",\n \"if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator which' '\",\n \"learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs)\",\n \"'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'),\",\n \"rate, optimizer and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate\",\n \"function, this can be called to return the current learning rate based on\",\n \"tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x)\",\n \"can take # more than one batch to be passed and we don't\",\n \"keep_data: frames_list = [] if optical_flow: frame_original_list = [] n = 0 def\",\n \"'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list\",\n \"np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]):\",\n \"lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float\",\n \"data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0)\",\n \"float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0):\",\n \"# np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy())\",\n \"* num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps *\",\n \"ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return\",\n \"= tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr +\",\n \"exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict)\",\n \"= np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels\",\n \"except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list,\",\n \"config = None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert\",\n \"axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list,\",\n \"return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\"\",\n \"rate during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32)\",\n \"tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup\",\n \"np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i]\",\n \"checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c) for\",\n \"= tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list =\",\n \"* num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats\",\n \"requested sequences.\\\"\\\"\\\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data,\",\n \"num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else:\",\n \"while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is\",\n \"def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save to buffer and return TB compatible\",\n \"to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict):\",\n \"import flags from absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg')\",\n \"\\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self): return time.time()\",\n \"invalid params have been passed in the config. \\\"\\\"\\\" lr_params = optimizer_config.LR #\",\n \"f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) +\",\n \"passed in config as no config.json file exists in ' '%s', logdir) with\",\n \"# Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn\",\n \"seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if\",\n \"fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params))\",\n \"that can take # more than one batch to be passed and we\",\n \"to call assign to update the learning rate. Args: optimizer_config: EasyDict, contains params\",\n \"visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\"\",\n \"def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE ==\",\n \"labels_list = [] steps_list = [] seq_lens_list = [] names_list = [] seq_labels_list\",\n \"steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps])\",\n \"num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride =\",\n \"self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time()\",\n \"f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels =\",\n \"* mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps),\",\n \"get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i\",\n \"overwriting a directory that already ' 'has train_logs. Please provide a new logdir\",\n \"optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): #\",\n \"optimizer based on config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt =\",\n \"cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list = [] steps_list\",\n \"if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness\",\n \"if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data\",\n \"num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps)\",\n \"for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data,\",\n \"max_embs # Make Recurrent Layers stateful, set batch size. # We do this\",\n \"from any pre-existing checkpoint. # Since model is a dict we can insert\",\n \"= tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done\",\n \"learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since\",\n \"tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else:\",\n \"* [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0)\",\n \"dist = e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate\",\n \"global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on config and learning\",\n \"from absl import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS =\",\n \"mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config = None if\",\n \"axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0)\",\n \"= sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e =\",\n \"tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type\",\n \"future. steps = np.arange(step - (num_steps - 1) * stride, step + stride,\",\n \"# Add the batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object):\",\n \"across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical\",\n \"frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers for\",\n \"global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps,\",\n \"None, \\\"config file is not provided or is corrupted\\\" return config def prepare_gpu(ind=-1):\",\n \"import CONFIG import json import tensorflow as tf import numpy as np import\",\n \"range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps = seq_len -\",\n \"one batch to be passed and we don't want to automatically # reset\",\n \"@tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if num_steps\",\n \"* [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps =\",\n \"on config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif\",\n \"CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list = []\",\n \"optimizer and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations learning_rate =\",\n \"random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate during warm up phase.\\\"\\\"\\\" if\",\n \"config is not None, \\\"config file is not provided or is corrupted\\\" return\",\n \"len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates\",\n \"needs to be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True)\",\n \"tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current learning rate\",\n \"ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if one exists\",\n \"axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames =\",\n \"frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset\",\n \"np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens,\",\n \"take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w,\",\n \"based on config. NOTE: This returns a function as in Eager we need\",\n \"checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(\",\n \"num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats =\",\n \"import logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def\",\n \"assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs)\",\n \"logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir,\",\n \"\\\"\\\"\\\"Sets the correct learning phase before calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls\",\n \"time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict\",\n \"batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps,\",\n \"axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size,\",\n \"%s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir =\",\n \"automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for\",\n \"False return dataset def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save to buffer and\",\n \"= tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im)\",\n \"return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if\",\n \"checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects):\",\n \"emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames =\",\n \"config import CONFIG import json import tensorflow as tf import numpy as np\",\n \"Set learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: #\",\n \"supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step():\",\n \"ValueError('You might be overwriting a directory that already ' 'has train_logs. Please provide\",\n \"prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1:\",\n \"lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in\",\n \"+ boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0]\",\n \"import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math import os\",\n \"* frames_per_batch curr_data = {} for k, v in data.items(): # Need to\",\n \"random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)),\",\n \"[] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for\",\n \"batch size. # We do this as we are embedding the whole sequence\",\n \"tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles\",\n \"'has train_logs. Please provide a new logdir name in ' 'config or pass\",\n \"= range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps\",\n \"sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r')\",\n \"os import time from absl import flags from absl import logging from easydict\",\n \"num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\" # Set learning_phase to False\",\n \"growth must be set before GPUs have been initialized logging.info(e) os.environ[\\\"CUDA_VISIBLE_DEVICES\\\"] = str(ind)\",\n \"GPUs\\\"]) except RuntimeError as e: # Memory growth must be set before GPUs\",\n \"don't want to automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE\",\n \"from __future__ import division from __future__ import print_function from config import CONFIG import\",\n \"power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not supported. Only support'\",\n \"tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats =\",\n \"im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for\",\n \"'GPU') try: # Currently, memory growth needs to be the same across GPUs\",\n \"emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx\",\n \"permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis),\",\n \"'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y):\",\n \"single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get\",\n \"< warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup *\",\n \"ValueError('Learning rate decay type %s not supported. Only support' 'the following decay types:\",\n \"return f(*args, **kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with\",\n \"a one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn\",\n \"emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save\",\n \"lr_fn: function, this can be called to return the current learning rate based\",\n \"data, training, num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if num_steps is None: if\",\n \"(num_steps - 1) * stride, step + stride, stride) return steps def get_indices(curr_idx,\",\n \"def cond(n): if max_embs is None: return True else: return n < max_embs\",\n \"tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]]\",\n \"before calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f after setting\",\n \"return True else: return n < max_embs # Make Recurrent Layers stateful, set\",\n \"type %s not supported. Only support' 'the following decay types: fixed, exp_decay, manual,'\",\n \"\\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries =\",\n \"training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using\",\n \"= [] if keep_data: frames_list = [] if optical_flow: frame_original_list = [] n\",\n \"= tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else\",\n \"\\\"\\\"\\\"Return a data dict which contains all the requested sequences.\\\"\\\"\\\" data = iterator.get_next()\",\n \"single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False,\",\n \"learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer':\",\n \"config. NOTE: This returns a function as in Eager we need to call\",\n \"in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in\",\n \"'Using config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as\",\n \"steps = np.arange(step - (num_steps - 1) * stride, step + stride, stride)\",\n \"tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup =\",\n \"return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data through\",\n \"embedding space.\\\"\\\"\\\" # Set learning_phase to False to use models in inference mode.\",\n \"provided config. Raises: ValueError: in case invalid params have been passed in the\",\n \"[name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels\",\n \"cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want\",\n \"* num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats\",\n \"data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if\",\n \"wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f: config\",\n \"num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) * frames_per_batch\",\n \"= iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes\",\n \"set_learning_phase decorator which' ' does not have training argument.') training = kwargs['training'] if\",\n \"= model['emb'] embs_list = [] labels_list = [] steps_list = [] seq_lens_list =\",\n \"evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for\",\n \"data, global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\" #\",\n \"plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0)\",\n \"np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0)\",\n \"take # more than one batch to be passed and we don't want\",\n \"range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def\",\n \"= random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning\",\n \"import numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io\",\n \"tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in\",\n \"this as some modalities might not exist. if len(v.shape) > 1 and v.shape[1]\",\n \"step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e /\",\n \"= np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names =\",\n \"1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to\",\n \"__init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed()\",\n \"'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does\",\n \"for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \\\"\\\"\\\"Create\",\n \"learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict\",\n \"frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if\",\n \"tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done =\",\n \"= tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames']\",\n \"warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) *\",\n \"+ num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def\",\n \"1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed':\",\n \"= emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx +\",\n \"support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning\",\n \"1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance)\",\n \"[] names = [] seq_labels = [] if keep_data: frames = [] if\",\n \"[] if keep_data: frames = [] if optical_flow: frame_original = [] # Reset\",\n \"frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs)\",\n \"to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split,\",\n \"have training argument.') training = kwargs['training'] if training: # Set learning_phase to True\",\n \"= [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i\",\n \"config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing\",\n \"import math import os import time from absl import flags from absl import\",\n \"return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay\",\n \"= steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs\",\n \"else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats\",\n \"from absl import flags from absl import logging from easydict import EasyDict import\",\n \"names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list\",\n \"return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current learning\",\n \"global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif\",\n \"] while cond(n): try: print(n) embs = [] labels = [] steps =\",\n \"get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1)\",\n \"{'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames']\",\n \"'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir)\",\n \"emb = model['emb'] embs_list = [] labels_list = [] steps_list = [] seq_lens_list\",\n \"e = np.exp(np.array(w) / t) dist = e / np.sum(e) return dist def\",\n \"statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for\",\n \"= False return dataset def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save to buffer\",\n \"to initialize the learning rate and the learning rate decay function. Returns: lr_fn:\",\n \"num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames\",\n \"ValueError: in case invalid params have been passed in the config. \\\"\\\"\\\" lr_params\",\n \"num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 *\",\n \"overwrite: logging.info( 'Using the existing passed in config as no config.json file exists\",\n \"callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name\",\n \"the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU')\",\n \"space.\\\"\\\"\\\" # Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0)\",\n \"[] names_list = [] seq_labels_list = [] if keep_data: frames_list = [] if\",\n \"to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image\",\n \"= tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently,\",\n \"data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0]\",\n \"return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps\",\n \"= 0 def cond(n): if max_embs is None: return True else: return n\",\n \"= np.exp(np.array(w) / t) dist = e / np.sum(e) return dist def random_choice_noreplace(m,\",\n \"RuntimeError as e: # Memory growth must be set before GPUs have been\",\n \"wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f after setting proper learning phase.\\\"\\\"\\\" if 'training'\",\n \"= tf.expand_dims(image, 0) return image class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\"\",\n \"try: # Currently, memory growth needs to be the same across GPUs for\",\n \"__future__ import print_function from config import CONFIG import json import tensorflow as tf\",\n \"rate decay function. Returns: lr_fn: function, this can be called to return the\",\n \"= [] names = [] seq_labels = [] if keep_data: frames = []\",\n \"if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt =\",\n \"while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir,\",\n \"* stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps\",\n \"= len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return\",\n \"/ warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32)\",\n \"True else: return n < max_embs # Make Recurrent Layers stateful, set batch\",\n \"axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames']\",\n \"learning rate, optimizer and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step = optimizer.iterations\",\n \"= keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb']\",\n \"0 def cond(n): if max_embs is None: return True else: return n <\",\n \"len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels)\",\n \"optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is\",\n \"tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path =\",\n \"logging from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data,\",\n \"cycle=False) else: raise ValueError('Learning rate decay type %s not supported. Only support' 'the\",\n \"**ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if one exists on the path).\\\"\\\"\\\" # Instantiate\",\n \"lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 -\",\n \"= np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels =\",\n \"growth needs to be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu,\",\n \"batch.\\\"\\\"\\\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2)\",\n \"plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert\",\n \"as we are embedding the whole sequence and that can take # more\",\n \"= data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn'\",\n \"stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps =\",\n \"'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist.\",\n \"self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase before calling function\",\n \"float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix,\",\n \"= True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs = []\",\n \"to automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru':\",\n \"lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on\",\n \"num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the\",\n \"except RuntimeError as e: # Memory growth must be set before GPUs have\",\n \"checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if one exists on the\",\n \"after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful =\",\n \"np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of\",\n \"logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list,\",\n \"data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries,\",\n \"pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\"\",\n \"for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n):\",\n \"for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for\",\n \"axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step)\",\n \"num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current\",\n \"dataset def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save to buffer and return TB\",\n \"use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path):\",\n \"frame_original_list = [] n = 0 def cond(n): if max_embs is None: return\",\n \"correct learning phase before calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function\",\n \"np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names,\",\n \"os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep\",\n \"curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in\",\n \"= tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the\",\n \"optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to\",\n \"[] labels_list = [] steps_list = [] seq_lens_list = [] names_list = []\",\n \"manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR\",\n \"learning rate based on config. NOTE: This returns a function as in Eager\",\n \"for k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups\",\n \"max_embs is None: return True else: return n < max_embs # Make Recurrent\",\n \"provide a new logdir name in ' 'config or pass --force_train while launching\",\n \"= [] seq_lens_list = [] names_list = [] seq_labels_list = [] if keep_data:\",\n \"[0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that\",\n \"steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except\",\n \"config. \\\"\\\"\\\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def\",\n \"cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps,\",\n \"raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer, MomentumOptimizer .')\",\n \"'the following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step:\",\n \"= emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2],\",\n \"for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting\",\n \"buf.seek(0) # Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) #\",\n \"models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config\",\n \"\\\"config file is not provided or is corrupted\\\" return config def prepare_gpu(ind=-1): ind\",\n \"passed and we don't want to automatically # reset hidden states after each\",\n \"+ stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx\",\n \"vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0),\",\n \"== 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False)\",\n \"in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats =\",\n \"cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS\",\n \"if training: # Set learning_phase to True to use models in training mode.\",\n \"warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int\",\n \"single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True,\",\n \"{} for k, v in data.items(): # Need to do this as some\",\n \"cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2\",\n \"config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE\",\n \"global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float\",\n \"+ 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE ==\",\n \"\\\"\\\"\\\"Create a pyplot, save to buffer and return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x,\",\n \"ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be\",\n \"for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config\",\n \"np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original\",\n \"= tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return\",\n \"'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence =\",\n \"def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step)\",\n \"learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn']\",\n \"len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names)\",\n \"np.exp(np.array(w) / t) dist = e / np.sum(e) return dist def random_choice_noreplace(m, n,\",\n \"self.elapsed() >= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct\",\n \"else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32)\",\n \"with set_learning_phase decorator which' ' does not have training argument.') training = kwargs['training']\",\n \"and return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout()\",\n \"summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix\",\n \"secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file)\",\n \"if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert\",\n \"[(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return\",\n \"cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats(\",\n \"function as in Eager we need to call assign to update the learning\",\n \"lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)()\",\n \"config. Raises: ValueError: in case invalid params have been passed in the config.\",\n \"'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep '\",\n \"labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0)\",\n \"= {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data:\",\n \"y): \\\"\\\"\\\"Create a pyplot, save to buffer and return TB compatible image.\\\"\\\"\\\" plt.figure()\",\n \"/ t) dist = e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1):\",\n \"keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list\",\n \"video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len\",\n \"curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps\",\n \"lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step):\",\n \"keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb =\",\n \"get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for\",\n \"in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1)\",\n \"isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k,\",\n \"config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else:\",\n \"1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps),\",\n \"can be called to return the current learning rate based on the provided\",\n \"def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time = time.time() def\",\n \"that already ' 'has train_logs. Please provide a new logdir name in '\",\n \"tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to\",\n \"step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0,\",\n \"data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label])\",\n \"curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data,\",\n \"i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data\",\n \"= tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint\",\n \"in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i],\",\n \"frame_original = [] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE ==\",\n \"to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb']\",\n \"+ 1) * frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch\",\n \"' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file:\",\n \"as plt # pylint: disable=g-import-not-at-top import io import math import os import time\",\n \"* warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup)\",\n \"current learning rate based on the provided config. Raises: ValueError: in case invalid\",\n \"training, num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if num_steps is None: if training:\",\n \"might be overwriting a directory that already ' 'has train_logs. Please provide a\",\n \"emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps,\",\n \"= frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list #\",\n \"f: config = json.load(f) assert config is not None, \\\"config file is not\",\n \"seq_lens = [] names = [] seq_labels = [] if keep_data: frames =\",\n \"= frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent layers\",\n \"k, v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory\",\n \"[] seq_labels = [] if keep_data: frames = [] if optical_flow: frame_original =\",\n \"We do this as we are embedding the whole sequence and that can\",\n \"tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list,\",\n \"and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes,\",\n \"= tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step -\",\n \"= [] if optical_flow: frame_original = [] # Reset GRU states for each\",\n \"as config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file,\",\n \"assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels)\",\n \"neighbours in embedding space.\\\"\\\"\\\" # Set learning_phase to False to use models in\",\n \"def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a\",\n \"pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step,\",\n \"= tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as e: #\",\n \"is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES\",\n \"Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix'\",\n \"i * frames_per_batch curr_data = {} for k, v in data.items(): # Need\",\n \"keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names =\",\n \"in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to\",\n \"tf import numpy as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import\",\n \"networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10,\",\n \"create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we can insert multiple\",\n \"1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list =\",\n \"be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus =\",\n \"modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir,\",\n \"emb(cnn_feats, 2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1)\",\n \":cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate during warm up\",\n \"num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps =\",\n \"- 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps,\",\n \"to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase\",\n \"io import math import os import time from absl import flags from absl\",\n \"else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data = {} for\",\n \"im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix\",\n \"0) return image class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\" def __init__(self):\",\n \"seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original =\",\n \"in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches =\",\n \"'%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k,\",\n \"axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step]\",\n \"gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save to buffer and return TB compatible image.\\\"\\\"\\\"\",\n \"seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels =\",\n \"steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs =\",\n \"Currently, memory growth needs to be the same across GPUs for gpu in\",\n \"json.load(f) assert config is not None, \\\"config file is not provided or is\",\n \"frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if\",\n \"[seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0]\",\n \"# Set learning_phase to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return\",\n \"seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels:\",\n \"as no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as\",\n \"tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image, 0) return image\",\n \"the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list,\",\n \"in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step -\",\n \"GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except\",\n \"= emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if\",\n \"opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else:\",\n \"v in config.items()]) else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for\",\n \"or is corrupted\\\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU')\",\n \"= tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img\",\n \"lr_params): \\\"\\\"\\\"Returns learning rate during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int\",\n \"phase.\\\"\\\"\\\" if 'training' not in kwargs: raise ValueError('Function called with set_learning_phase decorator which'\",\n \"False to use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb =\",\n \"lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def\",\n \"return image class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset()\",\n \"CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps, seq_len = get_data(iterator)\",\n \"return lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates):\",\n \"the learning rate. Args: optimizer_config: EasyDict, contains params required to initialize the learning\",\n \"tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float =\",\n \"keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0)\",\n \"not supported. Only support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.')\",\n \"gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs =\",\n \"np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0])\",\n \"initialize the learning rate and the learning rate decay function. Returns: lr_fn: function,\",\n \"logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as e:\",\n \"% split, summary_im, step=global_step) # Convert sim_matrix to float32 as summary_image doesn't take\",\n \"# np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len)\",\n \"tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the following'\",\n \"axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames\",\n \"* num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides\",\n \"= frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps)\",\n \"'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate):\",\n \"number %d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy())\",\n \"set batch size. # We do this as we are embedding the whole\",\n \"axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace(\",\n \"json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a\",\n \"get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if num_steps is None:\",\n \"frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) ==\",\n \"to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer\",\n \"json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You\",\n \"def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return\",\n \"absl import flags from absl import logging from easydict import EasyDict import matplotlib\",\n \"rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates *\",\n \"+= 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list,\",\n \"evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False\",\n \"restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore\",\n \"to be passed and we don't want to automatically # reset hidden states\",\n \"measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def\",\n \"cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr,\",\n \"force_train: raise ValueError('You might be overwriting a directory that already ' 'has train_logs.\",\n \"[num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1,\",\n \"'train.logs') if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a directory\",\n \"seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1\",\n \"im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in\",\n \"= [] seq_lens = [] names = [] seq_labels = [] if keep_data:\",\n \">= target_interval def reset(self): self.time = time.time() def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning\",\n \"else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data dict which contains all the\",\n \"= np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels\",\n \"for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v\",\n \"single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps =\",\n \"config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def\",\n \"frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels:\",\n \"* num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats,\",\n \"else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO:\",\n \"= labels_list # Reset statefulness to recurrent layers for other evaluation tasks. if\",\n \"logging.debug('On sequence number %d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy',\",\n \"cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We\",\n \"all the requested sequences.\\\"\\\"\\\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def\",\n \"base CNN.\\\"\\\"\\\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS\",\n \"image = tf.expand_dims(image, 0) return image class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed\",\n \"axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats =\",\n \"params required to initialize the learning rate and the learning rate decay function.\",\n \"Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self): return\",\n \"range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2\",\n \"training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps\",\n \"= tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done\",\n \"while cond(n): try: print(n) embs = [] labels = [] steps = []\",\n \"same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS),\",\n \"ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create\",\n \"size. # We do this as we are embedding the whole sequence and\",\n \"keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] =\",\n \"%d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) #\",\n \"v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json\",\n \"single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from\",\n \"emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step,\",\n \"num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step -\",\n \"True gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs = [] labels\",\n \"from a one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS\",\n \"matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math import os import\",\n \"that provides current learning rate based on config. NOTE: This returns a function\",\n \"keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list):\",\n \"if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config as\",\n \"1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens':\",\n \"# Currently, memory growth needs to be the same across GPUs for gpu\",\n \"is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr\",\n \"\\\"\\\"\\\"Calls the function f after setting proper learning phase.\\\"\\\"\\\" if 'training' not in\",\n \"data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps) emb_feats =\",\n \"Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in\",\n \"not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep ' ' %s\",\n \"range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE\",\n \"phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS,\",\n \"warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr = lr_params.INITIAL_LR *\",\n \"is not None, \\\"config file is not provided or is corrupted\\\" return config\",\n \"if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE,\",\n \"return single_steps def get_embeddings_dataset(model, iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings\",\n \"recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in\",\n \"# Instantiate checkpoint and restore from any pre-existing checkpoint. # Since model is\",\n \"None: return True else: return n < max_embs # Make Recurrent Layers stateful,\",\n \"dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO:\",\n \"global_step, lr_params): \\\"\\\"\\\"Returns learning rate during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0:\",\n \"learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr,\",\n \"def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise\",\n \"return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f:\",\n \"gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while cond(n): try:\",\n \"= tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32) warmup_percent_done = global_steps_float / warmup_steps_float warmup_lr\",\n \"tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0 - is_warmup) * lr + is_warmup\",\n \"from __future__ import absolute_import from __future__ import division from __future__ import print_function from\",\n \"GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: #\",\n \"tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep ' ' %s for\",\n \"embeddings from a one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step =\",\n \"* warmup_lr return lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step,\",\n \"for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels\",\n \"pylint: disable=g-import-not-at-top import io import math import os import time from absl import\",\n \"image class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def\",\n \"data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\"\",\n \"config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path, 'r') as config_file:\",\n \"def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e / np.sum(e)\",\n \"through base CNN.\\\"\\\"\\\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES *\",\n \"[] if keep_data: frames_list = [] if optical_flow: frame_original_list = [] n =\",\n \"lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES]\",\n \"n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size,\",\n \"seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn,\",\n \"tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return\",\n \"each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.reset_states() data, chosen_steps,\",\n \"from config import CONFIG import json import tensorflow as tf import numpy as\",\n \"= [] labels = [] steps = [] seq_lens = [] names =\",\n \"if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert\",\n \"steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps =\",\n \"= [] if optical_flow: frame_original_list = [] n = 0 def cond(n): if\",\n \"to be the same across GPUs for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus\",\n \"- 1) * stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps,\",\n \"= checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if isinstance(config, list): return [to_dict(c)\",\n \"embedding the dataset.') break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names':\",\n \"setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if\",\n \"# Generate m random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis)\",\n \"time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data dict which contains all\",\n \"CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats\",\n \"batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1)\",\n \"axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original =\",\n \"buf = io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image\",\n \"num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps =\",\n \"= tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j\",\n \"'Using the existing passed in config as no config.json file exists in '\",\n \"to update the learning rate. Args: optimizer_config: EasyDict, contains params required to initialize\",\n \"np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels =\",\n \"global_step def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we can\",\n \"argument.') training = kwargs['training'] if training: # Set learning_phase to True to use\",\n \"setting proper learning phase.\\\"\\\"\\\" if 'training' not in kwargs: raise ValueError('Function called with\",\n \"0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k]\",\n \"TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf =\",\n \"return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return\",\n \"checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return\",\n \"# Convert sim_matrix to float32 as summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32)\",\n \"Returns: lr_fn: function, this can be called to return the current learning rate\",\n \"embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps)\",\n \"if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)]\",\n \"logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k, v\",\n \"0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int,\",\n \"cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles\",\n \"warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int, tf.float32)\",\n \"mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats\",\n \"for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps =\",\n \"and the learning rate decay function. Returns: lr_fn: function, this can be called\",\n \"global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32)\",\n \"warmup_steps_float warmup_lr = lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr\",\n \"== 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate,\",\n \"insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager(\",\n \"learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE ==\",\n \"EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes\",\n \"if keep_data: frames_list = [] if optical_flow: frame_original_list = [] n = 0\",\n \"embedding the whole sequence and that can take # more than one batch\",\n \"def __init__(self): self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval): return\",\n \"def to_dict(config): if isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config,\",\n \"[] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean(\",\n \"CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames'])\",\n \"have been passed in the config. \\\"\\\"\\\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda\",\n \"CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1,\",\n \"return dict([(k, to_dict(v)) for k, v in config.items()]) else: return config def setup_train_dir(logdir,\",\n \"visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames'] frames_list = tf.unstack(frames,\",\n \"following' 'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer\",\n \"training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models in\",\n \"which contains all the requested sequences.\\\"\\\"\\\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens']\",\n \"= os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed\",\n \"if keep_data: frames = [] if optical_flow: frame_original = [] # Reset GRU\",\n \"iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb\",\n \"load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f)\",\n \"doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step) def\",\n \"= cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE #\",\n \"if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps\",\n \"tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data,\",\n \"curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps,\",\n \"= kwargs['training'] if training: # Set learning_phase to True to use models in\",\n \"use models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if\",\n \"get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR) global_step\",\n \"len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps)\",\n \"gru_layer.input_spec[0].shape = [1, ] while cond(n): try: print(n) embs = [] labels =\",\n \"not have training argument.') training = kwargs['training'] if training: # Set learning_phase to\",\n \"-1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs to be the same\",\n \"Eager we need to call assign to update the learning rate. Args: optimizer_config:\",\n \"rate and the learning rate decay function. Returns: lr_fn: function, this can be\",\n \"directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir, 'config.json') if not os.path.exists(config_path) or overwrite:\",\n \"if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break\",\n \"called to return the current learning rate based on the provided config. Raises:\",\n \"+ num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'),\",\n \"plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image image = tf.image.decode_png(buf.getvalue(),\",\n \"axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t) dist = e\",\n \"# Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer\",\n \"config to exist. Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds)\",\n \"logging.info('Waiting for config to exist. Going to sleep ' ' %s for secs.',\",\n \"steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps\",\n \"the correct learning phase before calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the\",\n \"- 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps,\",\n \"axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img)\",\n \"for i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list\",\n \"elif lr_params.DECAY_TYPE == 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f =\",\n \"with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) CONFIG.update(config_dict) train_logs_dir = os.path.join(logdir, 'train.logs')\",\n \"# pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr,\",\n \"callbacks=[]): \\\"\\\"\\\"Get embeddings from a one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and CONFIG.DATA.FRAME_LABELS\",\n \"steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels\",\n \"np.concatenate([get_context_steps(step) for step in steps]) single_steps = np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps\",\n \"want to automatically # reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE ==\",\n \"global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s not\",\n \"plt # pylint: disable=g-import-not-at-top import io import math import os import time from\",\n \"labels_list # Reset statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE\",\n \"num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' %\",\n \"compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0, 1) plt.tight_layout() buf = io.BytesIO()\",\n \"# Since model is a dict we can insert multiple modular networks in\",\n \"summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im,\",\n \"dict we can insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects)\",\n \"= np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens =\",\n \"else: logging.info( 'Using config from config.json that exists in %s.', logdir) with tf.io.gfile.GFile(config_path,\",\n \"for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats,\",\n \"axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32 as\",\n \"def get_data(iterator): \\\"\\\"\\\"Return a data dict which contains all the requested sequences.\\\"\\\"\\\" data\",\n \"return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer\",\n \"if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list =\",\n \"* tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current learning rate based\",\n \"* frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps\",\n \"in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0]\",\n \"restore=False, **ckpt_objects): # Since model is a dict we can insert multiple modular\",\n \"and that can take # more than one batch to be passed and\",\n \"in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2]\",\n \"CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists\",\n \"'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data,\",\n \"cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps = CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE\",\n \"summary_image doesn't take float64 sim_matrix = sim_matrix.astype(np.float32) tf.summary.image('%s/similarity_matrix' % split, np.expand_dims(sim_matrix, axis=3), step=global_step)\",\n \"keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels\",\n \"lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR)\",\n \"with open(config_path) as f: config = json.load(f) assert config is not None, \\\"config\",\n \"curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j]\",\n \"= CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list =\",\n \"inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config = None\",\n \"None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data dict which contains\",\n \"= CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the future.\",\n \"1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size):\",\n \"= [image_list[0] [num_frames_per_step - 1::num_frames_per_step][::2]] else: im_list = [image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix =\",\n \"status = checkpoint.restore( ckpt_manager.latest_checkpoint) if restore else -1 return ckpt_manager, status, checkpoint def\",\n \"# np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) #\",\n \"not os.path.exists(config_path) or overwrite: logging.info( 'Using the existing passed in config as no\",\n \"axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if\",\n \"== 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ]\",\n \"learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set\",\n \"reset hidden states after each batch. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in\",\n \"embs_list = [] labels_list = [] steps_list = [] seq_lens_list = [] names_list\",\n \"== len(steps) assert len(names) == len(steps) embs_list.append(embs) if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens)\",\n \"and restore checkpoint (if one exists on the path).\\\"\\\"\\\" # Instantiate checkpoint and\",\n \"assert config is not None, \\\"config file is not provided or is corrupted\\\"\",\n \"opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER,\",\n \"while not tf.io.gfile.exists(config_path): logging.info('Waiting for config to exist. Going to sleep ' '\",\n \"keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a one epoch iterator.\\\"\\\"\\\" keep_labels\",\n \"as some modalities might not exist. if len(v.shape) > 1 and v.shape[1] !=\",\n \"don't want to see the future. steps = np.arange(step - (num_steps - 1)\",\n \"status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if one exists on\",\n \"AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and global\",\n \"CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step):\",\n \"data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn, data, training, num_steps=None):\",\n \"__future__ import division from __future__ import print_function from config import CONFIG import json\",\n \"config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator):\",\n \"np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for f in callbacks: f(np.concatenate(embs), data, chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0])\",\n \"config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path = os.path.join(logdir,\",\n \"np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy',\",\n \"frames = [] if optical_flow: frame_original = [] # Reset GRU states for\",\n \"num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step\",\n \"tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models in inference mode.\",\n \"GPUS: if ind > -1: tf.config.experimental.set_visible_devices(GPUS[ind], 'GPU') try: # Currently, memory growth needs\",\n \"see the future. steps = np.arange(step - (num_steps - 1) * stride, step\",\n \"json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists in\",\n \"* num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step, axis=0)[num_frames_per_step\",\n \"EasyDict, contains params required to initialize the learning rate and the learning rate\",\n \"idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps,\",\n \"image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image,\",\n \"random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return random_cycles def get_warmup_lr(lr, global_step, lr_params): \\\"\\\"\\\"Returns learning rate\",\n \"> 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] =\",\n \"to exist. Going to sleep ' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while\",\n \"[] if optical_flow: frame_original_list = [] n = 0 def cond(n): if max_embs\",\n \"training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats =\",\n \"we don't want to automatically # reset hidden states after each batch. if\",\n \"checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint)\",\n \"as f: config = json.load(f) assert config is not None, \\\"config file is\",\n \"mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1)\",\n \"lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr,\",\n \"tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as e: # Memory\",\n \"indent=4) else: logging.info( 'Using config from config.json that exists in %s.', logdir) with\",\n \"elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0,\",\n \"= [] seq_labels_list = [] if keep_data: frames_list = [] if optical_flow: frame_original_list\",\n \"(batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats = emb_feats[i] if\",\n \"num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries =\",\n \"tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries)) def get_lr_fn(optimizer_config):\",\n \"= np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if keep_data:\",\n \"idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else: curr_data[k] =\",\n \"lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning\",\n \"c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v)) for k, v in\",\n \"manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config,\",\n \"model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 *\",\n \"== 'manual': lr_step_boundaries = [int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence\",\n \"get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 * num_steps)\",\n \"config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS: if ind\",\n \"if isinstance(config, list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return\",\n \"model['cnn'] emb = model['emb'] embs_list = [] labels_list = [] steps_list = []\",\n \"buffer to TF image image = tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension\",\n \"directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config): if\",\n \"in range(num_batches): if (i + 1) * frames_per_batch > seq_len: num_steps = seq_len\",\n \"lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float\",\n \"tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step, split): \\\"\\\"\\\"Visualize nearest\",\n \"model['cnn'] emb = model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data,\",\n \"a directory that already ' 'has train_logs. Please provide a new logdir name\",\n \"tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img =\",\n \"required to initialize the learning rate and the learning rate decay function. Returns:\",\n \"lr + is_warmup * warmup_lr return lr # Minimally adapted from Tensorflow object_detection\",\n \"iterator, frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a one epoch\",\n \"is_warmup) * lr + is_warmup * warmup_lr return lr # Minimally adapted from\",\n \"stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range(curr_idx, curr_idx + num_steps)\",\n \"in data.items(): # Need to do this as some modalities might not exist.\",\n \"a new logdir name in ' 'config or pass --force_train while launching script.')\",\n \"disable=g-import-not-at-top import io import math import os import time from absl import flags\",\n \"no config.json file exists in ' '%s', logdir) with tf.io.gfile.GFile(config_path, 'w') as config_file:\",\n \"seq_labels = [] if keep_data: frames = [] if optical_flow: frame_original = []\",\n \"n += 1 except tf.errors.OutOfRangeError: logging.info('Finished embedding the dataset.') break dataset = {'embs':\",\n \"(if one exists on the path).\\\"\\\"\\\" # Instantiate checkpoint and restore from any\",\n \"the existing passed in config as no config.json file exists in ' '%s',\",\n \"tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn'\",\n \"len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) == len(steps) embs_list.append(embs) if\",\n \"NumPy version: np.random.rand(m,n).argsort(axis=axis) return tf.cast(tf.argsort(tf.random.uniform((m, n)), axis=axis), tf.int64) def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate\",\n \"= emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps *\",\n \"config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config, config_file, sort_keys=True,\",\n \"sequence number %d, frames embedded %d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'),\",\n \"seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs, axis=0) labels = np.concatenate(labels, axis=0) steps\",\n \"[] labels = [] steps = [] seq_lens = [] names = []\",\n \"'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] =\",\n \"Instantiate checkpoint and restore from any pre-existing checkpoint. # Since model is a\",\n \"with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds)\",\n \"tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i],\",\n \"not None, \\\"config file is not provided or is corrupted\\\" return config def\",\n \"= tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data,\",\n \"types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step,\",\n \"num_frames_per_step = CONFIG.DATA.NUM_STEPS cnn = model['cnn'] emb = model['emb'] embs_list = [] labels_list\",\n \"num_steps = seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx =\",\n \"else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats = tf.stack(tf.split(emb_feats,\",\n \"not force_train: raise ValueError('You might be overwriting a directory that already ' 'has\",\n \"batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step,\",\n \"Generate m random permuations of range (0, n) # NumPy version: np.random.rand(m,n).argsort(axis=axis) return\",\n \"def get_cnn_feats(cnn, data, training, num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if num_steps is\",\n \"in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping( global_step, lr_step_boundaries, learning_rate_sequence) elif\",\n \"frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries = tf.concat(batch_list, axis=1)\",\n \"axis=0) labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0)\",\n \".') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and global step.\\\"\\\"\\\" optimizer\",\n \"optical_flow: frame_original = [] # Reset GRU states for each video. if CONFIG.MODEL.EMBEDDER_TYPE\",\n \"'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE\",\n \"name = data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels)\",\n \"= embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens) assert len(embs) == len(steps) assert len(names) ==\",\n \"= seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1)\",\n \"= lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)]\",\n \"candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0)\",\n \"labels = np.concatenate(labels, axis=0) steps = np.concatenate(steps, axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names\",\n \"= lr_params.INITIAL_LR * warmup_percent_done is_warmup = tf.cast(global_steps_int < warmup_steps_int, tf.float32) lr = (1.0\",\n \"in kwargs: raise ValueError('Function called with set_learning_phase decorator which' ' does not have\",\n \"seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0) if\",\n \"decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step),\",\n \"lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif lr_params.DECAY_TYPE == 'manual':\",\n \"np.concatenate(np.array(list(map(get_context_steps, np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps)\",\n \"corrupted\\\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if GPUS:\",\n \"mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models in inference\",\n \"def random_choice_noreplace(m, n, axis=-1): # Generate m random permuations of range (0, n)\",\n \"to False to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return\",\n \"frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps =\",\n \"num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries))\",\n \"def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE\",\n \"Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0]\",\n \"break dataset = {'embs': embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list}\",\n \"max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a one epoch iterator.\\\"\\\"\\\" keep_labels = keep_labels and\",\n \"config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists in %s.',\",\n \"CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0)\",\n \"in emb.gru_layers: gru_layer.stateful = False return dataset def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot,\",\n \"= [(lr_params.INITIAL_LR) * f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step):\",\n \"tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0) batch_summaries\",\n \"num_steps) emb_feats = tf.stack(tf.split(emb_feats, num_steps, axis=0), axis=1) query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames\",\n \"decay type %s not supported. Only support' 'the following decay types: fixed, exp_decay,\",\n \"seq_len): steps = range(curr_idx, curr_idx + num_steps) single_steps = np.concatenate([get_context_steps(step) for step in\",\n \"a batch.\\\"\\\"\\\" frames = data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list,\",\n \"if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32)\",\n \"= tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index,\",\n \"lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries\",\n \"matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames\",\n \"Make Recurrent Layers stateful, set batch size. # We do this as we\",\n \"get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on config and\",\n \"== 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay( lr, global_step, lr_params.EXP_DECAY_STEPS, lr_params.EXP_DECAY_RATE, staircase=True)() elif\",\n \"return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint (if one\",\n \"[] steps = [] seq_lens = [] names = [] seq_labels = []\",\n \"warmup_lr return lr # Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries,\",\n \"function f after setting proper learning phase.\\\"\\\"\\\" if 'training' not in kwargs: raise\",\n \"np.arange(curr_idx, curr_idx + num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return\",\n \"learning rate during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step,\",\n \"= tf.image.decode_png(buf.getvalue(), channels=4) # Add the batch dimension image = tf.expand_dims(image, 0) return\",\n \"if optical_flow: frame_original = [] # Reset GRU states for each video. if\",\n \"from easydict import EasyDict import matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step,\",\n \"\\\"\\\"\\\" lr_params = optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr,\",\n \"axis=0) seq_lens = np.concatenate(seq_lens, axis=0) names = np.concatenate(names, axis=0) seq_labels = np.concatenate(seq_labels, axis=0)\",\n \"calling function f.\\\"\\\"\\\" def wrapper(*args, **kwargs): \\\"\\\"\\\"Calls the function f after setting proper\",\n \"Recurrent Layers stateful, set batch size. # We do this as we are\",\n \"any pre-existing checkpoint. # Since model is a dict we can insert multiple\",\n \"ValueError('Function called with set_learning_phase decorator which' ' does not have training argument.') training\",\n \"[0] + boundaries num_boundaries = len(boundaries) rate_index = tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)),\",\n \"**ckpt_objects): # Since model is a dict we can insert multiple modular networks\",\n \"global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr,\",\n \"frames_per_batch, keep_data=False, optical_flow=False, keep_labels=True, max_embs=None, callbacks=[]): \\\"\\\"\\\"Get embeddings from a one epoch iterator.\\\"\\\"\\\"\",\n \"seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps = frames_per_batch curr_idx\",\n \"tf.reduce_max( tf.where( tf.greater_equal(global_step, boundaries), list(range(num_boundaries)), [0] * num_boundaries)) return tf.reduce_sum(rates * tf.one_hot(rate_index, depth=num_boundaries))\",\n \"def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and global step.\\\"\\\"\\\" optimizer = get_optimizer(CONFIG.OPTIMIZER, CONFIG.OPTIMIZER.LR.INITIAL_LR)\",\n \"tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path):\",\n \"chosen_steps, seq_len) steps.append(chosen_steps.numpy()[0]) seq_lens.append(seq_len * [seq_len]) all_labels = data['frame_labels'].numpy()[0] name = data['name'].numpy()[0] names.append(seq_len\",\n \"'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9)\",\n \"tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError\",\n \"params have been passed in the config. \\\"\\\"\\\" lr_params = optimizer_config.LR # pylint:\",\n \"directory that already ' 'has train_logs. Please provide a new logdir name in\",\n \"= (1.0 - is_warmup) * lr + is_warmup * warmup_lr return lr #\",\n \"as np import matplotlib.pyplot as plt # pylint: disable=g-import-not-at-top import io import math\",\n \"Add the batch dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \\\"\\\"\\\"Simple\",\n \"and not force_train: raise ValueError('You might be overwriting a directory that already '\",\n \"= data['frames'] frames_list = tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list =\",\n \"> seq_len: num_steps = seq_len - i * frames_per_batch else: num_steps = frames_per_batch\",\n \"is a dict we can insert multiple modular networks in this dict. checkpoint\",\n \"CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps =\",\n \"os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a directory that already\",\n \"tf.gather(v, idxes, axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step *\",\n \"decay function. Returns: lr_fn: function, this can be called to return the current\",\n \"for gpu in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus),\",\n \"% split, np.expand_dims(sim_matrix, axis=3), step=global_step) def softmax(w, t=1.0): e = np.exp(np.array(w) / t)\",\n \"rate. Args: optimizer_config: EasyDict, contains params required to initialize the learning rate and\",\n \"axis=-1): # Generate m random permuations of range (0, n) # NumPy version:\",\n \"= optimizer.iterations learning_rate = optimizer.learning_rate return learning_rate, optimizer, global_step def create_ckpt(logdir, restore=False, **ckpt_objects):\",\n \"dataset['frames'] = frames_list if optical_flow: dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list\",\n \"json import tensorflow as tf import numpy as np import matplotlib.pyplot as plt\",\n \"sequence and that can take # more than one batch to be passed\",\n \"print_function from config import CONFIG import json import tensorflow as tf import numpy\",\n \"some modalities might not exist. if len(v.shape) > 1 and v.shape[1] != 0:\",\n \"names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs = np.concatenate(embs,\",\n \"**kwargs) return wrapper def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as\",\n \"- self.time def done(self, target_interval): return self.elapsed() >= target_interval def reset(self): self.time =\",\n \"1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1])\",\n \"optimizer_config: EasyDict, contains params required to initialize the learning rate and the learning\",\n \"during warm up phase.\\\"\\\"\\\" if lr_params.NUM_WARMUP_STEPS > 0: global_steps_int = tf.cast(global_step, tf.int32) warmup_steps_int\",\n \"import json import tensorflow as tf import numpy as np import matplotlib.pyplot as\",\n \"def load_config(config_path): config = None if os.path.exists(config_path): with open(config_path) as f: config =\",\n \"in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1)\",\n \"[] seq_lens_list = [] names_list = [] seq_labels_list = [] if keep_data: frames_list\",\n \"1) * stride, step + stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len):\",\n \"the current learning rate based on the provided config. Raises: ValueError: in case\",\n \"in CONFIG.items()]) json.dump(config, config_file, sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that\",\n \"contains all the requested sequences.\\\"\\\"\\\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function\",\n \"be called to return the current learning rate based on the provided config.\",\n \"sort_keys=True, indent=4) else: logging.info( 'Using config from config.json that exists in %s.', logdir)\",\n \"[image_list[0][num_frames_per_step - 1::num_frames_per_step]] sim_matrix = np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in\",\n \"2 * num_steps) emb_feats = tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else:\",\n \"def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries = len(boundaries) rate_index\",\n \"axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1],\",\n \"exist. if len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps,\",\n \"num_steps))))) single_steps = np.maximum(0, single_steps) single_steps = np.minimum(seq_len, single_steps) return single_steps def get_embeddings_dataset(model,\",\n \"be overwriting a directory that already ' 'has train_logs. Please provide a new\",\n \"else: return config def setup_train_dir(logdir, overwrite=False, force_train=True): \\\"\\\"\\\"Setups directory for training.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) config_path\",\n \"split): \\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\" # Set learning_phase to False to\",\n \"to do this as some modalities might not exist. if len(v.shape) > 1\",\n \"the learning rate and the learning rate decay function. Returns: lr_fn: function, this\",\n \"want to see the future. steps = np.arange(step - (num_steps - 1) *\",\n \"True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is None:\",\n \"num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list = tf.unstack(image_list[i], num=num_steps * num_frames_per_step,\",\n \"assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names = names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)]\",\n \"which' ' does not have training argument.') training = kwargs['training'] if training: #\",\n \"after setting proper learning phase.\\\"\\\"\\\" if 'training' not in kwargs: raise ValueError('Function called\",\n \"config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if\",\n \"if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else: num_steps = CONFIG.EVAL.NUM_FRAMES * CONFIG.DATA.NUM_STEPS\",\n \"' ' %s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as\",\n \"tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict = json.load(config_file) if config_dict is None: time.sleep(config_timeout_seconds) else:\",\n \"axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0]) frame_original = np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)]\",\n \"= get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if\",\n \"data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in\",\n \"# Set learning_phase to True to use models in training mode. tf.keras.backend.set_learning_phase(1) else:\",\n \"= data['frames'] image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list =\",\n \"tf.io.gfile.GFile(config_path, 'w') as config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()])\",\n \"if os.path.exists(train_logs_dir) and not force_train: raise ValueError('You might be overwriting a directory that\",\n \"def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir,\",\n \"'optimizers: AdamOptimizer, MomentumOptimizer .') return opt def get_lr_opt_global_step(): \\\"\\\"\\\"Intializes learning rate, optimizer and\",\n \"in ' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1):\",\n \"= np.zeros( (batch_size-1, num_steps, num_steps), dtype=np.float32) for i in range(1, batch_size): candidate_feats =\",\n \"def create_ckpt(logdir, restore=False, **ckpt_objects): # Since model is a dict we can insert\",\n \"'w') as config_file: config = dict([(k, to_dict(v)) for k, v in CONFIG.items()]) json.dump(config,\",\n \"frames_list = [] if optical_flow: frame_original_list = [] n = 0 def cond(n):\",\n \"assign to update the learning rate. Args: optimizer_config: EasyDict, contains params required to\",\n \"= seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original\",\n \"num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step] nn_img_list = [] for j in range(num_steps): curr_query_feats =\",\n \"this can be called to return the current learning rate based on the\",\n \"= i * frames_per_batch curr_data = {} for k, v in data.items(): #\",\n \"tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for\",\n \"return n < max_embs # Make Recurrent Layers stateful, set batch size. #\",\n \"time from absl import flags from absl import logging from easydict import EasyDict\",\n \"update the learning rate. Args: optimizer_config: EasyDict, contains params required to initialize the\",\n \"emb(cnn_feats, num_steps) logging.debug('On sequence number %d, frames embedded %d', n, curr_idx + num_steps)\",\n \"CNN.\\\"\\\"\\\" if num_steps is None: if training: num_steps = CONFIG.TRAIN.NUM_FRAMES * CONFIG.DATA.NUM_STEPS else:\",\n \"os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config is not None,\",\n \"return dataset def gen_plot(x, y): \\\"\\\"\\\"Create a pyplot, save to buffer and return\",\n \"axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)]\",\n \"matplotlib matplotlib.use('Agg') FLAGS = flags.FLAGS def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\"\",\n \"query_feats = emb_feats[0] if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list\",\n \"# We do this as we are embedding the whole sequence and that\",\n \"if optical_flow: frame_original_list = [] n = 0 def cond(n): if max_embs is\",\n \"timer for measuring elapsed time.\\\"\\\"\\\" def __init__(self): self.reset() def elapsed(self): return time.time() -\",\n \"data['name'].numpy()[0] names.append(seq_len * [name]) seq_label = data['seq_labels'].numpy()[0] seq_labels.append(seq_len * [seq_label]) labels.append(all_labels) embs =\",\n \"- i * frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch\",\n \"curr_data = {} for k, v in data.items(): # Need to do this\",\n \"rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt = tf.keras.optimizers.Adam(learning_rate=learning_rate) elif optimizer_config.TYPE == 'MomentumOptimizer': opt\",\n \"in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to use models\",\n \"seq_len = get_data(iterator) seq_len = seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches):\",\n \"Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries = [0] + boundaries num_boundaries\",\n \"else: return n < max_embs # Make Recurrent Layers stateful, set batch size.\",\n \"= get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps) emb_feats = emb(cnn_feats, 2 *\",\n \"= get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats = emb(cnn_feats, num_steps) logging.debug('On sequence\",\n \"is corrupted\\\" return config def prepare_gpu(ind=-1): ind = int(ind) GPUS = tf.config.experimental.list_physical_devices('GPU') if\",\n \"= names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)]\",\n \"list): return [to_dict(c) for c in config] elif isinstance(config, EasyDict): return dict([(k, to_dict(v))\",\n \"' 'config or pass --force_train while launching script.') tf.io.gfile.makedirs(train_logs_dir) def setup_eval_dir(logdir, config_timeout_seconds=1): \\\"\\\"\\\"Setups\",\n \"if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2 * num_steps)\",\n \"'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = True gru_layer.input_spec[0].shape = [1, ] while\",\n \"tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager,\",\n \"[1, ] while cond(n): try: print(n) embs = [] labels = [] steps\",\n \"to buffer and return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy') plt.ylim(0,\",\n \"if restore else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and\",\n \"= tf.unstack(frames, num=num_steps, axis=1) frames_summaries = tf.concat(frames_list, axis=2) batch_list = tf.split(frames_summaries, batch_size, axis=0)\",\n \"optical_flow: frame_original_list = [] n = 0 def cond(n): if max_embs is None:\",\n \"keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert\",\n \"np.concatenate(frame_original, axis=0) if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels =\",\n \"len(v.shape) > 1 and v.shape[1] != 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k]\",\n \"call assign to update the learning rate. Args: optimizer_config: EasyDict, contains params required\",\n \"learning rate. Args: optimizer_config: EasyDict, contains params required to initialize the learning rate\",\n \"pyplot, save to buffer and return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val\",\n \"' 'has train_logs. Please provide a new logdir name in ' 'config or\",\n \"mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1, j] = softmax(-1.0 * mean_squared_distance) nn_img_list.append(img_list[tf.argmin(mean_squared_distance)])\",\n \"channels=4) # Add the batch dimension image = tf.expand_dims(image, 0) return image class\",\n \"1) * frames_per_batch > seq_len: num_steps = seq_len - i * frames_per_batch else:\",\n \"this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager = tf.train.CheckpointManager( checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status\",\n \"= e / np.sum(e) return dist def random_choice_noreplace(m, n, axis=-1): # Generate m\",\n \"dataset['labels'] = labels_list # Reset statefulness to recurrent layers for other evaluation tasks.\",\n \"%d', n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"])\",\n \"and restore from any pre-existing checkpoint. # Since model is a dict we\",\n \"directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not tf.io.gfile.exists(config_path):\",\n \"= np.concatenate(seq_labels, axis=0) if keep_data: frames.append(data['frames'].numpy()[0]) frames = np.concatenate(frames, axis=0) if optical_flow: frame_original.append(data['video_frames'].numpy()[0])\",\n \"* CONFIG.DATA.NUM_STEPS cnn.num_steps = num_steps cnn_feats = cnn(data['frames']) return cnn_feats def get_context_steps(step): num_steps\",\n \"dataset['frames_original'] = frame_original_list if keep_labels: dataset['labels'] = labels_list # Reset statefulness to recurrent\",\n \"elif lr_params.DECAY_TYPE == 'fixed': def lr_fn(lr, global_step): return lr_params.INITIAL_LR elif lr_params.DECAY_TYPE == 'poly':\",\n \"axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im = tf.expand_dims(tf.concat([vstack(im) for im\",\n \"def gen_cycles(num_cycles, batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:,\",\n \"tf.stack( tf.split(emb_feats, 2 * num_steps, axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False)\",\n \"for im in im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert\",\n \"'poly': def lr_fn(lr, global_step): return tf.train.polynomial_decay( lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else:\",\n \"decorator which' ' does not have training argument.') training = kwargs['training'] if training:\",\n \"!= 0: idxes = get_indices(curr_idx, num_steps, seq_len) curr_data[k] = tf.gather(v, idxes, axis=1) else:\",\n \"dimension image = tf.expand_dims(image, 0) return image class Stopwatch(object): \\\"\\\"\\\"Simple timer for measuring\",\n \"save to buffer and return TB compatible image.\\\"\\\"\\\" plt.figure() plt.plot(x, y) plt.title('Val Accuracy')\",\n \"lr, global_step, CONFIG.TRAIN.MAX_ITERS, end_learning_rate=0.0, power=1.0, cycle=False) else: raise ValueError('Learning rate decay type %s\",\n \"data dict which contains all the requested sequences.\\\"\\\"\\\" data = iterator.get_next() return data,\",\n \"support' 'the following decay types: fixed, exp_decay, manual,' 'and poly.') return (lambda lr,\",\n \"names[~np.isnan(embs).any(axis=1)] seq_lens = seq_lens[~np.isnan(embs).any(axis=1)] steps = steps[~np.isnan(embs).any(axis=1)] if keep_data: frames = frames[~np.isnan(embs).any(axis=1)] if\",\n \"Reset statefulness to recurrent layers for other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru':\",\n \"config_dict is None: time.sleep(config_timeout_seconds) else: break CONFIG.update(config_dict) def get_data(iterator): \\\"\\\"\\\"Return a data dict\",\n \"get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE ==\",\n \"the requested sequences.\\\"\\\"\\\" data = iterator.get_next() return data, data['chosen_steps'], data['seq_lens'] @tf.function def get_cnn_feats(cnn,\",\n \"np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) # np.save(tf.io.gfile.GFile('/air/team/saman/test_cnn_old.npy', 'w'), cnn_feats.numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_emb_old.npy', 'w'), emb_feats.numpy()) embs.append(emb_feats.numpy()) for\",\n \"if keep_labels: labels = labels[~np.isnan(embs).any(axis=1)] assert len(embs) == len(labels) seq_labels = seq_labels[~np.isnan(embs).any(axis=1)] names\",\n \"poly.') return (lambda lr, global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns\",\n \"[int(x) for x in lr_params.MANUAL_LR_STEP_BOUNDARIES] f = lr_params.MANUAL_LR_DECAY_RATE learning_rate_sequence = [(lr_params.INITIAL_LR) * f**p\",\n \"seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original) n += 1 except tf.errors.OutOfRangeError:\",\n \"checkpoint, directory=logdir, max_to_keep=10, keep_checkpoint_every_n_hours=1) status = checkpoint.restore(ckpt_manager.latest_checkpoint) return ckpt_manager, status, checkpoint def to_dict(config):\",\n \"models in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn'\",\n \"i in range(1, batch_size): candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list =\",\n \"contains params required to initialize the learning rate and the learning rate decay\",\n \"nn_img_list.append(img_list[tf.argmin(mean_squared_distance)]) nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im\",\n \"axis=1) else: curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False)\",\n \"frames_per_batch else: num_steps = frames_per_batch curr_idx = i * frames_per_batch curr_data = {}\",\n \"self.reset() def elapsed(self): return time.time() - self.time def done(self, target_interval): return self.elapsed() >=\",\n \"optimizer_config.LR # pylint: disable=g-long-lambda if lr_params.DECAY_TYPE == 'exp_decay': def lr_fn(lr, global_step): return tf.train.exponential_decay(\",\n \"function that provides current learning rate based on config. NOTE: This returns a\",\n \"use models in training mode. tf.keras.backend.set_learning_phase(1) else: # Set learning_phase to False to\",\n \"stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps =\",\n \"img_list = tf.unstack(image_list[i], num=2 * num_steps * num_frames_per_step, axis=0)[num_frames_per_step - 1::num_frames_per_step][::2] else: img_list\",\n \"as in Eager we need to call assign to update the learning rate.\",\n \"rate decay type %s not supported. Only support' 'the following decay types: fixed,\",\n \"elif optimizer_config.TYPE == 'MomentumOptimizer': opt = tf.keras.optimizers.SGD( learning_rate=learning_rate, momentum=0.9) else: raise ValueError('Optimizer %s\",\n \"num_steps=None): \\\"\\\"\\\"Passes data through base CNN.\\\"\\\"\\\" if num_steps is None: if training: num_steps\",\n \"None if os.path.exists(config_path): with open(config_path) as f: config = json.load(f) assert config is\",\n \"CONFIG.DATA.NUM_STEPS stride = CONFIG.DATA.FRAME_STRIDE # We don't want to see the future. steps\",\n \"seq_len.numpy()[0] num_batches = int(math.ceil(float(seq_len)/frames_per_batch)) for i in range(num_batches): if (i + 1) *\",\n \"whole sequence and that can take # more than one batch to be\",\n \"im_list], axis=0), axis=0) tf.summary.image('%s/nn' % split, summary_im, step=global_step) # Convert sim_matrix to float32\",\n \"= model['emb'] if 'tcn' in CONFIG.TRAINING_ALGO: cnn_feats = get_cnn_feats( cnn, data, training=False, num_steps=2\",\n \"# Minimally adapted from Tensorflow object_detection code. def manual_stepping(global_step, boundaries, rates): boundaries =\",\n \"* f**p for p in range(len(lr_step_boundaries) + 1)] def lr_fn(lr, global_step): return manual_stepping(\",\n \"= tf.concat(batch_list, axis=1) tf.summary.image('train_batch', batch_summaries, step=global_step) def visualize_nearest_neighbours(model, data, global_step, batch_size, num_steps, num_frames_per_step,\",\n \"restore checkpoint (if one exists on the path).\\\"\\\"\\\" # Instantiate checkpoint and restore\",\n \"%s for secs.', config_timeout_seconds) time.sleep(config_timeout_seconds) while True: with tf.io.gfile.GFile(config_path, 'r') as config_file: config_dict\",\n \"cond(n): if max_embs is None: return True else: return n < max_embs #\",\n \"= io.BytesIO() plt.savefig(buf, format='png') buf.seek(0) # Convert PNG buffer to TF image image\",\n \"image_list = tf.unstack(frames, num=batch_size, axis=0) if 'tcn' in CONFIG.TRAINING_ALGO: im_list = [image_list[0] [num_frames_per_step\",\n \"n, curr_idx + num_steps) # np.save(tf.io.gfile.GFile('/air/team/saman/test_weights_old.npy', 'w'), cnn.weights[0].numpy()) # np.save(tf.io.gfile.GFile('/air/team/saman/test_batch_old.npy', 'w'), curr_data[\\\"frames\\\"]) #\",\n \"if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful = False return dataset\",\n \"frames[~np.isnan(embs).any(axis=1)] if optical_flow: frame_original = frame_original[~np.isnan(embs).any(axis=1)] embs = embs[~np.isnan(embs).any(axis=1)] assert len(embs) == len(seq_lens)\",\n \"we need to call assign to update the learning rate. Args: optimizer_config: EasyDict,\",\n \"embs_list, 'seq_lens': seq_lens_list, 'steps': steps_list, 'names': names_list, 'seq_labels': seq_labels_list} if keep_data: dataset['frames'] =\",\n \"tf.int32) warmup_steps_int = tf.constant( lr_params.NUM_WARMUP_STEPS, dtype=tf.int32) global_steps_float = tf.cast(global_steps_int, tf.float32) warmup_steps_float = tf.cast(warmup_steps_int,\",\n \"on config. NOTE: This returns a function as in Eager we need to\",\n \"do this as we are embedding the whole sequence and that can take\",\n \"import time from absl import flags from absl import logging from easydict import\",\n \"to use models in inference mode. tf.keras.backend.set_learning_phase(0) return f(*args, **kwargs) return wrapper def\",\n \"range(num_steps): curr_query_feats = tf.tile(query_feats[j:j+1], [num_steps, 1]) mean_squared_distance = tf.reduce_mean( tf.math.squared_difference(curr_query_feats, candidate_feats), axis=1) sim_matrix[i-1,\",\n \"if CONFIG.OPTICALFLOW: frames = data['video_frames'] else: frames = data['frames'] image_list = tf.unstack(frames, num=batch_size,\",\n \"axis=0)[::2], axis=1) else: cnn_feats = get_cnn_feats(cnn, data, training=False) emb_feats = emb(cnn_feats, num_steps) emb_feats\",\n \"def visualize_batch(data, global_step, batch_size, num_steps): \\\"\\\"\\\"Visualizes a batch.\\\"\\\"\\\" frames = data['frames'] frames_list =\",\n \"in inference mode. tf.keras.backend.set_learning_phase(0) cnn = model['cnn'] emb = model['emb'] if 'tcn' in\",\n \"global_step: get_warmup_lr(lr_fn(lr, global_step), global_step, lr_params)) def get_optimizer(optimizer_config, learning_rate): \\\"\\\"\\\"Returns optimizer based on config\",\n \"\\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"]) except RuntimeError as e: # Memory growth must\",\n \"= [] if keep_data: frames = [] if optical_flow: frame_original = [] #\",\n \"momentum=0.9) else: raise ValueError('Optimizer %s not supported. Only support the following' 'optimizers: AdamOptimizer,\",\n \"in GPUS: tf.config.experimental.set_memory_growth(gpu, True) logical_gpus = tf.config.experimental.list_logical_devices('GPU') logging.info([len(GPUS), \\\"Physical GPUs,\\\", len(logical_gpus), \\\"Logical GPUs\\\"])\",\n \"is None: return True else: return n < max_embs # Make Recurrent Layers\",\n \"other evaluation tasks. if CONFIG.MODEL.EMBEDDER_TYPE == 'convgru': for gru_layer in emb.gru_layers: gru_layer.stateful =\",\n \"learning rate and the learning rate decay function. Returns: lr_fn: function, this can\",\n \"def set_learning_phase(f): \\\"\\\"\\\"Sets the correct learning phase before calling function f.\\\"\\\"\\\" def wrapper(*args,\",\n \"\\\"\\\"\\\"Setups directory for evaluation.\\\"\\\"\\\" tf.io.gfile.makedirs(logdir) tf.io.gfile.makedirs(os.path.join(logdir, 'eval_logs')) config_path = os.path.join(logdir, 'config.json') while not\",\n \"import absolute_import from __future__ import division from __future__ import print_function from config import\",\n \"we can insert multiple modular networks in this dict. checkpoint = tf.train.Checkpoint(**ckpt_objects) ckpt_manager\",\n \"def get_lr_fn(optimizer_config): \\\"\\\"\\\"Returns function that provides current learning rate based on config. NOTE:\",\n \"nn_img = tf.stack(nn_img_list, axis=0) im_list.append(nn_img) def vstack(im): return tf.concat(tf.unstack(im, num=num_steps), axis=1) summary_im =\",\n \"\\\"\\\"\\\"Returns optimizer based on config and learning rate.\\\"\\\"\\\" if optimizer_config.TYPE == 'AdamOptimizer': opt\",\n \"if keep_labels: labels_list.append(labels) seq_labels_list.append(seq_labels) steps_list.append(steps) seq_lens_list.append(seq_lens) names_list.append(names) if keep_data: frames_list.append(frames) if optical_flow: frame_original_list.append(frame_original)\",\n \"\\\"\\\"\\\"Visualize nearest neighbours in embedding space.\\\"\\\"\\\" # Set learning_phase to False to use\",\n \"EasyDict): return dict([(k, to_dict(v)) for k, v in config.items()]) else: return config def\",\n \"else -1 return ckpt_manager, status, checkpoint def restore_ckpt(logdir, **ckpt_objects): \\\"\\\"\\\"Create and restore checkpoint\",\n \"candidate_feats = emb_feats[i] if 'tcn' in CONFIG.TRAINING_ALGO: img_list = tf.unstack(image_list[i], num=2 * num_steps\",\n \"batch_size, cycle_len): \\\"\\\"\\\"Generate cycles for alignment.\\\"\\\"\\\" random_cycles = random_choice_noreplace( num_cycles, batch_size)[:, :cycle_len] return\"\n]","isConversational":false}}},{"rowIdx":3593,"cells":{"ngram":{"kind":"list like","value":["= reader.read_float() if version >= (5, 4, 2) or ( version >= (5,","(4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4,","self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >>","up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ]","2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in","Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures","= reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags =","# 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019","for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up","(2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version","in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up m_BonesSize =","writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits)","self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name =","# 2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): #",">= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot","range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up m_BonesSize = reader.read_int()",".Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr,","or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3","SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for","< (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >=","[ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5, 6): #","(2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2):","reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal =","writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if","up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up","(first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader)","and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture =","range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and up SubMeshesSize =","self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5,","2) or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch ): #","Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in","vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018","reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer =","if version >= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream()","PPtr(reader) # Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize =","version[:2] <= (4, 3): # 4.3 and down self.uv = reader.read_vector2() def save(self,","kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight =","# 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version","= reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down self.uv =",">= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes)","(2017,): # 2017 and up first = reader.read_bytes(16) # GUID second = reader.read_long()","self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version","and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): #","version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,):","PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum):","def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1","None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >=","reader): version = reader.version self.texture = PPtr(reader) # Texture2D if version >= (5,","2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version >=","and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4,","if version >= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones =","class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw","= VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)]","# 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation =","self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw","self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of the","if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version","writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset)","writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6","self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize)","# 4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017","writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if","reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down","version = reader.version self.texture = PPtr(reader) # Texture2D if version >= (5, 2):","<= (4, 3): # 4.3 and down self.uv = reader.read_vector2() def save(self, writer,","kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle =","m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if","self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6","2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in","# Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture =","if version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version","range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer is None: writer","= self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1)","version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class","def save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2): #","self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up","= reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >=","and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize)","enum import IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject","and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3):","writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and",">= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if version >=","] if version >= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones","and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5):","reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _","SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def","(5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,):","= reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up first =","and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6):","self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize =","PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import","..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject):","and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up","EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version","< (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect","6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _","SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) &","2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize)","reader.read_vector2() if version >= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4()","self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if","5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up","if version >= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if","= (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1","the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader):","version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if",".PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from","reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices","# 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): #","version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self,","version >= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes =","(5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >=","reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer)","] if version >= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int()","# 2017 and up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey","from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType","version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3","SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value","image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f()","reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >=",">= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version","(5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if",">= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer,","self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5","Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D","version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5","self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if","return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw &","SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of the bits are","self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if","writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID","self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) &","up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version)","2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in","save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical","writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >=","if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape:","# 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5","import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader)","version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone)","self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys","version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer)","version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer,","SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def","tex.save(writer) if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh","reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin","reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): #","(1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight",">= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader)","import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr","for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): #","up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect)","if version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures =","mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices:","2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) #","SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <=","version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >=","self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,)","self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down self.uv","up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon)",">> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1)","writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if","# 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3","3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017","writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if","reader.read_u_int() if version >= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean()","): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): #","6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): #",">= (2017,): # 2017 and up first = reader.read_bytes(16) # GUID second =","= reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw =","version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version","= PPtr(reader) # Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize","self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier =","import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property","= _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >>","reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6 and up","# Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone","reader.read_vector2() if version >= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2()","& 1) # 1 # rest of the bits are reserved def save(self,","reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2):","1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw","self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up","(2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer)","writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name","def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2()","and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and","PPtr(reader) # Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture","(2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for","writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader) #","super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >=","reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value","= [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream()","if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def","= [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >=","_ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up m_BonesSize","reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value):","writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version","1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) #","(2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2","self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter","return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset","def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,)","down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and","def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def","and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0])","self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for","2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): #","= reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3 and","version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version","Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version","up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2)","up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize =","(2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for","self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version >=","= self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and","(5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): #","= (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 =","= PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017","4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and","3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude =","_value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1)","are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version =","# 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017 and up","down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader)","= (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,)","2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6):","= reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int()","= reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): #","5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream()","down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4,","_ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader)","version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer)","reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if","reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5 and up","# GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas","# Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int()","for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): #","vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose)","( version >= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and","& 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4","# SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and up","and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize)","bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version","and up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and up","version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)]","GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas =","= reader.read_vector2() if version >= (5, 6): # 5.6 and up self.atlasRectOffset =","version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >=",">= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version):","up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version","SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def","from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import","1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude","4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of","@value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 #","in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and","version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in","and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5,","__init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if",">= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): #","1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version","writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version <","version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes:","(4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if","2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,):","= reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version","version >= (2017,): # 2017 and up first = reader.read_bytes(16) # GUID second","5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017 and","VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices","2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset =","# 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,):","kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings:",">= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if","[SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,):","and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer)","5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and","6) & 1) # 1 # rest of the bits are reserved def","PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and","version): self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2 and up","if version[:2] <= (4, 3): # 4.3 and down self.uv = reader.read_vector2() def","writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else:","= reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose =","= (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD =","# 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer)","writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer)","(2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas","4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self,","second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader)","# 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _","kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value =","@property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect","4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture","value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw","if version >= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes","and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up","if version >= (2017,): # 2017 and up first = reader.read_bytes(16) # GUID","value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode","# 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >=","NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums import","up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize","3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >=","writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version >= (5, 4,","reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,):","second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if","save(self, writer: EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version","= reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version","up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second)","self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize =","class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader):","_value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode =","[BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): #","reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5 and up","# 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture:","__init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3):","and up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): # 2019","save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos =","(5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for","5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)]","1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType","verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array()","2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6","= PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class","(2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle","self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) #","def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed =","if version >= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits","if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset)","up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures))","class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null()","reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas","and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self,","version >= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if version","bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader)","m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def","and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if","and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and up","<= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader):","= reader.read_u_int() if version >= (5, 3): # 5.3 and up self.m_IsPolygon =","version >= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits =","self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3 and up self.m_IsPolygon","SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape","5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017","writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in","= 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self):","writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv)","if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if","rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def","6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version","(2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect =","= SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of the bits","3): # 4.3 and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos)","self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): # 2019 and up","and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2 down","version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2):","VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export import","up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform)","= reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int()","def __init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D if version","self.texture = PPtr(reader) # Texture2D if version >= (5, 2): # 5.2 and","if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version","range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize","4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017 and up writer.write_float(self.downscaleMultiplier)","is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version","# 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType =","def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and","# 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >=","4, 2) or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch ):",">= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): #","self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for","# 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >=","writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream()","reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >=","_ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer is","phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones))","up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): #","(2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version","self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in","5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and","writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData:","reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if","= [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6):","& 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1","version >= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up","= SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int()","# 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in","version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4,","= 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def","for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer","_ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018","# 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer,","Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader)","if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if","self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,):","SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset =","version >= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [","writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version)","self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version","# rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex:","__init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer):","5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2)","class SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2]","from enum import IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import","for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture =","3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version =","# 1 # rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw)","SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize",">= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >=","= SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2)","SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export","5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5,","(4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017","1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation","self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >>","version >= (5, 4, 2) or ( version >= (5, 4, 1, 3)","= [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): #","up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018","version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version <","in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for","reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down self.uv = reader.read_vector2()","# 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >=","if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin =","(2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class","writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version >= (5,","self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or","self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw","self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5","class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 =","writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2 and","version >= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if",">= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018,","if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version","EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version",">= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,):","up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version","from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class","class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D","import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self)","5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version","# 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >=","IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr","and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <=","writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream()","writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and","tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and up","in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer is None:","super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border)","and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if","_ in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and up","..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return","[ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018","& 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1","self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or ( version >=","Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone =","kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90","(4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version","if version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self,","self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1","reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for","reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version","(5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version","if version >= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw","self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up","in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and","writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4,","up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version >=","def save(self, writer: EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian)","reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3","reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array()","kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class","6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer))",">> 6) & 1) # 1 # rest of the bits are reserved","): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version","(5, 4, 2) or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch","(0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property","if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones:","= self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5):","for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData =","= reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal","__init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self,",">> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF)","2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f()","(0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4","up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): #","= PPtr(reader) # Texture2D if version >= (5, 2): # 5.2 and up","4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or (","# GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,):","reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array()",">= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader)","= None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer)","(4, 3): # 4.3 and down self.uv = reader.read_vector2() def save(self, writer, version):","= reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or (","writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180","mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for","reader.read_float() if version >= (5, 4, 2) or ( version >= (5, 4,","reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >=","2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D","3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,):","SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) &","# 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset","self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5 and up self.uvTransform","and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize)",">= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if","for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices))","0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 #","2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version)","if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags)","else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices =","= reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5 and","reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and","[ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None):","EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): #","1) # 1 # rest of the bits are reserved def save(self, writer):","reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def","in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex","= reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) #","[SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData","= reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6 and","of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self,","self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD","def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos","# Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex","5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4,","1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return","for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and","5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): #","else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >=",">= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array()","for _ in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and","SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter","def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self,","up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader):","# 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): #","(2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5,","4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5,","# 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _","@property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed","self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3):","self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6 and up self.atlasRectOffset","self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in","if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in","..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader):","= [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5, 6):","# 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down",">= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4,","version >= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D","1 # rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class","writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017","= EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5):","reader.version self.texture = PPtr(reader) # Texture2D if version >= (5, 2): # 5.2","= reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version","writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up","and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version","up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) #","up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier","version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [","up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ]","in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes))","writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,):","(5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude)","self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices =","self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose","writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture","in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and up SubMeshesSize",">= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader)","if version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape =","# Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) #","in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else:","version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [","2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2","reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version","reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3 and up",".NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from","writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if","4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if","down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2()","= reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float()","from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper","and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer,","= reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and","up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ]","if version >= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) #","class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version =","writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if","= reader.read_vector2() if version >= (4, 5): # 4.5 and up self.m_Border =","writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3()","up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): # 2019 and","(5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5):","self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up first","and up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first,","= reader.version self.texture = PPtr(reader) # Texture2D if version >= (5, 2): #",">= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot)","= reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self,","# 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) #","reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or ( version","reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up first = reader.read_bytes(16)","self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name)","1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw","= reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5,","range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up","version >= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw =","GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): #","save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down","= SpriteSettings(reader) if version >= (4, 5): # 4.5 and up self.uvTransform =",">= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array()","self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer)","5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and","(2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for","= (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum):","self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3):","= (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int()","BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from","in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) #","from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self):","= [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter =","(4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,):","# 4.3 and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if","# SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape))","version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version","import IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from","import PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams","self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D","if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures:","def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4,","# 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or","2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def","4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2()","first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags","self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version","version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >=",">= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float()","if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >=","= reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5 and","self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5,","SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6","self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone","and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize","] def save(self, writer: EndianBinaryWriter = None): if writer is None: writer =","save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter","kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def","writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices)","(5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot =","if version >= (5, 4, 2) or ( version >= (5, 4, 1,","(3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class","writer: EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version =","None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect)",">= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if","reader.align_stream() if version >= (2017,): # 2017 and up first = reader.read_bytes(16) #","writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1])","writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical =","save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2","# 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and","SpriteSettings(reader) if version >= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4()","self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3","= SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6)",">= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5,","for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up","and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int()","import NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums","from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self,","5): # 4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,): #","self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): #","SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int()","(5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version)","reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if","self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,)",">= (5, 4, 2) or ( version >= (5, 4, 1, 3) and","= reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader)",">= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) #","def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect =","self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): #","SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,)","4.3 and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2]","SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D if","__init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D if version >=","2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5,","= reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream()","# 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _","version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys)","= reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer","2017 and up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey =","up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): #","self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5 and up self.m_Border","reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer:","# 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest","secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if","IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize =","if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018,","self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset)","version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer,"],"string":"[\n \"= reader.read_float() if version >= (5, 4, 2) or ( version >= (5,\",\n \"(4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4,\",\n \"self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >>\",\n \"up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ]\",\n \"2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in\",\n \"Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures\",\n \"= reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags =\",\n \"# 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019\",\n \"for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up\",\n \"(2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version\",\n \"in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up m_BonesSize =\",\n \"writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits)\",\n \"self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name =\",\n \"# 2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): #\",\n \">= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot\",\n \"range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up m_BonesSize = reader.read_int()\",\n \".Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr,\",\n \"or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3\",\n \"SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for\",\n \"< (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >=\",\n \"[ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5, 6): #\",\n \"(2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2):\",\n \"reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal =\",\n \"writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if\",\n \"up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up\",\n \"(first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader)\",\n \"and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture =\",\n \"range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and up SubMeshesSize =\",\n \"self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5,\",\n \"2) or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch ): #\",\n \"Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in\",\n \"vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018\",\n \"reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer =\",\n \"if version >= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream()\",\n \"PPtr(reader) # Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize =\",\n \"version[:2] <= (4, 3): # 4.3 and down self.uv = reader.read_vector2() def save(self,\",\n \"kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight =\",\n \"# 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version\",\n \"= reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down self.uv =\",\n \">= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes)\",\n \"(2017,): # 2017 and up first = reader.read_bytes(16) # GUID second = reader.read_long()\",\n \"self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version\",\n \"and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): #\",\n \"version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,):\",\n \"PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum):\",\n \"def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1\",\n \"None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >=\",\n \"reader): version = reader.version self.texture = PPtr(reader) # Texture2D if version >= (5,\",\n \"2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version >=\",\n \"and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4,\",\n \"if version >= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones =\",\n \"class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw\",\n \"= VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)]\",\n \"# 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation =\",\n \"self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw\",\n \"self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of the\",\n \"if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version\",\n \"writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset)\",\n \"writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6\",\n \"self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize)\",\n \"# 4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017\",\n \"writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if\",\n \"reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down\",\n \"version = reader.version self.texture = PPtr(reader) # Texture2D if version >= (5, 2):\",\n \"<= (4, 3): # 4.3 and down self.uv = reader.read_vector2() def save(self, writer,\",\n \"kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle =\",\n \"m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if\",\n \"self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6\",\n \"2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in\",\n \"# Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture =\",\n \"if version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version\",\n \"range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer is None: writer\",\n \"= self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1)\",\n \"version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class\",\n \"def save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2): #\",\n \"self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up\",\n \"= reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >=\",\n \"and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize)\",\n \"enum import IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject\",\n \"and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3):\",\n \"writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and\",\n \">= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if version >=\",\n \"] if version >= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones\",\n \"and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5):\",\n \"reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _\",\n \"SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def\",\n \"(5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,):\",\n \"= reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up first =\",\n \"and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6):\",\n \"self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize =\",\n \"PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import\",\n \"..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject):\",\n \"and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up\",\n \"EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version\",\n \"< (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect\",\n \"6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _\",\n \"SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) &\",\n \"2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize)\",\n \"reader.read_vector2() if version >= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4()\",\n \"self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if\",\n \"5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up\",\n \"if version >= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if\",\n \"= (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1\",\n \"the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader):\",\n \"version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if\",\n \".PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from\",\n \"reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices\",\n \"# 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): #\",\n \"version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self,\",\n \"version >= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes =\",\n \"(5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >=\",\n \"reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer)\",\n \"] if version >= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int()\",\n \"# 2017 and up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey\",\n \"from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType\",\n \"version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3\",\n \"SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value\",\n \"image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f()\",\n \"reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >=\",\n \">= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version\",\n \"(5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if\",\n \">= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer,\",\n \"self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5\",\n \"Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D\",\n \"version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5\",\n \"self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if\",\n \"return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw &\",\n \"SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of the bits are\",\n \"self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if\",\n \"writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID\",\n \"self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) &\",\n \"up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version)\",\n \"2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in\",\n \"save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical\",\n \"writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >=\",\n \"if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape:\",\n \"# 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5\",\n \"import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader)\",\n \"version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone)\",\n \"self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys\",\n \"version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer)\",\n \"version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer,\",\n \"SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def\",\n \"tex.save(writer) if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh\",\n \"reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin\",\n \"reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): #\",\n \"(1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight\",\n \">= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader)\",\n \"import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr\",\n \"for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): #\",\n \"up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect)\",\n \"if version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures =\",\n \"mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices:\",\n \"2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) #\",\n \"SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <=\",\n \"version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >=\",\n \"self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,)\",\n \"self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down self.uv\",\n \"up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon)\",\n \">> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1)\",\n \"writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if\",\n \"# 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3\",\n \"3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017\",\n \"writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if\",\n \"reader.read_u_int() if version >= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean()\",\n \"): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): #\",\n \"6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): #\",\n \">= (2017,): # 2017 and up first = reader.read_bytes(16) # GUID second =\",\n \"= reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw =\",\n \"version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version\",\n \"= PPtr(reader) # Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize\",\n \"self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier =\",\n \"import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property\",\n \"= _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >>\",\n \"reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6 and up\",\n \"# Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone\",\n \"reader.read_vector2() if version >= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2()\",\n \"& 1) # 1 # rest of the bits are reserved def save(self,\",\n \"reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2):\",\n \"1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw\",\n \"self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up\",\n \"(2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer)\",\n \"writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name\",\n \"def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2()\",\n \"and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and\",\n \"PPtr(reader) # Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture\",\n \"(2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for\",\n \"writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader) #\",\n \"super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >=\",\n \"reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value\",\n \"= [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream()\",\n \"if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def\",\n \"= [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >=\",\n \"_ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and up m_BonesSize\",\n \"reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value):\",\n \"writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version\",\n \"1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) #\",\n \"(2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2\",\n \"self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter\",\n \"return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset\",\n \"def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,)\",\n \"down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and\",\n \"def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def\",\n \"and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0])\",\n \"self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for\",\n \"2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): #\",\n \"= reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3 and\",\n \"version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version\",\n \"Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version\",\n \"up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2)\",\n \"up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize =\",\n \"(2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for\",\n \"self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version >=\",\n \"= self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and\",\n \"(5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): #\",\n \"= (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 =\",\n \"= PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017\",\n \"4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and\",\n \"3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude =\",\n \"_value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1)\",\n \"are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version =\",\n \"# 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017 and up\",\n \"down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader)\",\n \"= (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,)\",\n \"2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6):\",\n \"= reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int()\",\n \"= reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): #\",\n \"5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream()\",\n \"down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4,\",\n \"_ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader)\",\n \"version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer)\",\n \"reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if\",\n \"reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5 and up\",\n \"# GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas\",\n \"# Texture2D if version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int()\",\n \"for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): #\",\n \"vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose)\",\n \"( version >= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and\",\n \"& 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4\",\n \"# SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and up\",\n \"and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize)\",\n \"bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version\",\n \"and up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and up\",\n \"version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)]\",\n \"GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas =\",\n \"= reader.read_vector2() if version >= (5, 6): # 5.6 and up self.atlasRectOffset =\",\n \"version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >=\",\n \">= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version):\",\n \"up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version\",\n \"SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def\",\n \"from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import\",\n \"1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude\",\n \"4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of\",\n \"@value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 #\",\n \"in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and\",\n \"version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in\",\n \"and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5,\",\n \"__init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if\",\n \">= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): #\",\n \"1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version\",\n \"writer.align_stream() if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version <\",\n \"version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes:\",\n \"(4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if\",\n \"2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,):\",\n \"= reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version\",\n \"version >= (2017,): # 2017 and up first = reader.read_bytes(16) # GUID second\",\n \"5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017 and\",\n \"VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices\",\n \"2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset =\",\n \"# 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,):\",\n \"kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings:\",\n \">= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if\",\n \"[SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,):\",\n \"and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer)\",\n \"5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and\",\n \"6) & 1) # 1 # rest of the bits are reserved def\",\n \"PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and\",\n \"version): self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2 and up\",\n \"if version[:2] <= (4, 3): # 4.3 and down self.uv = reader.read_vector2() def\",\n \"writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else:\",\n \"= reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose =\",\n \"= (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD =\",\n \"# 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer)\",\n \"writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer)\",\n \"(2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas\",\n \"4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self,\",\n \"second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader)\",\n \"# 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _\",\n \"kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value =\",\n \"@property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect\",\n \"4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture\",\n \"value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw\",\n \"if version >= (5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes\",\n \"and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up\",\n \"if version >= (2017,): # 2017 and up first = reader.read_bytes(16) # GUID\",\n \"value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode\",\n \"# 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >=\",\n \"NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums import\",\n \"up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize\",\n \"3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >=\",\n \"writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version >= (5, 4,\",\n \"reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,):\",\n \"second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if\",\n \"save(self, writer: EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version\",\n \"= reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version\",\n \"up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second)\",\n \"self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize =\",\n \"class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader):\",\n \"_value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1 # 1 self.packingMode =\",\n \"[BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): #\",\n \"reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5 and up\",\n \"# 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture:\",\n \"__init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3):\",\n \"and up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): # 2019\",\n \"save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos =\",\n \"(5, 6): # 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for\",\n \"5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)]\",\n \"1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType\",\n \"verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array()\",\n \"2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6\",\n \"= PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class\",\n \"(2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle\",\n \"self.m_settingsRaw & 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) #\",\n \"def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed =\",\n \"if version >= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits\",\n \"if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset)\",\n \"up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures))\",\n \"class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null()\",\n \"reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas\",\n \"and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self,\",\n \"version >= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if version\",\n \"bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader)\",\n \"m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def\",\n \"and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if\",\n \"and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and up\",\n \"<= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader):\",\n \"= reader.read_u_int() if version >= (5, 3): # 5.3 and up self.m_IsPolygon =\",\n \"version >= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits =\",\n \"self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3 and up self.m_IsPolygon\",\n \"SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape\",\n \"5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017\",\n \"writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in\",\n \"= 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self):\",\n \"writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv)\",\n \"if version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if\",\n \"rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def\",\n \"6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version\",\n \"(2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect =\",\n \"= SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest of the bits\",\n \"3): # 4.3 and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos)\",\n \"self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): # 2019 and up\",\n \"and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2 down\",\n \"version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2):\",\n \"VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export import\",\n \"up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform)\",\n \"= reader.read_matrix_array() if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int()\",\n \"def __init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D if version\",\n \"self.texture = PPtr(reader) # Texture2D if version >= (5, 2): # 5.2 and\",\n \"if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version\",\n \"range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize\",\n \"4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017 and up writer.write_float(self.downscaleMultiplier)\",\n \"is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version\",\n \"# 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType =\",\n \"def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and\",\n \"# 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >=\",\n \"4, 2) or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch ):\",\n \">= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): #\",\n \"self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for\",\n \"# 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >=\",\n \"writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream()\",\n \"reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >=\",\n \"_ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer is\",\n \"phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones))\",\n \"up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): #\",\n \"(2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version\",\n \"self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in\",\n \"5.2 and up self.alphaTexture.save(writer) # Texture2D if version >= (2019,): # 2019 and\",\n \"writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData:\",\n \"reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if\",\n \"= [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6):\",\n \"& 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1\",\n \"version >= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up\",\n \"= SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int()\",\n \"# 5.6 and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in\",\n \"version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4,\",\n \"= 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class SpriteSettings: def\",\n \"for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer\",\n \"_ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018\",\n \"# 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer,\",\n \"Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader)\",\n \"if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if\",\n \"self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,):\",\n \"SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset =\",\n \"version >= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [\",\n \"writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version)\",\n \"self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version\",\n \"# rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex:\",\n \"__init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer):\",\n \"5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2)\",\n \"class SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2]\",\n \"from enum import IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import\",\n \"for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture =\",\n \"3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version =\",\n \"# 1 # rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw)\",\n \"SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): # 2017 and up m_PhysicsShapeSize\",\n \">= (5, 2): # 5.2 and up self.alphaTexture.save(writer) # Texture2D if version >=\",\n \"= SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2)\",\n \"SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export\",\n \"5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5,\",\n \"(4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >= (2017,): # 2017\",\n \"1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation\",\n \"self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >>\",\n \"version >= (5, 4, 2) or ( version >= (5, 4, 1, 3)\",\n \"= [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): #\",\n \"up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018\",\n \"version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version <\",\n \"in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for\",\n \"reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and down self.uv = reader.read_vector2()\",\n \"# 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >=\",\n \"if version < (2018, 2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin =\",\n \"(2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class\",\n \"writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version >= (5,\",\n \"self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or\",\n \"self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw\",\n \"self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5\",\n \"class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 =\",\n \"writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2 and\",\n \"version >= (5, 3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if\",\n \">= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018,\",\n \"if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version\",\n \"EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version\",\n \">= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,):\",\n \"up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version\",\n \"from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class\",\n \"class SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D\",\n \"import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self)\",\n \"5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version\",\n \"# 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >=\",\n \"IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr\",\n \"and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <=\",\n \"writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream()\",\n \"writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and\",\n \"tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and up\",\n \"in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None): if writer is None:\",\n \"super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border)\",\n \"and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if\",\n \"_ in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and up\",\n \"..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return\",\n \"[ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018\",\n \"& 1 # 1 self.packingMode = SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1\",\n \"self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or ( version >=\",\n \"Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone =\",\n \"kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90\",\n \"(4, 3): # 4.3 and down writer.write__vector2(self.uv) class SpriteRenderData: def __init__(self, reader): version\",\n \"if version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def save(self,\",\n \"self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed = self.m_settingsRaw & 1\",\n \"reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for\",\n \"reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version\",\n \"(5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version\",\n \"if version >= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw\",\n \"self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up\",\n \"in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,): # 2018 and\",\n \"writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4,\",\n \"up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version >=\",\n \"def save(self, writer: EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian)\",\n \"reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3\",\n \"reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array()\",\n \"kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class\",\n \"6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer))\",\n \">> 6) & 1) # 1 # rest of the bits are reserved\",\n \"): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version\",\n \"(5, 4, 2) or ( version >= (5, 4, 1, 3) and self.build_type.IsPatch\",\n \"(0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property\",\n \"if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones:\",\n \"= self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5):\",\n \"for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData =\",\n \"= reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal\",\n \"__init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter def value(self,\",\n \">> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF)\",\n \"2): # 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f()\",\n \"(0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4\",\n \"up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): #\",\n \"= PPtr(reader) # Texture2D if version >= (5, 2): # 5.2 and up\",\n \"4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or (\",\n \"# GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,):\",\n \"reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array()\",\n \">= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader)\",\n \"= None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer)\",\n \"(4, 3): # 4.3 and down self.uv = reader.read_vector2() def save(self, writer, version):\",\n \"= reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or (\",\n \"writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180\",\n \"mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for\",\n \"reader.read_float() if version >= (5, 4, 2) or ( version >= (5, 4,\",\n \"reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >=\",\n \"2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D\",\n \"3): # 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,):\",\n \"SpritePackingMode((self.m_settingsRaw >> 1) & 1) # 1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) &\",\n \"# 2018.2 down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset\",\n \"self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5 and up self.uvTransform\",\n \"and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _ in range(m_PhysicsShapeSize)\",\n \">= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if\",\n \"for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices))\",\n \"0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 #\",\n \"2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version)\",\n \"if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags)\",\n \"else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices =\",\n \"= reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): # 4.5 and\",\n \"reader.read_vector2_array() for _ in range(m_PhysicsShapeSize) ] if version >= (2018,): # 2018 and\",\n \"[ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter = None):\",\n \"EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5): #\",\n \"1) # 1 # rest of the bits are reserved def save(self, writer):\",\n \"reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float() def\",\n \"in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex\",\n \"= reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) #\",\n \"[SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData\",\n \"= reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6 and\",\n \"of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self,\",\n \"self.m_RenderDataKey = (first, second) self.m_AtlasTags = reader.read_string_array() self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD\",\n \"def save(self, writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos\",\n \"# Texture2D if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex\",\n \"5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4,\",\n \"1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return\",\n \"for tex in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and\",\n \"5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): #\",\n \"else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >=\",\n \">= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array()\",\n \"for _ in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and\",\n \"SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def value(self): return self.m_settingsRaw @value.setter\",\n \"def __init__(self, reader): self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self,\",\n \"up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader):\",\n \"# 5.3 and up self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): #\",\n \"(2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5,\",\n \"4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5,\",\n \"# 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for _\",\n \"@property def value(self): return self.m_settingsRaw @value.setter def value(self, _value): self.m_settingsRaw = _value self.packed\",\n \"self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3):\",\n \"self.textureRectOffset = reader.read_vector2() if version >= (5, 6): # 5.6 and up self.atlasRectOffset\",\n \"self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader) for _ in\",\n \"if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in\",\n \"..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader):\",\n \"= [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5, 6):\",\n \"# 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018, 2): # 2018.2 down\",\n \">= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4,\",\n \"version >= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D\",\n \"1 # rest of the bits are reserved def save(self, writer): writer.write_u_int(self.m_settingsRaw) class\",\n \"writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017\",\n \"= EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect) writer.write_vector2(self.m_Offset) if version >= (4, 5):\",\n \"reader.version self.texture = PPtr(reader) # Texture2D if version >= (5, 2): # 5.2\",\n \"= reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version\",\n \"writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up\",\n \"and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or ( version\",\n \"up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) #\",\n \"up self.uvTransform = reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier\",\n \"version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [\",\n \"up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ]\",\n \"in self.secondaryTextures: tex.save(writer) if version >= (5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes))\",\n \"writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices) writer.align_stream() if version >= (2018,):\",\n \"(5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude)\",\n \"self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices =\",\n \"self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose\",\n \"writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture\",\n \"in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6 and up SubMeshesSize\",\n \">= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader)\",\n \"if version >= (2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape =\",\n \"# Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer) #\",\n \"in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else:\",\n \"version >= (2019,): # 2019 and up secondaryTexturesSize = reader.read_int() self.secondaryTextures = [\",\n \"2018 and up self.m_Bindpose = reader.read_matrix_array() if version < (2018, 2): # 2018.2\",\n \"reader): super().__init__(reader=reader) version = self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version\",\n \"reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3 and up\",\n \".NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from\",\n \"writer.write_vector2(self.m_Offset) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if\",\n \"4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if\",\n \"down self.m_SourceSkinSize = reader.read_int() self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2()\",\n \"= reader.read_vector4() if version >= (2017,): # 2017 and up self.downscaleMultiplier = reader.read_float()\",\n \"from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper\",\n \"and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer,\",\n \"= reader.version self.pos = reader.read_vector3() if version[:2] <= (4, 3): # 4.3 and\",\n \"up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ]\",\n \"if version >= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) #\",\n \"class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version =\",\n \"writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if\",\n \"= reader.read_vector2() if version >= (4, 5): # 4.5 and up self.m_Border =\",\n \"writer): writer.write_u_int(self.m_settingsRaw) class SpriteVertex: def __init__(self, reader): version = reader.version self.pos = reader.read_vector3()\",\n \"up self.alphaTexture = PPtr(reader) # Texture2D if version >= (2019,): # 2019 and\",\n \"(5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset) self.settingsRaw.save(writer) if version >= (4, 5):\",\n \"self.m_IsPolygon = reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up first\",\n \"and up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first,\",\n \"= reader.version self.texture = PPtr(reader) # Texture2D if version >= (5, 2): #\",\n \">= (5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot)\",\n \"= reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self,\",\n \"# 2017 and up self.downscaleMultiplier = reader.read_float() def save(self, writer, version): self.texture.save(writer) #\",\n \"reader.read_vector4() self.m_PixelsToUnits = reader.read_float() if version >= (5, 4, 2) or ( version\",\n \"reader.read_boolean() reader.align_stream() if version >= (2017,): # 2017 and up first = reader.read_bytes(16)\",\n \"self.texture = PPtr(reader) # Texture2D self.name = reader.read_string_to_null() def save(self, writer): self.texture.save(writer) writer.write_string_to_null(self.name)\",\n \"1 self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw\",\n \"= reader.read_float() def save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5,\",\n \"range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream() if version >= (2018,): # 2018 and up\",\n \"version >= (5, 6): # 5.6 and up self.atlasRectOffset = reader.read_vector2() self.settingsRaw =\",\n \"GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) # SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): #\",\n \"save(self, writer, version): writer.write_vector3(self.pos) if version[:2] <= (4, 3): # 4.3 and down\",\n \"= SpriteSettings(reader) if version >= (4, 5): # 4.5 and up self.uvTransform =\",\n \">= (2018,): # 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array()\",\n \"self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2 and up self.alphaTexture.save(writer)\",\n \"5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3): # 5.3 and\",\n \"(2017,): # 2017 and up m_PhysicsShapeSize = reader.read_int() self.m_PhysicsShape = [ reader.read_vector2_array() for\",\n \"= (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum):\",\n \"self.build_type.IsPatch ): # 5.4.1p3 and up writer.write_vector2(self.m_Pivot) writer.write_u_int(self.m_Extrude) if version >= (5, 3):\",\n \"= (0,) kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int()\",\n \"BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from .PPtr import PPtr, save_ptr from\",\n \"in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) #\",\n \"from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self):\",\n \"= [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer: EndianBinaryWriter =\",\n \"(4, 5): # 4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,):\",\n \"# 4.3 and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if\",\n \"# SpriteAtlas self.m_RD.save(writer, version) if version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape))\",\n \"version < (2018, 2): # 2018.2 down writer.write_int(self.m_SourceSkinSize) self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version\",\n \"import IntEnum from .Mesh import BoneWeights4, SubMesh, VertexData from .NamedObject import NamedObject from\",\n \"import PPtr, save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams\",\n \"self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def __init__(self, reader): self.texture = PPtr(reader) # Texture2D\",\n \"if version >= (2019,): # 2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures:\",\n \"def __init__(self, reader): version = reader.version self.pos = reader.read_vector3() if version[:2] <= (4,\",\n \"# 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4, 2) or\",\n \"2018 and up writer.write_int(len(self.m_Bones)) for bone in self.m_Bones: writer.write_vector2_array(bone) self.set_raw_data(writer.bytes) class SecondarySpriteTexture: def\",\n \"4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2()\",\n \"first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey = (first, second) self.m_AtlasTags\",\n \"self.settingsRaw.save(writer) if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version\",\n \"version >= (5, 3): # 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >=\",\n \">= (4, 5): # 4.5 and up self.m_Border = reader.read_vector4() self.m_PixelsToUnits = reader.read_float()\",\n \"if version >= (4, 5): # 4.5 and up writer.write_vector4(self.uvTransform) if version >=\",\n \"= reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5 and\",\n \"self.m_SourceSkin = [BoneWeights4(reader)] self.textureRect = reader.read_rectangle_f() self.textureRectOffset = reader.read_vector2() if version >= (5,\",\n \"SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if version >= (5, 6): # 5.6\",\n \"self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up writer.write_int(len(self.m_Bones)) for bone\",\n \"and up SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize\",\n \"] def save(self, writer: EndianBinaryWriter = None): if writer is None: writer =\",\n \"save_ptr from ..export import SpriteHelper from ..enums import SpriteMeshType from ..streams import EndianBinaryWriter\",\n \"kSPMRectangle = 1 class SpriteSettings: def __init__(self, reader): self.value = reader.read_u_int() @property def\",\n \"writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer, version) writer.write_u_short_array(self.indices)\",\n \"(5, 4, 1, 3) and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot =\",\n \"if version >= (5, 4, 2) or ( version >= (5, 4, 1,\",\n \"(3,) kSPRRotate90 = 4 class SpritePackingMode(IntEnum): kSPMTight = (0,) kSPMRectangle = 1 class\",\n \"writer: EndianBinaryWriter = None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version =\",\n \"None): if writer is None: writer = EndianBinaryWriter(endian=self.reader.endian) version = self.version super().save(writer) writer.write_rectangle_f(self.m_Rect)\",\n \">= (5, 2): # 5.2 and up self.alphaTexture = PPtr(reader) # Texture2D if\",\n \"reader.align_stream() if version >= (2017,): # 2017 and up first = reader.read_bytes(16) #\",\n \"writer.align_stream() if version >= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1])\",\n \"writer): self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical =\",\n \"save(self, writer, version): self.texture.save(writer) # Texture2D if version >= (5, 2): # 5.2\",\n \"# 5.3 and up writer.write_boolean(self.m_IsPolygon) writer.align_stream() if version >= (2017,): # 2017 and\",\n \"SpriteSettings(reader) if version >= (4, 5): # 4.5 and up self.uvTransform = reader.read_vector4()\",\n \"self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int() if version >= (5, 3): # 5.3\",\n \"= SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6)\",\n \">= (4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5,\",\n \"for phys in self.m_PhysicsShape: writer.write_vector2_array(phys) if version >= (2018,): # 2018 and up\",\n \"and self.build_type.IsPatch ): # 5.4.1p3 and up self.m_Pivot = reader.read_vector2() self.m_Extrude = reader.read_u_int()\",\n \"import NamedObject from .PPtr import PPtr, save_ptr from ..export import SpriteHelper from ..enums\",\n \"from ..streams import EndianBinaryWriter class Sprite(NamedObject): @property def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self,\",\n \"5): # 4.5 and up self.uvTransform = reader.read_vector4() if version >= (2017,): #\",\n \"self.version self.m_Rect = reader.read_rectangle_f() self.m_Offset = reader.read_vector2() if version >= (4, 5): #\",\n \"SubMeshesSize = reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int()\",\n \"(5, 6): # 5.6 and up writer.write_int(len(self.m_SubMeshes)) for mesh in self.m_SubMeshes: mesh.save(writer, version)\",\n \"reader.align_stream() if version >= (2018,): # 2018 and up self.m_Bindpose = reader.read_matrix_array() if\",\n \"self.texture.save(writer) writer.write_string_to_null(self.name) class SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,)\",\n \">= (5, 4, 2) or ( version >= (5, 4, 1, 3) and\",\n \"= reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize = reader.read_int() self.vertices = [SpriteVertex(reader)\",\n \">= (2017,): # 2017 and up writer.write_bytes(self.m_RenderDataKey[0]) # GUID writer.write_long(self.m_RenderDataKey[1]) writer.write_string_array(self.m_AtlasTags) self.m_SpriteAtlas.save(writer) #\",\n \"def image(self): return SpriteHelper.get_image_from_sprite(self) def __init__(self, reader): super().__init__(reader=reader) version = self.version self.m_Rect =\",\n \"self.m_SpriteAtlas = PPtr(reader) # SpriteAtlas self.m_RD = SpriteRenderData(reader) if version >= (2017,): #\",\n \"SpritePackingRotation(IntEnum): kSPRNone = (0,) kSPRFlipHorizontal = (1,) kSPRFlipVertical = (2,) kSPRRotate180 = (3,)\",\n \"4.3 and down self.uv = reader.read_vector2() def save(self, writer, version): writer.write_vector3(self.pos) if version[:2]\",\n \"SpriteRenderData: def __init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D if\",\n \"__init__(self, reader): version = reader.version self.texture = PPtr(reader) # Texture2D if version >=\",\n \"2019 and up writer.write_int(len(self.secondaryTextures)) for tex in self.secondaryTextures: tex.save(writer) if version >= (5,\",\n \"= reader.read_int() self.vertices = [SpriteVertex(reader) for _ in range(verticesSize)] self.indices = reader.read_u_short_array() reader.align_stream()\",\n \"# 2018 and up m_BonesSize = reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _\",\n \"version >= (2017,): # 2017 and up writer.write_int(len(self.m_PhysicsShape)) for phys in self.m_PhysicsShape: writer.write_vector2_array(phys)\",\n \"= reader.read_int() self.m_SubMeshes = [SubMesh(reader) for _ in range(SubMeshesSize)] IndexBufferSize = reader.read_int() self.m_IndexBuffer\",\n \"2017 and up first = reader.read_bytes(16) # GUID second = reader.read_long() self.m_RenderDataKey =\",\n \"up self.atlasRectOffset = reader.read_vector2() self.settingsRaw = SpriteSettings(reader) if version >= (4, 5): #\",\n \"self.m_Offset = reader.read_vector2() if version >= (4, 5): # 4.5 and up self.m_Border\",\n \"reader.read_int() self.m_Bones = [ reader.read_vector2_array() for _ in range(m_BonesSize) ] def save(self, writer:\",\n \"# 4 self.meshType = SpriteMeshType((self.m_settingsRaw >> 6) & 1) # 1 # rest\",\n \"secondaryTexturesSize = reader.read_int() self.secondaryTextures = [ SecondarySpriteTexture(reader) for _ in range(secondaryTexturesSize) ] if\",\n \"IndexBufferSize = reader.read_int() self.m_IndexBuffer = reader.read_bytes(IndexBufferSize) reader.align_stream() self.m_VertexData = VertexData(reader) else: verticesSize =\",\n \"if version >= (2018,): # 2018 and up writer.write_matrix_array(self.m_Bindpose) if version < (2018,\",\n \"self.m_SourceSkin[0].save(writer) writer.write_rectangle_f(self.textureRect) writer.write_vector2(self.textureRectOffset) if version >= (5, 6): # 5.6 and up writer.write_vector2(self.atlasRectOffset)\",\n \"version) writer.write_int(len(self.m_IndexBuffer)) writer.write_bytes(self.m_IndexBuffer) writer.align_stream() self.m_VertexData.save(writer, version) else: writer.write_int(len(self.vertices)) for vertex in self.vertices: vertex.save(writer,\"\n]","isConversational":false}}},{"rowIdx":3594,"cells":{"ngram":{"kind":"list like","value":["self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None):","= 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self):","ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'):","\"\"\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path =","else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines()","http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \"\"\"","row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a string that can be put in the","Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans = [] else:","right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0. else:","in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object","Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x =","last one # new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave =","else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) |","case=False, verbose=True): \"\"\" Search filter names for ``search_string``. If ``case`` is True, then","i1): \"\"\" Return unit-indexed filter, e.g., 161 = 2mass-j \"\"\" return self.filters[i1-1] def","self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if bp is not","\"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den","filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class","# new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) #","np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c","= [] header = '' for line in lines: if 'lambda_c' in line:","= FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput =","`pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\" self.name = name self.wave = wave self.throughput","Search filter names for ``search_string``. If ``case`` is True, then match case. \"\"\"","= 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get the MW","161 = 2mass-j \"\"\" return self.filters[i1-1] def names(self, verbose=True): \"\"\" Print the filter","search_string, case=False, verbose=True): \"\"\" Search filter names for ``search_string``. If ``case`` is True,","# Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans = []","f-lambda flux density, cgs # Interpolate to wavelength grid, no extrapolation vega_full =","is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self):","= [] for i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name","Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta =","is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with","grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput and Vega spectrum","rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth / self.throughput.max() return","'' for line in lines: if 'lambda_c' in line: if len(wave) > 0:","file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r')","is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0)","910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv)","lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for","conversion \"\"\" from astropy.constants import c import astropy.units as u try: import grizli.utils_c","{1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78:","astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda)","= plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if","object `pysynphot` filter bandpass \"\"\" self.name = name self.wave = wave self.throughput =","[] for i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name =","integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot =","self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \"\"\" Filter equivalent width","= FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) #","interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter","new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header =","def names(self, verbose=True): \"\"\" Print the filter names. \"\"\" if verbose: for i","if self.wave is None: print('Filter not defined.') return False if source_flux is None:","ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else:","'r') as fp: lines = fp.readlines() self.filename = file_path filters = [] wave","<filename>eazy/filters.py import numpy as np import os from astropy.table import Table from .","def NFILT(self): \"\"\" Number of filters in the list \"\"\" return len(self.filters) def","ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\"","filters.append(new_filter) self.filters = filters @property def NFILT(self): \"\"\" Number of filters in the","if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__()","class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c =","return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor","in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \"\"\" Dump the filter information","= [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i,","if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d}","Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\" self.name","interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same units absp =","= np.sqrt(num/den) return pivot @property def equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\"","Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full,","path='./'): \"\"\" Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as","source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0,","@property def NFILT(self): \"\"\" Number of filters in the list \"\"\" return len(self.filters)","np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator","from astropy.table import Table from . import utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"]","c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None,","Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum,","filter file. \"\"\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name))","width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self):","utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']])","search(self, search_string, case=False, verbose=True): \"\"\" Search filter names for ``search_string``. If ``case`` is","return rect @property def ctw95(self): \"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl","wave = [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) #","den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html","self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \"\"\" Compute","new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave =","utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE)","left=0, right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate","string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file))","array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp :","the MW extinction correction within the filter. Optionally supply a source spectrum. \"\"\"","if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path,","throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans)","full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot wavelength","[\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in","is True, then match case. \"\"\" import re if not case: search_string =","= search_string.upper() matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name if","self.filters[i1-1] def names(self, verbose=True): \"\"\" Print the filter names. \"\"\" if verbose: for",": optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\" self.name = name self.wave =","(w, t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'):","bp.name self.norm = 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave)","\"\"\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i","for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a string that can be put in","return '\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY filter","{1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \"\"\"","bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name","for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False)","\"\"\" Return unit-indexed filter, e.g., 161 = 2mass-j \"\"\" return self.filters[i1-1] def names(self,","= self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def","self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self,","= utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1,","Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None,","6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta","= res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput')","of f-lambda flux density, cgs # Interpolate to wavelength grid, no extrapolation vega_full","= np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is not None:","return string_list def write(self, file='xxx.res', verbose=True): \"\"\" Dump the filter information to a","\"\"\" Search filter names for ``search_string``. If ``case`` is True, then match case.","filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1,","print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20,","full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda","verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1,","{1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0],","# new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter)","if not case: search_string = search_string.upper() matched = [] for i in range(len(self.filters)):","= name self.wave = wave self.throughput = throughput self.Aflux = 1. # pysynphot","lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return","/ np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self):","from . import utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits')","self.throughput))] return '\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY","filter information to a filter file. \"\"\" fp = open(file,'w') for filter in","None: print('Filter not defined.') return False if source_flux is None: source_flux = self.throughput*0.+1","= os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as","label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is None: file_path","unit-indexed filter, e.g., 161 = 2mass-j \"\"\" return self.filters[i1-1] def names(self, verbose=True): \"\"\"","self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \"\"\" 95% cumulative throughput width","search_string.upper() matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name if not","{1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list","ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\" from astropy.constants import c import astropy.units as","``case`` is True, then match case. \"\"\" import re if not case: search_string","i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for","i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if","import numpy as np import os from astropy.table import Table from . import","fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w')","from astropy.constants import c import astropy.units as u try: import grizli.utils_c interp =","in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string,","None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if","False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam,","\"\"\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:])","= 2mass-j \"\"\" return self.filters[i1-1] def names(self, verbose=True): \"\"\" Print the filter names.","def rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth / self.throughput.max()","new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter)","return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'):","fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5)","eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax","else: return 1./delta @property def ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\" from astropy.constants","3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den","new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property","= interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0)","throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter","bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\" self.name = name self.wave","= np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \"\"\"","Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den =","np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e}","http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave)","[row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class","source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda","of filters in the list \"\"\" return len(self.filters) def __getitem__(self, i1): \"\"\" Return","to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full","in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i],","np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0,","throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] #","cgs # Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'],","| (self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda","= os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float)","case. \"\"\" import re if not case: search_string = search_string.upper() matched = []","# new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter","filters @property def NFILT(self): \"\"\" Number of filters in the list \"\"\" return","in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave =","---------- name : str Label name wave : array Wavelength array, in `astropy.units.Angstrom`.","\"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\"","Print the filter names. \"\"\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1,","wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full =","= utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave)","try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union","for line in lines: if 'lambda_c' in line: if len(wave) > 0: #","len(self.filters) def __getitem__(self, i1): \"\"\" Return unit-indexed filter, e.g., 161 = 2mass-j \"\"\"","\"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl)","wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave,","'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition:","[header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w,","except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not defined.') return","= np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0]","self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if","def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\" import","'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines","for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \"\"\" Dump the","Dump the filter information to a filter file. \"\"\" fp = open(file,'w') for","within the filter. Optionally supply a source spectrum. \"\"\" import astropy.units as u","self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a string that","self.throughput.max() return rect @property def ctw95(self): \"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\"","np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self):","import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax =","np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return","range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in","2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\" from","= np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux","None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self,","'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \"\"\"","def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY filter file. .. plot:: :include-source:","# pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput =","os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class","fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string,","'lambda_c' in line: if len(wave) > 0: # Make filter from lines already","10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get the MW extinction","= '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines =","not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose:","grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not defined.')","return self.filters[i1-1] def names(self, verbose=True): \"\"\" Print the filter names. \"\"\" if verbose:","f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV,","1. # pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput","trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one #","Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return","filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a string","2mass-j \"\"\" return self.filters[i1-1] def names(self, verbose=True): \"\"\" Print the filter names. \"\"\"","self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i]))","# Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0,","= self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \"\"\" 95% cumulative throughput","can be put in the EAZY filter file \"\"\" header = '{0} {1}","AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)]","EAZY filter file \"\"\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}'","dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return","res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend()","= np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\"","pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html","class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY filter file. ..","interp = utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x =","None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1.","search_string = search_string.upper() matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name","mag=True, source_lam=None, source_flux=None): \"\"\" Get the MW extinction correction within the filter. Optionally","f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux,","ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is","self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave,","= VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass object","names. \"\"\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list","i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \"\"\" Dump the filter","string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self,","= np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self,","np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density,","class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass object Parameters ----------","`pysynphot` filter bandpass \"\"\" self.name = name self.wave = wave self.throughput = throughput","= np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta","'.join(line.split()[1:]) wave = [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2])","read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave","filter names for ``search_string``. If ``case`` is True, then match case. \"\"\" import","same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num","np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters =","__repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction","Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput,","np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property","if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \"\"\" Compute AB-Vega","Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag:","import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is","absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full,","self.filters = filters @property def NFILT(self): \"\"\" Number of filters in the list","plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path","filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if","self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\"","self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list","filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans =","VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def","u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self,","# new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters","if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux,","NFILT(self): \"\"\" Number of filters in the list \"\"\" return len(self.filters) def __getitem__(self,","in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i","def ctw95(self): \"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt","filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name","def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\"","os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp:","verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter","\"\"\" import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError:","def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\" import astropy.units as u f99","95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95","left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0,","= [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last","new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \"\"\"","Compute AB-Vega conversion \"\"\" from astropy.constants import c import astropy.units as u try:","- lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber =","VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] =","filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose:","width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975],","else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def","lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber = int(line.split('Filter","> 0: # Make filter from lines already read in new_filter = FilterDefinition(name=header,","ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters',","self.Aflux = 1. # pysynphot Bandpass if bp is not None: self.wave =","extinction correction within the filter. Optionally supply a source spectrum. \"\"\" import astropy.units","Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from","return len(self.filters) def __getitem__(self, i1): \"\"\" Return unit-indexed filter, e.g., 161 = 2mass-j","__str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\" import astropy.units","Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber","np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def","file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res","names(self, verbose=True): \"\"\" Print the filter names. \"\"\" if verbose: for i in","in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp =","f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else:","e.g., 161 = 2mass-j \"\"\" return self.filters[i1-1] def names(self, verbose=True): \"\"\" Print the","is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition):","if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return","utils.interp_conserve if self.wave is None: print('Filter not defined.') return False if source_flux is","put in the EAZY filter file \"\"\" header = '{0} {1} lambda_c= {2:.4e}","grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None:","integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \"\"\" Filter equivalent","return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect","case: search_string = search_string.upper() matched = [] for i in range(len(self.filters)): filt_name =","flux density, cgs # Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x,","mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \"\"\" Compute AB-Vega conversion","self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def","Table from . import utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(),","the filter names. \"\"\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name))","= 1. # pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave)","1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return","(self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99 =","{2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close()","if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \"\"\" Search filter names","astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp =","rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth / self.throughput.max() return rect @property def","bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is None:","VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass","np import os from astropy.table import Table from . import utils __all__ =","Return a string that can be put in the EAZY filter file \"\"\"","= f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\"","# Vega spectrum, units of f-lambda flux density, cgs # Interpolate to wavelength","i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest',","@property def rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth /","0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a","line: if len(wave) > 0: # Make filter from lines already read in","np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:])",": array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp","< 910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV,","np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans","new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput","{3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines","bandpass \"\"\" self.name = name self.wave = wave self.throughput = throughput self.Aflux =","VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass object Parameters","= integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property","os from astropy.table import Table from . import utils __all__ = [\"FilterDefinition\", \"FilterFile\",","throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025,","'\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY filter file.","COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber","arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\" self.name =","self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def","# Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x","file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename","return pivot @property def equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput,","def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a string that can be put","filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1,","plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None)","= 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x)","[] header = '' for line in lines: if 'lambda_c' in line: if","information to a filter file. \"\"\" fp = open(file,'w') for filter in self.filters:","np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot","header = ' '.join(line.split()[1:]) wave = [] trans = [] else: lspl =","str Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array","as np import os from astropy.table import Table from . import utils __all__","else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) /","N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1,","len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t)","Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get the MW extinction correction within the filter.","\"\"\" from astropy.constants import c import astropy.units as u try: import grizli.utils_c interp","not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def","filter bandpass \"\"\" self.name = name self.wave = wave self.throughput = throughput self.Aflux","get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\" import astropy.units as u f99 =","if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='#","Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth / self.throughput.max() return rect @property","ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is None: file_path =","if 'lambda_c' in line: if len(wave) > 0: # Make filter from lines","ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1])","__getitem__(self, i1): \"\"\" Return unit-indexed filter, e.g., 161 = 2mass-j \"\"\" return self.filters[i1-1]","[] wave = [] trans = [] header = '' for line in","\"\"\" self.name = name self.wave = wave self.throughput = throughput self.Aflux = 1.","\"\"\" Number of filters in the list \"\"\" return len(self.filters) def __getitem__(self, i1):","names for ``search_string``. If ``case`` is True, then match case. \"\"\" import re","matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt -","interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4):","object Parameters ---------- name : str Label name wave : array Wavelength array,","arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of","num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self):","= interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() >","not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm =","\"\"\" Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt","# new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans))","self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None:","equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\" Filter rectangular","verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \"\"\" Search filter names for","= '' for line in lines: if 'lambda_c' in line: if len(wave) >","filter names. \"\"\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else:","self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self):","one # new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave)","self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt","AB-Vega conversion \"\"\" from astropy.constants import c import astropy.units as u try: import","True, then match case. \"\"\" import re if not case: search_string = search_string.upper()","then match case. \"\"\" import re if not case: search_string = search_string.upper() matched","\"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c]","= os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename =","\"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth / self.throughput.max() return rect","self.name = bp.name self.norm = 1. if self.throughput is not None: self.norm =","density, cgs # Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'],","fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote","\"\"\" Bandpass object Parameters ---------- name : str Label name wave : array","optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\" self.name = name self.wave = wave","EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV,","__all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for","self.wave, self.throughput, left=0, right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2","= utils.interp_conserve if self.wave is None: print('Filter not defined.') return False if source_flux","0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam","self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in","= [] wave = [] trans = [] header = '' for line","def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num = integrator(self.wave,","file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot","throughput self.Aflux = 1. # pysynphot Bandpass if bp is not None: self.wave","path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file)","# new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header","AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass,","wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if bp is","fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \"\"\" Search filter","self.wave = wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if","return 1./delta @property def ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\" from astropy.constants import","extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get the MW extinction correction within","self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t)","{wave:.5e} {thru:.5e}'): \"\"\" Return a string that can be put in the EAZY","= fp.readlines() self.filename = file_path filters = [] wave = [] trans =","for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e}","\"\"\" return len(self.filters) def __getitem__(self, i1): \"\"\" Return unit-indexed filter, e.g., 161 =","astropy.table import Table from . import utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE","pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput)","= full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs # Interpolate","http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(),","' '.join(line.split()[1:]) wave = [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1])","np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \"\"\"","c import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError:","ImportError: interp = utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x","VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB",": array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass","pivot @property def equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave)","new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter =","FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY filter file. .. plot::","None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path,","string_list def write(self, file='xxx.res', verbose=True): \"\"\" Dump the filter information to a filter","in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None):","\"\"\" rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \"\"\" 95%","\"\"\" return self.filters[i1-1] def names(self, verbose=True): \"\"\" Print the filter names. \"\"\" if","if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99","__init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass object Parameters ---------- name : str","source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max()","import matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp","= utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave,","range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \"\"\" Dump the filter information to","= np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = '","= FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput,","enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a",". import utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA","= np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave = []","Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None,","return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a string that can","= 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam","astropy.constants import c import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c","Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\" Filter","fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in","@property def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num =","lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() #","list \"\"\" return len(self.filters) def __getitem__(self, i1): \"\"\" Return unit-indexed filter, e.g., 161","dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def","in the list \"\"\" return len(self.filters) def __getitem__(self, i1): \"\"\" Return unit-indexed filter,","range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info',","\"\"\" import re if not case: search_string = search_string.upper() matched = [] for","np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is","units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num =","@property def ctw95(self): \"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave)","re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched)","in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read","for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None,","line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4]","def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get the MW extinction correction","open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d}","= bp.name self.norm = 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave,","def ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\" from astropy.constants import c import astropy.units","FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0])","[] trans = [] header = '' for line in lines: if 'lambda_c'","<{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \"\"\" Search filter names for ``search_string``. If","grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x,","self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is not","\"\"\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux","interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput and","as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve","pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput)","header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter","(self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda =","trans = [] header = '' for line in lines: if 'lambda_c' in","np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name =","= np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6}","full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs # Interpolate to","a source spectrum. \"\"\" import astropy.units as u try: import grizli.utils_c interp =","FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \"\"\" Number of","for ``search_string``. If ``case`` is True, then match case. \"\"\" import re if","= wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if bp","wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput =","None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__()","from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig,","verbose=True): \"\"\" Print the filter names. \"\"\" if verbose: for i in range(len(self.filters)):","= ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res',","def write(self, file='xxx.res', verbose=True): \"\"\" Dump the filter information to a filter file.","= np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters","0: # Make filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave),","= len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w,","for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile: def __init__(self,","= open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)):","= interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same units absp","self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def","self.norm = 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def","wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \"\"\" Number of filters","\"\"\" Print the filter names. \"\"\" if verbose: for i in range(len(self.filters)): print('{0:5d}","u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if","VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB =","/ self.throughput.max() return rect @property def ctw95(self): \"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords","source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min()","os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path","= header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize","f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get","[] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter =","spectrum, units of f-lambda flux density, cgs # Interpolate to wavelength grid, no","def search(self, search_string, case=False, verbose=True): \"\"\" Search filter names for ``search_string``. If ``case``","-2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz num",":include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters))","return -2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator = np.trapz","file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path filters","bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms')","spectrum. \"\"\" import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except","cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 =","filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile","fp: lines = fp.readlines() self.filename = file_path filters = [] wave = []","np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect =","den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \"\"\"","numpy as np import os from astropy.table import Table from . import utils","\"\"\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\"","over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x)","wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name = header","fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \"\"\" Search","If ``case`` is True, then match case. \"\"\" import re if not case:","spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units","{2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega,","in the EAZY filter file \"\"\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega=","utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if","MW extinction correction within the filter. Optionally supply a source spectrum. \"\"\" import","utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True,","try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave","throughput=None, bp=None): \"\"\" Bandpass object Parameters ---------- name : str Label name wave","not defined.') return False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux","FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput","for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name)","print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))]","not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return","bp=None): \"\"\" Bandpass object Parameters ---------- name : str Label name wave :","self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get the","Bandpass object Parameters ---------- name : str Label name wave : array Wavelength","integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def","= integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \"\"\" Filter","self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))]","VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs","= open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False,","except ImportError: interp = utils.interp_conserve # Union of throughput and Vega spectrum arrays","of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)]","filter file \"\"\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N","= grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput and Vega","{thru:.5e}'): \"\"\" Return a string that can be put in the EAZY filter","\"\"\" Extinction factor \"\"\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda","fp.readlines() self.filename = file_path filters = [] wave = [] trans = []","return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\"","left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0.","return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\" import astropy.units as","as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def","# Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den =","`astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot`","rect @property def ctw95(self): \"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl =","no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave,","lines: if 'lambda_c' in line: if len(wave) > 0: # Make filter from","if len(wave) > 0: # Make filter from lines already read in new_filter","\"\"\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave)","the filter information to a filter file. \"\"\" fp = open(file,'w') for filter","vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0,","Return unit-indexed filter, e.g., 161 = 2mass-j \"\"\" return self.filters[i1-1] def names(self, verbose=True):","full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs #","line in lines: if 'lambda_c' in line: if len(wave) > 0: # Make","correction within the filter. Optionally supply a source spectrum. \"\"\" import astropy.units as","if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name =","ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file)","full_x) return -2.5*np.log10(num/den) @property def pivot(self): \"\"\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" integrator =","for i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper()","Number of filters in the list \"\"\" return len(self.filters) def __getitem__(self, i1): \"\"\"","filters = [] wave = [] trans = [] header = '' for","throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \"\"\" Number of filters in","self.wave) @property def rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" rect = self.equivwidth","source_lam=None, source_flux=None): \"\"\" Get the MW extinction correction within the filter. Optionally supply","width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\" Filter rectangular width","open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path filters = []","['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True):","http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \"\"\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html","= header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header,","file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines =","as fp: lines = fp.readlines() self.filename = file_path filters = [] wave =","u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve #","already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header #","self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1):","np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is not None: self.norm","= self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() <","Vega spectrum, units of f-lambda flux density, cgs # Interpolate to wavelength grid,","a filter file. \"\"\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave),","EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import","FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4))","to a filter file. \"\"\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d}","is None: print('Filter not defined.') return False if source_flux is None: source_flux =","= filters @property def NFILT(self): \"\"\" Number of filters in the list \"\"\"","EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \"\"\" Get the MW extinction correction within the","plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp)","w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines +=","string that can be put in the EAZY filter file \"\"\" header =","lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header","``search_string``. If ``case`` is True, then match case. \"\"\" import re if not","file='xxx.res', verbose=True): \"\"\" Dump the filter information to a filter file. \"\"\" fp","new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) #","else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition()",": str Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput :","def __init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass object Parameters ---------- name :","{1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list)","header = '' for line in lines: if 'lambda_c' in line: if len(wave)","t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \"\"\"","= Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self,","name : str Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput","grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput","filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self,","= filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name))","np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\" Return a string that can be","the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return","filter. Optionally supply a source spectrum. \"\"\" import astropy.units as u try: import","Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name","source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda =","header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave),","ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not defined.') return False","import re if not case: search_string = search_string.upper() matched = [] for i","= 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda),","delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return","lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot,","with open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path filters =","print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid()","array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass`","in lines: if 'lambda_c' in line: if len(wave) > 0: # Make filter","FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans)","i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp","normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\" self.name = name","= f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta)","__init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name =","in line: if len(wave) > 0: # Make filter from lines already read","file. \"\"\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for","= throughput self.Aflux = 1. # pysynphot Bandpass if bp is not None:","thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile: def","Parameters ---------- name : str Label name wave : array Wavelength array, in","trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name = header #","\"\"\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot","file_path filters = [] wave = [] trans = [] header = ''","= np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput","open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True):","self.wave is None: print('Filter not defined.') return False if source_flux is None: source_flux","self.filename = file_path filters = [] wave = [] trans = [] header","\"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames:","\"\"\" Dump the filter information to a filter file. \"\"\" fp = open(file,'w')","# new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave","print('Filter not defined.') return False if source_flux is None: source_flux = self.throughput*0.+1 else:","= [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter","import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux =","import utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA =","import c import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except","wave = [] trans = [] header = '' for line in lines:","extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput,","flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property","= FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \"\"\" Number","new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave)","def equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property def","equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self):","[] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one","self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \"\"\" Extinction factor \"\"\" import astropy.units as u","fig.tight_layout(pad=0.5) \"\"\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path","res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave,","# last one # new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave","matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name if not case:","wave=None, throughput=None, bp=None): \"\"\" Bandpass object Parameters ---------- name : str Label name","import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of","if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if","Optionally supply a source spectrum. \"\"\" import astropy.units as u try: import grizli.utils_c","interp = utils.interp_conserve if self.wave is None: print('Filter not defined.') return False if","print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \"\"\" Search filter names for ``search_string``.","and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega","write(self, file='xxx.res', verbose=True): \"\"\" Dump the filter information to a filter file. \"\"\"","import Table from . import utils __all__ = [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE =","Extinction factor \"\"\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda =","self.throughput, left=0, right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 #","RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0])","self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber = int(line.split('Filter #')[1].split(',')[0])","match case. \"\"\" import re if not case: search_string = search_string.upper() matched =","self.name = name self.wave = wave self.throughput = throughput self.Aflux = 1. #","import os from astropy.table import Table from . import utils __all__ = [\"FilterDefinition\",","units of f-lambda flux density, cgs # Interpolate to wavelength grid, no extrapolation","rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \"\"\" 95% cumulative","wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization","as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205]","= np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den)","the EAZY filter file \"\"\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f}","name self.wave = wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass","#20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber =","right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over","source spectrum. \"\"\" import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c","is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm","self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910)","name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass object Parameters ---------- name : str Label","np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return","filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list =","+= [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines)","from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name =","matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp =","import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp","np.sqrt(num/den) return pivot @property def equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return","the filter. Optionally supply a source spectrum. \"\"\" import astropy.units as u try:","filter, e.g., 161 = 2mass-j \"\"\" return self.filters[i1-1] def names(self, verbose=True): \"\"\" Print","@property def ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\" from astropy.constants import c import","supply a source spectrum. \"\"\" import astropy.units as u try: import grizli.utils_c interp","fp.write('{0:6d} {1}\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close()","a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters","case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d}","bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den)","Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0)","= np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \"\"\"","not case: search_string = search_string.upper() matched = [] for i in range(len(self.filters)): filt_name","= ' '.join(line.split()[1:]) wave = [] trans = [] else: lspl = np.cast[float](line.split())","filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is","= grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not","len(wave) > 0: # Make filter from lines already read in new_filter =","def __getitem__(self, i1): \"\"\" Return unit-indexed filter, e.g., 161 = 2mass-j \"\"\" return","right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same","factor \"\"\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave)","# AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the","'{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N,","\"\"\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list =","Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional,","\"\"\" Return a string that can be put in the EAZY filter file","filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i)","range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name)","self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \"\"\" Dump","\"\"\" Compute AB-Vega conversion \"\"\" from astropy.constants import c import astropy.units as u","return False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave,","# Make filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans))","print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength,","VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\"","file \"\"\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N =","0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave)","\"\"\" Get the MW extinction correction within the filter. Optionally supply a source","a string that can be put in the EAZY filter file \"\"\" header","= np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name","the list \"\"\" return len(self.filters) def __getitem__(self, i1): \"\"\" Return unit-indexed filter, e.g.,","interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) #","that can be put in the EAZY filter file \"\"\" header = '{0}","filters in the list \"\"\" return len(self.filters) def __getitem__(self, i1): \"\"\" Return unit-indexed","Rv=3.1): \"\"\" Extinction factor \"\"\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv)","FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \"\"\" Bandpass object Parameters ---------- name","wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\n'.join(lines) class FilterFile:","array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \"\"\"","verbose=True): \"\"\" Search filter names for ``search_string``. If ``case`` is True, then match","1./delta @property def ABVega(self): \"\"\" Compute AB-Vega conversion \"\"\" from astropy.constants import c","defined.') return False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux =","Make filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) #","= np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def","ctw95(self): \"\"\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \"\"\" dl = np.diff(self.wave) filt =","= self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not","= [] trans = [] header = '' for line in lines: if","num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot","= [\"FilterDefinition\", \"FilterFile\", \"ParamFilter\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c","Get the MW extinction correction within the filter. Optionally supply a source spectrum.","{1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return",".. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res =","source_flux=None): \"\"\" Get the MW extinction correction within the filter. Optionally supply a","name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary","lines = fp.readlines() self.filename = file_path filters = [] wave = [] trans","= file_path filters = [] wave = [] trans = [] header =","Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \"\"\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'),","verbose=True): \"\"\" Dump the filter information to a filter file. \"\"\" fp =","be put in the EAZY filter file \"\"\" header = '{0} {1} lambda_c=","re if not case: search_string = search_string.upper() matched = [] for i in","> 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave)","thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same units","filter header = ' '.join(line.split()[1:]) wave = [] trans = [] else: lspl","@property def equivwidth(self): \"\"\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \"\"\" return np.trapz(self.throughput, self.wave) @property","def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name","header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines","__init__(self, file='FILTER.RES.latest', path='./'): \"\"\" Read a EAZY filter file. .. plot:: :include-source: import"],"string":"[\n \"self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None):\",\n \"= 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self):\",\n \"ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'):\",\n \"\\\"\\\"\\\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path =\",\n \"else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines()\",\n \"http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \\\"\\\"\\\"\",\n \"row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string that can be put in the\",\n \"Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans = [] else:\",\n \"right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0. else:\",\n \"in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object\",\n \"Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x =\",\n \"last one # new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave =\",\n \"else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) |\",\n \"case=False, verbose=True): \\\"\\\"\\\" Search filter names for ``search_string``. If ``case`` is True, then\",\n \"i1): \\\"\\\"\\\" Return unit-indexed filter, e.g., 161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def\",\n \"self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if bp is not\",\n \"\\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den\",\n \"filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class\",\n \"# new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) #\",\n \"np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c\",\n \"= [] header = '' for line in lines: if 'lambda_c' in line:\",\n \"= FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput =\",\n \"`pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\" self.name = name self.wave = wave self.throughput\",\n \"Search filter names for ``search_string``. If ``case`` is True, then match case. \\\"\\\"\\\"\",\n \"= 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW\",\n \"161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print the filter\",\n \"search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter names for ``search_string``. If ``case`` is True,\",\n \"# Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans = []\",\n \"f-lambda flux density, cgs # Interpolate to wavelength grid, no extrapolation vega_full =\",\n \"is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self):\",\n \"= [] for i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name\",\n \"Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta =\",\n \"is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with\",\n \"grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput and Vega spectrum\",\n \"rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth / self.throughput.max() return\",\n \"'' for line in lines: if 'lambda_c' in line: if len(wave) > 0:\",\n \"file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r')\",\n \"is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0)\",\n \"910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv)\",\n \"lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for\",\n \"conversion \\\"\\\"\\\" from astropy.constants import c import astropy.units as u try: import grizli.utils_c\",\n \"{1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78:\",\n \"astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda)\",\n \"= plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if\",\n \"object `pysynphot` filter bandpass \\\"\\\"\\\" self.name = name self.wave = wave self.throughput =\",\n \"[] for i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name =\",\n \"integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot =\",\n \"self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \\\"\\\"\\\" Filter equivalent width\",\n \"= FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) #\",\n \"interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter\",\n \"new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header =\",\n \"def names(self, verbose=True): \\\"\\\"\\\" Print the filter names. \\\"\\\"\\\" if verbose: for i\",\n \"if self.wave is None: print('Filter not defined.') return False if source_flux is None:\",\n \"ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else:\",\n \"'r') as fp: lines = fp.readlines() self.filename = file_path filters = [] wave\",\n \"<filename>eazy/filters.py import numpy as np import os from astropy.table import Table from .\",\n \"def NFILT(self): \\\"\\\"\\\" Number of filters in the list \\\"\\\"\\\" return len(self.filters) def\",\n \"ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\"\",\n \"filters.append(new_filter) self.filters = filters @property def NFILT(self): \\\"\\\"\\\" Number of filters in the\",\n \"if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__()\",\n \"class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c =\",\n \"return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor\",\n \"in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter information\",\n \"= [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i,\",\n \"if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d}\",\n \"Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\" self.name\",\n \"interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same units absp =\",\n \"= np.sqrt(num/den) return pivot @property def equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\"\",\n \"Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full,\",\n \"path='./'): \\\"\\\"\\\" Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as\",\n \"source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0,\",\n \"@property def NFILT(self): \\\"\\\"\\\" Number of filters in the list \\\"\\\"\\\" return len(self.filters)\",\n \"np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator\",\n \"from astropy.table import Table from . import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"]\",\n \"c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None,\",\n \"Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum,\",\n \"filter file. \\\"\\\"\\\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name))\",\n \"width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self):\",\n \"utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']])\",\n \"search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter names for ``search_string``. If ``case`` is\",\n \"return rect @property def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl\",\n \"wave = [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) #\",\n \"den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html\",\n \"self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \\\"\\\"\\\" Compute\",\n \"new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave =\",\n \"utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE)\",\n \"left=0, right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate\",\n \"string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file))\",\n \"array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp :\",\n \"the MW extinction correction within the filter. Optionally supply a source spectrum. \\\"\\\"\\\"\",\n \"if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path,\",\n \"throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans)\",\n \"full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot wavelength\",\n \"[\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in\",\n \"is True, then match case. \\\"\\\"\\\" import re if not case: search_string =\",\n \"= search_string.upper() matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name if\",\n \"self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print the filter names. \\\"\\\"\\\" if verbose: for\",\n \": optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\" self.name = name self.wave =\",\n \"(w, t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'):\",\n \"bp.name self.norm = 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave)\",\n \"\\\"\\\"\\\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i\",\n \"for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string that can be put in\",\n \"return '\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY filter\",\n \"{1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\"\",\n \"bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name\",\n \"for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False)\",\n \"\\\"\\\"\\\" Return unit-indexed filter, e.g., 161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self,\",\n \"= self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def\",\n \"self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self,\",\n \"= utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1,\",\n \"Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None,\",\n \"6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta\",\n \"= res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput')\",\n \"of f-lambda flux density, cgs # Interpolate to wavelength grid, no extrapolation vega_full\",\n \"= np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is not None:\",\n \"return string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter information to a\",\n \"\\\"\\\"\\\" Search filter names for ``search_string``. If ``case`` is True, then match case.\",\n \"filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1,\",\n \"print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20,\",\n \"full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda\",\n \"verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1,\",\n \"{1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0],\",\n \"# new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter)\",\n \"if not case: search_string = search_string.upper() matched = [] for i in range(len(self.filters)):\",\n \"= name self.wave = wave self.throughput = throughput self.Aflux = 1. # pysynphot\",\n \"lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return\",\n \"/ np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self):\",\n \"from . import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits')\",\n \"self.throughput))] return '\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY\",\n \"filter information to a filter file. \\\"\\\"\\\" fp = open(file,'w') for filter in\",\n \"None: print('Filter not defined.') return False if source_flux is None: source_flux = self.throughput*0.+1\",\n \"= os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as\",\n \"label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None: file_path\",\n \"unit-indexed filter, e.g., 161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\"\",\n \"self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width\",\n \"search_string.upper() matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name if not\",\n \"{1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list\",\n \"ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants import c import astropy.units as\",\n \"``case`` is True, then match case. \\\"\\\"\\\" import re if not case: search_string\",\n \"i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for\",\n \"i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if\",\n \"import numpy as np import os from astropy.table import Table from . import\",\n \"fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w')\",\n \"from astropy.constants import c import astropy.units as u try: import grizli.utils_c interp =\",\n \"in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string,\",\n \"None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if\",\n \"False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam,\",\n \"\\\"\\\"\\\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:])\",\n \"= 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print the filter names.\",\n \"def rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth / self.throughput.max()\",\n \"new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter)\",\n \"return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'):\",\n \"fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5)\",\n \"eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax\",\n \"else: return 1./delta @property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants\",\n \"3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den\",\n \"new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property\",\n \"= interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0)\",\n \"throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter\",\n \"bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\" self.name = name self.wave\",\n \"= np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\"\",\n \"Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den =\",\n \"np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e}\",\n \"http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave)\",\n \"[row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class\",\n \"source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda\",\n \"of filters in the list \\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return\",\n \"to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full\",\n \"in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i],\",\n \"np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0,\",\n \"throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] #\",\n \"cgs # Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'],\",\n \"| (self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda\",\n \"= os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float)\",\n \"case. \\\"\\\"\\\" import re if not case: search_string = search_string.upper() matched = []\",\n \"# new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter\",\n \"filters @property def NFILT(self): \\\"\\\"\\\" Number of filters in the list \\\"\\\"\\\" return\",\n \"in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave =\",\n \"---------- name : str Label name wave : array Wavelength array, in `astropy.units.Angstrom`.\",\n \"\\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\"\",\n \"Print the filter names. \\\"\\\"\\\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1,\",\n \"wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full =\",\n \"= utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave)\",\n \"try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union\",\n \"for line in lines: if 'lambda_c' in line: if len(wave) > 0: #\",\n \"len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter, e.g., 161 = 2mass-j \\\"\\\"\\\"\",\n \"\\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl)\",\n \"wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave,\",\n \"'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition:\",\n \"[header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w,\",\n \"except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not defined.') return\",\n \"= np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0]\",\n \"self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if\",\n \"def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import\",\n \"'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines\",\n \"for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the\",\n \"Dump the filter information to a filter file. \\\"\\\"\\\" fp = open(file,'w') for\",\n \"within the filter. Optionally supply a source spectrum. \\\"\\\"\\\" import astropy.units as u\",\n \"self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string that\",\n \"self.throughput.max() return rect @property def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\"\",\n \"np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self):\",\n \"import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax =\",\n \"np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return\",\n \"range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in\",\n \"2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from\",\n \"= np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux\",\n \"None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self,\",\n \"'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\"\",\n \"def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY filter file. .. plot:: :include-source:\",\n \"# pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput =\",\n \"os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class\",\n \"fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string,\",\n \"'lambda_c' in line: if len(wave) > 0: # Make filter from lines already\",\n \"10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW extinction\",\n \"= '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines =\",\n \"not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose:\",\n \"grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not defined.')\",\n \"return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print the filter names. \\\"\\\"\\\" if verbose:\",\n \"f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV,\",\n \"1. # pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput\",\n \"trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one #\",\n \"Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return\",\n \"filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string\",\n \"2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print the filter names. \\\"\\\"\\\"\",\n \"self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i]))\",\n \"# Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0,\",\n \"= self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput\",\n \"can be put in the EAZY filter file \\\"\\\"\\\" header = '{0} {1}\",\n \"AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)]\",\n \"EAZY filter file \\\"\\\"\\\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}'\",\n \"dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return\",\n \"res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend()\",\n \"= np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\"\",\n \"pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html\",\n \"class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY filter file. ..\",\n \"interp = utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x =\",\n \"None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1.\",\n \"search_string = search_string.upper() matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name\",\n \"mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW extinction correction within the filter. Optionally\",\n \"f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux,\",\n \"ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is\",\n \"self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave,\",\n \"= VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object\",\n \"names. \\\"\\\"\\\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list\",\n \"i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter\",\n \"string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self,\",\n \"= np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self,\",\n \"np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density,\",\n \"class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters ----------\",\n \"`pysynphot` filter bandpass \\\"\\\"\\\" self.name = name self.wave = wave self.throughput = throughput\",\n \"= np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta\",\n \"'.join(line.split()[1:]) wave = [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2])\",\n \"read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave\",\n \"filter names for ``search_string``. If ``case`` is True, then match case. \\\"\\\"\\\" import\",\n \"same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num\",\n \"np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters =\",\n \"__repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction\",\n \"Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput,\",\n \"np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property\",\n \"if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega\",\n \"Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag:\",\n \"import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is\",\n \"absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full,\",\n \"self.filters = filters @property def NFILT(self): \\\"\\\"\\\" Number of filters in the list\",\n \"plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path\",\n \"filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if\",\n \"self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\"\",\n \"self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list\",\n \"filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans =\",\n \"VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def\",\n \"u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self,\",\n \"# new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters\",\n \"if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux,\",\n \"NFILT(self): \\\"\\\"\\\" Number of filters in the list \\\"\\\"\\\" return len(self.filters) def __getitem__(self,\",\n \"in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i\",\n \"def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt\",\n \"filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name\",\n \"def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\"\",\n \"os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp:\",\n \"verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter\",\n \"\\\"\\\"\\\" import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError:\",\n \"def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units as u f99\",\n \"95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95\",\n \"left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0,\",\n \"= [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last\",\n \"new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \\\"\\\"\\\"\",\n \"Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants import c import astropy.units as u try:\",\n \"- lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber =\",\n \"VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] =\",\n \"filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose:\",\n \"width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975],\",\n \"else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def\",\n \"lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber = int(line.split('Filter\",\n \"> 0: # Make filter from lines already read in new_filter = FilterDefinition(name=header,\",\n \"ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters',\",\n \"self.Aflux = 1. # pysynphot Bandpass if bp is not None: self.wave =\",\n \"extinction correction within the filter. Optionally supply a source spectrum. \\\"\\\"\\\" import astropy.units\",\n \"Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from\",\n \"return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter, e.g., 161 = 2mass-j\",\n \"__str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units\",\n \"Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber\",\n \"np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def\",\n \"file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res\",\n \"names(self, verbose=True): \\\"\\\"\\\" Print the filter names. \\\"\\\"\\\" if verbose: for i in\",\n \"in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp =\",\n \"f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else:\",\n \"e.g., 161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print the\",\n \"is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition):\",\n \"if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return\",\n \"utils.interp_conserve if self.wave is None: print('Filter not defined.') return False if source_flux is\",\n \"put in the EAZY filter file \\\"\\\"\\\" header = '{0} {1} lambda_c= {2:.4e}\",\n \"grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None:\",\n \"integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \\\"\\\"\\\" Filter equivalent\",\n \"return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect\",\n \"case: search_string = search_string.upper() matched = [] for i in range(len(self.filters)): filt_name =\",\n \"flux density, cgs # Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x,\",\n \"mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion\",\n \"self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def\",\n \"Table from . import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(),\",\n \"the filter names. \\\"\\\"\\\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name))\",\n \"= 1. # pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave)\",\n \"1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return\",\n \"(self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99 =\",\n \"{2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close()\",\n \"if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter names\",\n \"astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp =\",\n \"rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth / self.throughput.max() return rect @property def\",\n \"bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None:\",\n \"VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass\",\n \"np import os from astropy.table import Table from . import utils __all__ =\",\n \"Return a string that can be put in the EAZY filter file \\\"\\\"\\\"\",\n \"= f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\"\",\n \"# Vega spectrum, units of f-lambda flux density, cgs # Interpolate to wavelength\",\n \"i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest',\",\n \"@property def rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth /\",\n \"0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a\",\n \"line: if len(wave) > 0: # Make filter from lines already read in\",\n \"np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:])\",\n \": array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp\",\n \"< 910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV,\",\n \"np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave = [] trans\",\n \"new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput\",\n \"{3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines\",\n \"bandpass \\\"\\\"\\\" self.name = name self.wave = wave self.throughput = throughput self.Aflux =\",\n \"VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters\",\n \"= integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property\",\n \"os from astropy.table import Table from . import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\",\",\n \"throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025,\",\n \"'\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY filter file.\",\n \"COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber\",\n \"arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\" self.name =\",\n \"self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def\",\n \"# Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x\",\n \"file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename\",\n \"return pivot @property def equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput,\",\n \"def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string that can be put\",\n \"filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1,\",\n \"plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None)\",\n \"= 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x)\",\n \"[] header = '' for line in lines: if 'lambda_c' in line: if\",\n \"information to a filter file. \\\"\\\"\\\" fp = open(file,'w') for filter in self.filters:\",\n \"np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot\",\n \"header = ' '.join(line.split()[1:]) wave = [] trans = [] else: lspl =\",\n \"str Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array\",\n \"as np import os from astropy.table import Table from . import utils __all__\",\n \"else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) /\",\n \"N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1,\",\n \"len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t)\",\n \"Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW extinction correction within the filter.\",\n \"\\\"\\\"\\\" from astropy.constants import c import astropy.units as u try: import grizli.utils_c interp\",\n \"not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def\",\n \"filter bandpass \\\"\\\"\\\" self.name = name self.wave = wave self.throughput = throughput self.Aflux\",\n \"get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units as u f99 =\",\n \"if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='#\",\n \"Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth / self.throughput.max() return rect @property\",\n \"ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None: file_path =\",\n \"if 'lambda_c' in line: if len(wave) > 0: # Make filter from lines\",\n \"ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1])\",\n \"__getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter, e.g., 161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1]\",\n \"[] wave = [] trans = [] header = '' for line in\",\n \"\\\"\\\"\\\" self.name = name self.wave = wave self.throughput = throughput self.Aflux = 1.\",\n \"\\\"\\\"\\\" Number of filters in the list \\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1):\",\n \"names for ``search_string``. If ``case`` is True, then match case. \\\"\\\"\\\" import re\",\n \"matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt -\",\n \"interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4):\",\n \"object Parameters ---------- name : str Label name wave : array Wavelength array,\",\n \"arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of\",\n \"num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def pivot(self):\",\n \"= interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() >\",\n \"not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm =\",\n \"\\\"\\\"\\\" Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt\",\n \"# new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans))\",\n \"self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None:\",\n \"equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter rectangular\",\n \"verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter names for\",\n \"= '' for line in lines: if 'lambda_c' in line: if len(wave) >\",\n \"filter names. \\\"\\\"\\\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else:\",\n \"self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self):\",\n \"one # new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave)\",\n \"self.filters[i].name)) matched.append(i) return np.array(matched) class ParamFilter(FilterDefinition): def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt\",\n \"AB-Vega conversion \\\"\\\"\\\" from astropy.constants import c import astropy.units as u try: import\",\n \"True, then match case. \\\"\\\"\\\" import re if not case: search_string = search_string.upper()\",\n \"then match case. \\\"\\\"\\\" import re if not case: search_string = search_string.upper() matched\",\n \"\\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c]\",\n \"= os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename =\",\n \"\\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth / self.throughput.max() return rect\",\n \"self.name = bp.name self.norm = 1. if self.throughput is not None: self.norm =\",\n \"density, cgs # Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'],\",\n \"fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote\",\n \"\\\"\\\"\\\" Bandpass object Parameters ---------- name : str Label name wave : array\",\n \"optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\" self.name = name self.wave = wave\",\n \"EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV,\",\n \"__all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for\",\n \"self.wave, self.throughput, left=0, right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2\",\n \"= utils.interp_conserve if self.wave is None: print('Filter not defined.') return False if source_flux\",\n \"0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam\",\n \"self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in\",\n \"= [] wave = [] trans = [] header = '' for line\",\n \"def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num = integrator(self.wave,\",\n \"file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot\",\n \"throughput self.Aflux = 1. # pysynphot Bandpass if bp is not None: self.wave\",\n \"path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file)\",\n \"# new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header\",\n \"AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass,\",\n \"wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if bp is\",\n \"fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter\",\n \"self.wave = wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if\",\n \"return 1./delta @property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants import\",\n \"extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW extinction correction within\",\n \"self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t)\",\n \"{wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string that can be put in the EAZY\",\n \"= fp.readlines() self.filename = file_path filters = [] wave = [] trans =\",\n \"for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e}\",\n \"\\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter, e.g., 161 =\",\n \"astropy.table import Table from . import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE\",\n \"pysynphot Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput)\",\n \"= full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs # Interpolate\",\n \"http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(),\",\n \"' '.join(line.split()[1:]) wave = [] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1])\",\n \"np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \\\"\\\"\\\"\",\n \"c import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError:\",\n \"ImportError: interp = utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x\",\n \"VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB\",\n \": array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass\",\n \"pivot @property def equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave)\",\n \"new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter =\",\n \"FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY filter file. .. plot::\",\n \"None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path = os.path.join(path, file) with open(file_path,\",\n \"string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter information to a filter\",\n \"in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None):\",\n \"\\\"\\\"\\\" rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \\\"\\\"\\\" 95%\",\n \"\\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print the filter names. \\\"\\\"\\\" if\",\n \"if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0. else: f99\",\n \"__init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters ---------- name : str\",\n \"source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max()\",\n \"import matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp\",\n \"= utils.interp_conserve # Union of throughput and Vega spectrum arrays full_x = np.hstack([self.wave,\",\n \"range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter information to\",\n \"= np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = '\",\n \"= FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput,\",\n \"enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a\",\n \". import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA\",\n \"= np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave = []\",\n \"Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None,\",\n \"return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string that can\",\n \"= 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam\",\n \"astropy.constants import c import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c\",\n \"Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter\",\n \"fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in\",\n \"@property def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num =\",\n \"lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() #\",\n \"list \\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter, e.g., 161\",\n \"dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property def\",\n \"in the list \\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter,\",\n \"range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info',\",\n \"\\\"\\\"\\\" import re if not case: search_string = search_string.upper() matched = [] for\",\n \"np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is\",\n \"units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the bandpass, flam dlam num =\",\n \"@property def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave)\",\n \"re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i) return np.array(matched)\",\n \"in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read\",\n \"for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None,\",\n \"line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4]\",\n \"def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW extinction correction\",\n \"open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d}\",\n \"= bp.name self.norm = 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave,\",\n \"def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants import c import astropy.units\",\n \"FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0])\",\n \"[] trans = [] header = '' for line in lines: if 'lambda_c'\",\n \"<{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter names for ``search_string``. If\",\n \"grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x,\",\n \"self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is not\",\n \"\\\"\\\"\\\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux\",\n \"interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput and\",\n \"as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve\",\n \"pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput)\",\n \"header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter\",\n \"(self.wave.max() > 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda =\",\n \"trans = [] header = '' for line in lines: if 'lambda_c' in\",\n \"np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name =\",\n \"= np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6}\",\n \"full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs # Interpolate to\",\n \"a source spectrum. \\\"\\\"\\\" import astropy.units as u try: import grizli.utils_c interp =\",\n \"FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \\\"\\\"\\\" Number of\",\n \"for ``search_string``. If ``case`` is True, then match case. \\\"\\\"\\\" import re if\",\n \"= wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass if bp\",\n \"wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput =\",\n \"None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__()\",\n \"from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig,\",\n \"verbose=True): \\\"\\\"\\\" Print the filter names. \\\"\\\"\\\" if verbose: for i in range(len(self.filters)):\",\n \"= ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res',\",\n \"def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter information to a filter file.\",\n \"= np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters\",\n \"0: # Make filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave),\",\n \"= len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines += [row_str.format(i=i+1, wave=w,\",\n \"for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile: def __init__(self,\",\n \"= open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)):\",\n \"= interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same units absp\",\n \"self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def\",\n \"self.norm = 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def\",\n \"wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \\\"\\\"\\\" Number of filters\",\n \"\\\"\\\"\\\" Print the filter names. \\\"\\\"\\\" if verbose: for i in range(len(self.filters)): print('{0:5d}\",\n \"u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if\",\n \"VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB =\",\n \"/ self.throughput.max() return rect @property def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords\",\n \"source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min()\",\n \"os.path.join(path, file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path\",\n \"= header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize\",\n \"f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get\",\n \"[] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter =\",\n \"spectrum, units of f-lambda flux density, cgs # Interpolate to wavelength grid, no\",\n \"def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter names for ``search_string``. If ``case``\",\n \"-2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz num\",\n \":include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters))\",\n \"return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator = np.trapz\",\n \"file) with open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path filters\",\n \"bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms')\",\n \"spectrum. \\\"\\\"\\\" import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except\",\n \"cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 =\",\n \"filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile\",\n \"fp: lines = fp.readlines() self.filename = file_path filters = [] wave = []\",\n \"np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect =\",\n \"den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \\\"\\\"\\\"\",\n \"numpy as np import os from astropy.table import Table from . import utils\",\n \"\\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\"\",\n \"over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x)\",\n \"wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name = header\",\n \"fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search\",\n \"If ``case`` is True, then match case. \\\"\\\"\\\" import re if not case:\",\n \"spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units\",\n \"{2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega,\",\n \"in the EAZY filter file \\\"\\\"\\\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega=\",\n \"utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if\",\n \"MW extinction correction within the filter. Optionally supply a source spectrum. \\\"\\\"\\\" import\",\n \"utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True,\",\n \"try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave\",\n \"throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters ---------- name : str Label name wave\",\n \"not defined.') return False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux\",\n \"FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput\",\n \"for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name)\",\n \"print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))]\",\n \"not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return\",\n \"bp=None): \\\"\\\"\\\" Bandpass object Parameters ---------- name : str Label name wave :\",\n \"self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the\",\n \"Bandpass object Parameters ---------- name : str Label name wave : array Wavelength\",\n \"integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def\",\n \"= integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot @property def equivwidth(self): \\\"\\\"\\\" Filter\",\n \"self.ctw95)] lines += [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))]\",\n \"VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs\",\n \"= open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False,\",\n \"except ImportError: interp = utils.interp_conserve # Union of throughput and Vega spectrum arrays\",\n \"of throughput and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)]\",\n \"filter file \\\"\\\"\\\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N\",\n \"= grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput and Vega\",\n \"{thru:.5e}'): \\\"\\\"\\\" Return a string that can be put in the EAZY filter\",\n \"\\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda\",\n \"fp.readlines() self.filename = file_path filters = [] wave = [] trans = []\",\n \"return 2.5*np.log10(delta) else: return 1./delta @property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\"\",\n \"left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda = 0.\",\n \"return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units as\",\n \"as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def\",\n \"# Integrate over the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den =\",\n \"`astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot`\",\n \"rect @property def ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl =\",\n \"no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave,\",\n \"lines: if 'lambda_c' in line: if len(wave) > 0: # Make filter from\",\n \"if len(wave) > 0: # Make filter from lines already read in new_filter\",\n \"\\\"\\\"\\\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave)\",\n \"the filter information to a filter file. \\\"\\\"\\\" fp = open(file,'w') for filter\",\n \"vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0,\",\n \"Return unit-indexed filter, e.g., 161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True):\",\n \"full_x = full_x[np.argsort(full_x)] # Vega spectrum, units of f-lambda flux density, cgs #\",\n \"line in lines: if 'lambda_c' in line: if len(wave) > 0: # Make\",\n \"correction within the filter. Optionally supply a source spectrum. \\\"\\\"\\\" import astropy.units as\",\n \"if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name =\",\n \"ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file)\",\n \"full_x) return -2.5*np.log10(num/den) @property def pivot(self): \\\"\\\"\\\" Pivot wavelength http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" integrator =\",\n \"for i in range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper()\",\n \"Number of filters in the list \\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\"\",\n \"filters = [] wave = [] trans = [] header = '' for\",\n \"throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \\\"\\\"\\\" Number of filters in\",\n \"self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" rect = self.equivwidth\",\n \"source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW extinction correction within the filter. Optionally supply\",\n \"width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter rectangular width\",\n \"open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path filters = []\",\n \"['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True):\",\n \"http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self): \\\"\\\"\\\" Filter rectangular width http://pysynphot.readthedocs.io/en/latest/properties.html\",\n \"= header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header,\",\n \"file) else: file_path = os.path.join(path, file) with open(file_path, 'r') as fp: lines =\",\n \"as fp: lines = fp.readlines() self.filename = file_path filters = [] wave =\",\n \"u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve #\",\n \"already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header #\",\n \"self.wave) def __repr__(self): return self.name.__repr__() def __str__(self): return self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1):\",\n \"np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput is not None: self.norm\",\n \"= self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() <\",\n \"Vega spectrum, units of f-lambda flux density, cgs # Interpolate to wavelength grid,\",\n \"a filter file. \\\"\\\"\\\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave),\",\n \"EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import\",\n \"FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4))\",\n \"to a filter file. \\\"\\\"\\\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d}\",\n \"is None: print('Filter not defined.') return False if source_flux is None: source_flux =\",\n \"= filters @property def NFILT(self): \\\"\\\"\\\" Number of filters in the list \\\"\\\"\\\"\",\n \"EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None): \\\"\\\"\\\" Get the MW extinction correction within the\",\n \"plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp)\",\n \"w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot, self.ABVega, self.ctw95)] lines +=\",\n \"string that can be put in the EAZY filter file \\\"\\\"\\\" header =\",\n \"lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header\",\n \"``search_string``. If ``case`` is True, then match case. \\\"\\\"\\\" import re if not\",\n \"file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter information to a filter file. \\\"\\\"\\\" fp\",\n \"new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) filters.append(new_filter) #\",\n \"else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition()\",\n \": str Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput :\",\n \"def __init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters ---------- name :\",\n \"{1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp = open(file+'.info', 'w') fp.writelines(string_list)\",\n \"header = '' for line in lines: if 'lambda_c' in line: if len(wave)\",\n \"t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile: def __init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\"\",\n \"= Table.read(VEGA_FILE) for c in VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self,\",\n \"name : str Label name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput\",\n \"grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of throughput\",\n \"filt = np.cumsum((self.wave*self.throughput)[1:]*dl) ctw95 = np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self,\",\n \"= filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name))\",\n \"np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\" Return a string that can be\",\n \"the bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return\",\n \"filter. Optionally supply a source spectrum. \\\"\\\"\\\" import astropy.units as u try: import\",\n \"Bandpass if bp is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name\",\n \"source_flux, left=0, right=0) if (self.wave.min() < 910) | (self.wave.max() > 6.e4): Alambda =\",\n \"header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave),\",\n \"ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not defined.') return False\",\n \"import re if not case: search_string = search_string.upper() matched = [] for i\",\n \"= 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda),\",\n \"delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta) else: return\",\n \"lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N, self.name.split('lambda_c')[0], self.pivot,\",\n \"with open(file_path, 'r') as fp: lines = fp.readlines() self.filename = file_path filters =\",\n \"print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength, Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid()\",\n \"array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass`\",\n \"in lines: if 'lambda_c' in line: if len(wave) > 0: # Make filter\",\n \"FilterDefinition() # new_filter.name = header # new_filter.wave = np.cast[float](wave) # new_filter.throughput = np.cast[float](trans)\",\n \"i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list = self.names(verbose=False) fp\",\n \"normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\" self.name = name\",\n \"= f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave) / np.trapz(self.throughput*source_flux, self.wave) if mag: return 2.5*np.log10(delta)\",\n \"__init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name =\",\n \"in line: if len(wave) > 0: # Make filter from lines already read\",\n \"file. \\\"\\\"\\\" fp = open(file,'w') for filter in self.filters: fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for\",\n \"= throughput self.Aflux = 1. # pysynphot Bandpass if bp is not None:\",\n \"thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile: def\",\n \"Parameters ---------- name : str Label name wave : array Wavelength array, in\",\n \"trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name = header #\",\n \"\\\"\\\"\\\" integrator = np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot\",\n \"file_path filters = [] wave = [] trans = [] header = ''\",\n \"= np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm = 1. if self.throughput\",\n \"open(file+'.info', 'w') fp.writelines(string_list) fp.close() if verbose: print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True):\",\n \"self.wave is None: print('Filter not defined.') return False if source_flux is None: source_flux\",\n \"self.filename = file_path filters = [] wave = [] trans = [] header\",\n \"\\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c in VEGA.colnames:\",\n \"\\\"\\\"\\\" Dump the filter information to a filter file. \\\"\\\"\\\" fp = open(file,'w')\",\n \"# new_filter.throughput = np.cast[float](trans) filters.append(new_filter) # Initialize filter header = ' '.join(line.split()[1:]) wave\",\n \"print('Filter not defined.') return False if source_flux is None: source_flux = self.throughput*0.+1 else:\",\n \"= [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter\",\n \"import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave) self.Aflux =\",\n \"import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA =\",\n \"import c import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except\",\n \"wave = [] trans = [] header = '' for line in lines:\",\n \"extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput,\",\n \"flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den) @property\",\n \"= FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def NFILT(self): \\\"\\\"\\\" Number\",\n \"new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name = header # new_filter.wave = np.cast[float](wave)\",\n \"def equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def\",\n \"equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property def rectwidth(self):\",\n \"[] trans = [] else: lspl = np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one\",\n \"self.name.__str__() def get_extinction(self, EBV=0, Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units as u\",\n \"fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'), 'filters', file) else: file_path\",\n \"res = FilterFile(path=None) print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave,\",\n \"# last one # new_filter = FilterDefinition() # new_filter.name = header # new_filter.wave\",\n \"matched = [] for i in range(len(self.filters)): filt_name = self.filters[i].name if not case:\",\n \"wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters ---------- name : str Label name\",\n \"import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve # Union of\",\n \"if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if\",\n \"Optionally supply a source spectrum. \\\"\\\"\\\" import astropy.units as u try: import grizli.utils_c\",\n \"interp = utils.interp_conserve if self.wave is None: print('Filter not defined.') return False if\",\n \"print('Wrote <{0}[.info]>'.format(file)) def search(self, search_string, case=False, verbose=True): \\\"\\\"\\\" Search filter names for ``search_string``.\",\n \"and Vega spectrum arrays full_x = np.hstack([self.wave, VEGA['WAVELENGTH']]) full_x = full_x[np.argsort(full_x)] # Vega\",\n \"write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump the filter information to a filter file. \\\"\\\"\\\"\",\n \"import Table from . import utils __all__ = [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE =\",\n \"Extinction factor \\\"\\\"\\\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda =\",\n \"self.throughput, left=0, right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 #\",\n \"RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0])\",\n \"self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber = int(line.split('RES#')[1].split(':')[0]) self.cnumber = int(line.split('Filter #')[1].split(',')[0])\",\n \"match case. \\\"\\\"\\\" import re if not case: search_string = search_string.upper() matched =\",\n \"self.name = name self.wave = wave self.throughput = throughput self.Aflux = 1. #\",\n \"import os from astropy.table import Table from . import utils __all__ = [\\\"FilterDefinition\\\",\",\n \"units of f-lambda flux density, cgs # Interpolate to wavelength grid, no extrapolation\",\n \"rect = self.equivwidth / self.throughput.max() return rect @property def ctw95(self): \\\"\\\"\\\" 95% cumulative\",\n \"wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization\",\n \"as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp = res[205]\",\n \"= np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den)\",\n \"the EAZY filter file \\\"\\\"\\\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f}\",\n \"name self.wave = wave self.throughput = throughput self.Aflux = 1. # pysynphot Bandpass\",\n \"#20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name = line.split()[4] self.fnumber =\",\n \"right=0) # AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over\",\n \"source spectrum. \\\"\\\"\\\" import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c\",\n \"is not None: self.wave = np.cast[np.double](bp.wave) self.throughput = np.cast[np.double](bp.throughput) self.name = bp.name self.norm\",\n \"self.throughput*0.+1 else: source_flux = interp(self.wave, source_lam, source_flux, left=0, right=0) if (self.wave.min() < 910)\",\n \"name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters ---------- name : str Label\",\n \"np.trapz num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return\",\n \"filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close() string_list =\",\n \"+= [row_str.format(i=i+1, wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines)\",\n \"from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) # new_filter.name =\",\n \"matplotlib.pyplot as plt from eazy.filters import FilterFile res = FilterFile(path=None) print(len(res.filters)) bp =\",\n \"import astropy.units as u try: import grizli.utils_c interp = grizli.utils_c.interp.interp_conserve_c except ImportError: interp\",\n \"np.sqrt(num/den) return pivot @property def equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return\",\n \"the filter. Optionally supply a source spectrum. \\\"\\\"\\\" import astropy.units as u try:\",\n \"filter, e.g., 161 = 2mass-j \\\"\\\"\\\" return self.filters[i1-1] def names(self, verbose=True): \\\"\\\"\\\" Print\",\n \"@property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants import c import\",\n \"supply a source spectrum. \\\"\\\"\\\" import astropy.units as u try: import grizli.utils_c interp\",\n \"fp.write('{0:6d} {1}\\\\n'.format(len(filter.wave), filter.name)) for i in range(len(filter.wave)): fp.write('{0:6d} {1:.5e} {2:.5e}\\\\n'.format(i+1, filter.wave[i], filter.throughput[i])) fp.close()\",\n \"a EAZY filter file. .. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters\",\n \"case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d}\",\n \"bandpass, flam dlam num = np.trapz(vega_full*thru_full, full_x) den = np.trapz(absp*thru_full, full_x) return -2.5*np.log10(num/den)\",\n \"Interpolate to wavelength grid, no extrapolation vega_full = interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0)\",\n \"= np.interp([0.025, 0.975], filt/filt.max(), self.wave[1:]) return np.diff(ctw95)[0] def for_filter_file(self, row_str='{i:6} {wave:.5e} {thru:.5e}'): \\\"\\\"\\\"\",\n \"not case: search_string = search_string.upper() matched = [] for i in range(len(self.filters)): filt_name\",\n \"= ' '.join(line.split()[1:]) wave = [] trans = [] else: lspl = np.cast[float](line.split())\",\n \"filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is\",\n \"= grizli.utils_c.interp.interp_conserve_c except ImportError: interp = utils.interp_conserve if self.wave is None: print('Filter not\",\n \"len(wave) > 0: # Make filter from lines already read in new_filter =\",\n \"def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter, e.g., 161 = 2mass-j \\\"\\\"\\\" return\",\n \"right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same\",\n \"factor \\\"\\\"\\\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) self.Alambda = f99(self.wave)\",\n \"# AB = 0, same units absp = 3631*1e-23*c.to(u.m/u.s).value*1.e10/full_x**2 # Integrate over the\",\n \"'{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines = [header.format(N,\",\n \"\\\"\\\"\\\" if verbose: for i in range(len(self.filters)): print('{0:5d} {1}'.format(i+1, self.filters[i].name)) else: string_list =\",\n \"Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary normalization bp : optional,\",\n \"\\\"\\\"\\\" Return a string that can be put in the EAZY filter file\",\n \"filt_name.upper() if re.search(search_string, filt_name) is not None: if verbose: print('{0:5d} {1}'.format(i+1, self.filters[i].name)) matched.append(i)\",\n \"range(len(self.filters)): filt_name = self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name)\",\n \"self.filters[i].name) for i in range(len(self.filters))] return string_list def write(self, file='xxx.res', verbose=True): \\\"\\\"\\\" Dump\",\n \"\\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants import c import astropy.units as u\",\n \"return False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux = interp(self.wave,\",\n \"# Make filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans))\",\n \"print(len(res.filters)) bp = res[205] print(bp) fig, ax = plt.subplots(1,1,figsize=(6,4)) ax.plot(bp.wave, bp.throughput, label=bp.name.split()[0]) ax.set_xlabel('wavelength,\",\n \"VEGA.colnames: VEGA[c] = VEGA[c].astype(float) class FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\"\",\n \"file \\\"\\\"\\\" header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N =\",\n \"0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave) delta = np.trapz(self.throughput*source_flux*10**(-0.4*Alambda), self.wave)\",\n \"\\\"\\\"\\\" Get the MW extinction correction within the filter. Optionally supply a source\",\n \"a string that can be put in the EAZY filter file \\\"\\\"\\\" header\",\n \"= np.cast[float](line.split()) wave.append(lspl[1]) trans.append(lspl[2]) # last one # new_filter = FilterDefinition() # new_filter.name\",\n \"the list \\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed filter, e.g.,\",\n \"interp(full_x, VEGA['WAVELENGTH'], VEGA['FLUX'], left=0, right=0) thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) #\",\n \"that can be put in the EAZY filter file \\\"\\\"\\\" header = '{0}\",\n \"filters in the list \\\"\\\"\\\" return len(self.filters) def __getitem__(self, i1): \\\"\\\"\\\" Return unit-indexed\",\n \"Rv=3.1): \\\"\\\"\\\" Extinction factor \\\"\\\"\\\" import astropy.units as u f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv)\",\n \"FilterDefinition: def __init__(self, name=None, wave=None, throughput=None, bp=None): \\\"\\\"\\\" Bandpass object Parameters ---------- name\",\n \"wave=w, thru=t) for i, (w, t) in enumerate(zip(self.wave, self.throughput))] return '\\\\n'.join(lines) class FilterFile:\",\n \"array Throughput, arbitrary normalization bp : optional, `pysynphot.obsbandpass` object `pysynphot` filter bandpass \\\"\\\"\\\"\",\n \"verbose=True): \\\"\\\"\\\" Search filter names for ``search_string``. If ``case`` is True, then match\",\n \"1./delta @property def ABVega(self): \\\"\\\"\\\" Compute AB-Vega conversion \\\"\\\"\\\" from astropy.constants import c\",\n \"defined.') return False if source_flux is None: source_flux = self.throughput*0.+1 else: source_flux =\",\n \"Make filter from lines already read in new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) #\",\n \"= np.cast[float](trans) new_filter = FilterDefinition(name=header, wave=np.cast[float](wave), throughput=np.cast[float](trans)) filters.append(new_filter) self.filters = filters @property def\",\n \"ctw95(self): \\\"\\\"\\\" 95% cumulative throughput width http://www.stsci.edu/hst/acs/analysis/bandwidths/#keywords \\\"\\\"\\\" dl = np.diff(self.wave) filt =\",\n \"= self.filters[i].name if not case: filt_name = filt_name.upper() if re.search(search_string, filt_name) is not\",\n \"= [] trans = [] header = '' for line in lines: if\",\n \"num = integrator(self.wave, self.wave*self.throughput) den = integrator(self.wave, self.throughput/self.wave) pivot = np.sqrt(num/den) return pivot\",\n \"= [\\\"FilterDefinition\\\", \\\"FilterFile\\\", \\\"ParamFilter\\\"] VEGA_FILE = os.path.join(utils.path_to_eazy_data(), 'alpha_lyr_stis_008.fits') VEGA = Table.read(VEGA_FILE) for c\",\n \"Get the MW extinction correction within the filter. Optionally supply a source spectrum.\",\n \"{1}'.format(i+1, self.filters[i].name)) else: string_list = ['{0:5d} {1}\\\\n'.format(i+1, self.filters[i].name) for i in range(len(self.filters))] return\",\n \".. plot:: :include-source: import matplotlib.pyplot as plt from eazy.filters import FilterFile res =\",\n \"source_flux=None): \\\"\\\"\\\" Get the MW extinction correction within the filter. Optionally supply a\",\n \"name wave : array Wavelength array, in `astropy.units.Angstrom`. throughput : array Throughput, arbitrary\",\n \"lines = fp.readlines() self.filename = file_path filters = [] wave = [] trans\",\n \"= file_path filters = [] wave = [] trans = [] header =\",\n \"Angstroms') ax.set_ylabel('throughput') ax.legend() ax.grid() fig.tight_layout(pad=0.5) \\\"\\\"\\\" if path is None: file_path = os.path.join(os.getenv('EAZYCODE'),\",\n \"verbose=True): \\\"\\\"\\\" Dump the filter information to a filter file. \\\"\\\"\\\" fp =\",\n \"be put in the EAZY filter file \\\"\\\"\\\" header = '{0} {1} lambda_c=\",\n \"re if not case: search_string = search_string.upper() matched = [] for i in\",\n \"> 6.e4): Alambda = 0. else: f99 = utils.GalacticExtinction(EBV=EBV, Rv=Rv) Alambda = f99(self.wave)\",\n \"thru_full = interp(full_x, self.wave, self.throughput, left=0, right=0) # AB = 0, same units\",\n \"filter header = ' '.join(line.split()[1:]) wave = [] trans = [] else: lspl\",\n \"@property def equivwidth(self): \\\"\\\"\\\" Filter equivalent width http://pysynphot.readthedocs.io/en/latest/properties.html \\\"\\\"\\\" return np.trapz(self.throughput, self.wave) @property\",\n \"def __init__(self, line='# Filter #20, RES#78: COSMOS/SUBARU_filter_B.txt - lambda_c=4458.276253'): self.lambda_c = float(line.split('lambda_c=')[1]) self.name\",\n \"header = '{0} {1} lambda_c= {2:.4e} AB-Vega= {3:.3f} w95={4:.1f}' N = len(self.wave) lines\",\n \"__init__(self, file='FILTER.RES.latest', path='./'): \\\"\\\"\\\" Read a EAZY filter file. .. plot:: :include-source: import\"\n]","isConversational":false}}},{"rowIdx":3595,"cells":{"ngram":{"kind":"list like","value":["None r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:],","class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder:","TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index], postorder) return","self.right = right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode:","<reponame>KevinTMtz/CompetitiveProgramming # # LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/","- https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. # class TreeNode:","not inorder: return None r = postorder.pop() root = TreeNode(r) index = inorder.index(r)","right=None): # self.val = val # self.left = left # self.right = right","# # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for","val=0, left=None, right=None): # self.val = val # self.left = left # self.right","inorder: return None r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right","# self.left = left # self.right = right class Solution: def buildTree(self, inorder:","TreeNode: if not inorder: return None r = postorder.pop() root = TreeNode(r) index","a binary tree node. # class TreeNode: # def __init__(self, val=0, left=None, right=None):","# Definition for a binary tree node. # class TreeNode: # def __init__(self,","def __init__(self, val=0, left=None, right=None): # self.val = val # self.left = left","LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition","106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node.","= right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if","# URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. #","# # Definition for a binary tree node. # class TreeNode: # def","def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return None","https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. # class TreeNode: #","root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index],","self.val = val # self.left = left # self.right = right class Solution:","if not inorder: return None r = postorder.pop() root = TreeNode(r) index =","List[int], postorder: List[int]) -> TreeNode: if not inorder: return None r = postorder.pop()","# self.right = right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) ->","Definition for a binary tree node. # class TreeNode: # def __init__(self, val=0,","tree node. # class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val","postorder: List[int]) -> TreeNode: if not inorder: return None r = postorder.pop() root","return None r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right =","TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val = val # self.left","# LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # #","left # self.right = right class Solution: def buildTree(self, inorder: List[int], postorder: List[int])","postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left =","# self.val = val # self.left = left # self.right = right class","Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary","= left # self.right = right class Solution: def buildTree(self, inorder: List[int], postorder:","buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return None r","# # LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ #","= postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left","= TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index], postorder)","index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index], postorder) return root","class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val = val #","= val # self.left = left # self.right = right class Solution: def","-> TreeNode: if not inorder: return None r = postorder.pop() root = TreeNode(r)","URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. # class","# def __init__(self, val=0, left=None, right=None): # self.val = val # self.left =","val # self.left = left # self.right = right class Solution: def buildTree(self,","node. # class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val =","left=None, right=None): # self.val = val # self.left = left # self.right =","inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return None r =","r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder)","Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return","__init__(self, val=0, left=None, right=None): # self.val = val # self.left = left #","right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not","self.left = left # self.right = right class Solution: def buildTree(self, inorder: List[int],","binary tree node. # class TreeNode: # def __init__(self, val=0, left=None, right=None): #","for a binary tree node. # class TreeNode: # def __init__(self, val=0, left=None,","- 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree","List[int]) -> TreeNode: if not inorder: return None r = postorder.pop() root =","# Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a","# class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val = val"],"string":"[\n \"None r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:],\",\n \"class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder:\",\n \"TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index], postorder) return\",\n \"self.right = right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode:\",\n \"<reponame>KevinTMtz/CompetitiveProgramming # # LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/\",\n \"- https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. # class TreeNode:\",\n \"not inorder: return None r = postorder.pop() root = TreeNode(r) index = inorder.index(r)\",\n \"right=None): # self.val = val # self.left = left # self.right = right\",\n \"# # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for\",\n \"val=0, left=None, right=None): # self.val = val # self.left = left # self.right\",\n \"inorder: return None r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right\",\n \"# self.left = left # self.right = right class Solution: def buildTree(self, inorder:\",\n \"TreeNode: if not inorder: return None r = postorder.pop() root = TreeNode(r) index\",\n \"a binary tree node. # class TreeNode: # def __init__(self, val=0, left=None, right=None):\",\n \"# Definition for a binary tree node. # class TreeNode: # def __init__(self,\",\n \"def __init__(self, val=0, left=None, right=None): # self.val = val # self.left = left\",\n \"LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition\",\n \"106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node.\",\n \"= right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if\",\n \"# URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. #\",\n \"# # Definition for a binary tree node. # class TreeNode: # def\",\n \"def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return None\",\n \"https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. # class TreeNode: #\",\n \"root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index],\",\n \"self.val = val # self.left = left # self.right = right class Solution:\",\n \"if not inorder: return None r = postorder.pop() root = TreeNode(r) index =\",\n \"List[int], postorder: List[int]) -> TreeNode: if not inorder: return None r = postorder.pop()\",\n \"# self.right = right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) ->\",\n \"Definition for a binary tree node. # class TreeNode: # def __init__(self, val=0,\",\n \"tree node. # class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val\",\n \"postorder: List[int]) -> TreeNode: if not inorder: return None r = postorder.pop() root\",\n \"return None r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right =\",\n \"TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val = val # self.left\",\n \"# LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # #\",\n \"left # self.right = right class Solution: def buildTree(self, inorder: List[int], postorder: List[int])\",\n \"postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left =\",\n \"# self.val = val # self.left = left # self.right = right class\",\n \"Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary\",\n \"= left # self.right = right class Solution: def buildTree(self, inorder: List[int], postorder:\",\n \"buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return None r\",\n \"# # LeetCode # # Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ #\",\n \"= postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left\",\n \"= TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index], postorder)\",\n \"index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder) root.left = self.buildTree(inorder[:index], postorder) return root\",\n \"class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val = val #\",\n \"= val # self.left = left # self.right = right class Solution: def\",\n \"-> TreeNode: if not inorder: return None r = postorder.pop() root = TreeNode(r)\",\n \"URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree node. # class\",\n \"# def __init__(self, val=0, left=None, right=None): # self.val = val # self.left =\",\n \"val # self.left = left # self.right = right class Solution: def buildTree(self,\",\n \"node. # class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val =\",\n \"left=None, right=None): # self.val = val # self.left = left # self.right =\",\n \"inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return None r =\",\n \"r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:], postorder)\",\n \"Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder: return\",\n \"__init__(self, val=0, left=None, right=None): # self.val = val # self.left = left #\",\n \"right class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not\",\n \"self.left = left # self.right = right class Solution: def buildTree(self, inorder: List[int],\",\n \"binary tree node. # class TreeNode: # def __init__(self, val=0, left=None, right=None): #\",\n \"for a binary tree node. # class TreeNode: # def __init__(self, val=0, left=None,\",\n \"- 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a binary tree\",\n \"List[int]) -> TreeNode: if not inorder: return None r = postorder.pop() root =\",\n \"# Problem - 106 # URL - https://leetcode.com/problems/construct-binary-tree-from-inorder-and-postorder-traversal/ # # Definition for a\",\n \"# class TreeNode: # def __init__(self, val=0, left=None, right=None): # self.val = val\"\n]","isConversational":false}}},{"rowIdx":3596,"cells":{"ngram":{"kind":"list like","value":["[] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted:","try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores","percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id,","not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) +","requirement that automl search has been run. Arguments: pipeline (PipelineBase): pipeline to train","> best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score","Function called when `search()` errors and raises an Exception. Callback function takes three","Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" )","the X_train argument') if y_train is None: raise ValueError('Must specify training data target","pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN:","max_batches >= {ensemble_nth_batch + 1} to run ensembling.\") else: logger.info(f\"Ensembling will run every","objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and","return False # check for early stopping if self.patience is None or self.tolerance","{**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class,","event add_to_rankings is called before search _baseline_cv_scores will be empty so we will","self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value)","o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise","obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc","or patience < 0): raise ValueError(\"patience value must be a positive integer. Received","(list(str, ModelFamily)): The model families to search. The default of None searches over","[fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if","for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) /","be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot in Jupyter","= [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores:","data target values as a 1d vector using the y_train argument') try: self.problem_type","splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is not compatible with a","post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for","full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the current AutoML batch","self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as","logger.info(\"Input for ensembler are pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training","better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and score pipelines.\") if","of the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results","if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping search early...\".format(self.patience)) return False","None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines allowed in","object at file path Arguments: file_path (str): location to save file pickle_protocol (int):","(callable): Function called when `search()` errors and raises an Exception. Callback function takes","if patience and (not isinstance(patience, int) or patience < 0): raise ValueError(\"patience value","pipelines to train for every batch after the first one. The first batch","time budget (if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return","families: %s\\n\" % \", \".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if","raise AutoMLSearchException(f\"All pipelines in the current AutoML batch produced a score of np.nan","parameters needed to configure the search. For example, in time series problems, values","self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline:","data as a 2d array using the X_train argument') if y_train is None:","and - R2 for regression problems. max_iterations (int): Maximum number of iterations to","information such as type of pipeline components, problem, training time, cross validation, etc.","Exception raised, the traceback, and the AutoMLSearch object. Must also accepts kwargs, so","+ n_jobs) are used. ensembling (boolean): If True, runs ensembling in a separate","self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1","= y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def","\"\"\"Fits a baseline pipeline to the data. This is the first pipeline fit","'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates","be specified as seconds, minutes, or hours. patience (int): Number of iterations without","training results for the new pipeline, an untrained_pipeline containing the parameters used during","False. max_batches (int): The maximum number of batches of pipelines to search. Parameters","np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return","{}\".format(self.objective)) logger.info(\"Total training time (including CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\",","self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score","errors and raises an Exception. Callback function takes three positional parameters: the Exception","X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features]. Required. y_train","seconds.\\n\" % self.max_time) logger.info(\"Allowed model families: %s\\n\" % \", \".join([model.value for model in","False. Returns: Description of specified pipeline. Includes information such as type of pipeline","also fields like \"# Training\" and \"# Testing\", so we want to exclude","will be displayed in the log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout,","(list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable,","0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if","\"\"\"Returns a pandas.DataFrame with scoring results from all pipelines searched\"\"\" ascending = True","= split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train,","self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time Series","curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change:","have precedence over stopping the search. problem_configuration (dict, None): Additional parameters needed to","additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] +","'all_objective_scores' field contains scores for all objectives # but also fields like \"#","in automl.\") return objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name)","Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] ==","return True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional","break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True","any pipelines that error out during training will not be included in the","self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best","best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True","Setting this field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The","2d array using the X_train argument') if y_train is None: raise ValueError('Must specify","user asking if they want to stop the search. Returns: bool: If True,","all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of var\", c] = std / mean if","will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot","obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in","search. The default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to","self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline","= self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not","Received {max_time}.\") if max_batches is not None and max_batches < 0: raise ValueError(f\"Parameter","not be included in the dictionary but the exception and stacktrace will be","features will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score","if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger,","minutes, or hours. patience (int): Number of iterations without improvement to stop search","n_jobs (int or None): Non-negative integer describing level of parallelism used for pipelines.","from the current best pipeline before training and thresholding\"\"\" if len(self.rankings) == 0:","if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) //","start_iteration_callback (callable): Function called before each pipeline training iteration. Callback function takes three","appropriate parameters by default. Defaults to None, which will call `log_error_callback`. additional_objectives (list):","if X_train is None: raise ValueError('Must specify training data as a 2d array","the search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original stopping criterion and","None): Additional parameters needed to configure the search. For example, in time series","the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able to pass","None or non-negative. Received {max_batches}.\") if max_iterations is not None and max_iterations <","{self.max_iterations} pipelines. \") elif self.max_iterations is not None: logger.info(\"Searching up to %s pipelines.","(dict, None): Additional parameters needed to configure the search. For example, in time","if max_iterations is not None and max_iterations < 0: raise ValueError(f\"Parameter max_iterations must","as score improvement for early stopping. Only applicable if patience is not None.","import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators","The first batch will train a baseline pipline + one of each pipeline","= pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\"# Training\", \"# Validation\"]:","thresholding\"\"\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and","StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether","to train and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in","0): raise ValueError(\"patience value must be a positive integer. Received {} instead\".format(patience)) if","copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from the highest-scoring","non-negative. Received {max_time}.\") if max_batches is not None and max_batches < 0: raise","return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and logs a","= {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name","best pipeline before training and thresholding\"\"\" if len(self.rankings) == 0: return best_pipeline =","make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives","describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict","found in automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a","/ n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time =","at least one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines ==","AutoML object at file path Arguments: file_path (str): location to save file pickle_protocol","self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True,","pipelines after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed model families: %s\\n\" % \", \".join([model.value","logger.info(\"Objective to optimize binary classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training time (including","\"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family","objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for objective, score in","logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40 # Necessary","be included in the dictionary but the exception and stacktrace will be displayed","custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\"No allowed pipelines","= self._num_pipelines() if num_pipelines == 0: return True # check max_time and max_iterations","multiclass classification problems, and - R2 for regression problems. max_iterations (int): Maximum number","random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__,","to make sure it is different from the current best pipeline before training","this loop does not count towards the time budget (if set) time_in_loop =","= time.time() - start_of_loop self._start += time_in_loop return False else: leading_char = \"\"","of np.nan on the primary objective {self.objective}.\") self.search_duration = time.time() - self._start elapsed_time","pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for","pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations,","evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types","if tolerance and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\"tolerance value","pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name","\"Do you really want to exit search (y/n)? \").strip().lower() if choice == \"y\":","float]]: Dictionary keyed by pipeline name that maps to a dictionary of scores.","(int): The number of pipelines to train for every batch after the first","} self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try: self.plot =","family allowed in the search. \"\"\" if X_train is None: raise ValueError('Must specify","save file pickle_protocol (int): the pickle data stream format. Returns: None \"\"\" with","specify training data target values as a 1d vector using the y_train argument')","pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by","pipelines that error out during training will not be included in the dictionary","self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc =","None: raise ValueError('Must specify training data as a 2d array using the X_train","prompt to the user asking if they want to stop the search. Returns:","to use. Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults to","= infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type,","ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import","for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start","a baseline pipeline to the data. This is the first pipeline fit during","Note that if allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator):","for training ensemble metalearners. Only used when ensembling is True. Must be between","> AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number","all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\",","used. For n_jobs below -1, (n_cpus + 1 + n_jobs) are used. ensembling","ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner","= [] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines =","cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback:","that if allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data","enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already been run and will not run","or not all(p in problem_configuration for p in required_parameters): raise ValueError(\"user_parameters must be","num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping search","not compatible with a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives =","TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\":","\"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df =","import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import","for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv","pipeline components, problem, training time, cross validation, etc. \"\"\" if pipeline_id not in","train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to the","is set to False, returns an untrained pipeline instance. \"\"\" if not self._best_pipeline:","the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout","in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is not compatible with","tolerance and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\"tolerance value must","pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with the","Defaults to False. max_batches (int): The maximum number of batches of pipelines to","pass along other appropriate parameters by default. Defaults to None, which will call","infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed)","str, type(None))): raise TypeError(f\"Parameter max_time must be a float, int, string or None.","{str(max_time)}..\") if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\"Parameter max_time must","pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted =","specified pipeline initialized with the parameters used to train that pipeline during automl","that maps to the fitted pipeline. Note that the any pipelines that error","len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0 <","used to train that pipeline during automl search. Arguments: pipeline_id (int): pipeline to","objective (str, ObjectiveBase): The objective to optimize for. Used to propose and rank","\"Time Series Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class):","\"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df","True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\",","within cross validation scores. {pipeline_name} may not perform as estimated on unseen data.\")","not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id]","list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline","= get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not","evaluations which have been made Returns: int: the number of pipeline evaluations made","yes, False if no. \"\"\" if self._interrupted: return False # for add_to_rankings if","\" \"to get all objective names allowed in automl.\") return objective() return objective","Description of specified pipeline. Includes information such as type of pipeline components, problem,","must be between 0 and 1 exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _,","{self.max_batches} batches for a total of {self.max_iterations} pipelines. \") elif self.max_iterations is not","on the primary objective {self.objective}.\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start)","during training will not be included in the dictionary but the exception and","search. Parameters max_time, and max_iterations have precedence over stopping the search. problem_configuration (dict,","pipeline to the data. This is the first pipeline fit during search. \"\"\"","_pipelines_per_batch (int): The number of pipelines to train for every batch after the","from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds,","pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool):","out during scoring will not be included in the dictionary but the exception","tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to","\"Beginning pipeline search\") logger.info(\"Optimizing for %s. \" % self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater'","== ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger,","positional parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback","logger.info(\"Allowed model families: %s\\n\" % \", \".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot =","for a full list. objective (str, ObjectiveBase): The objective to optimize for. Used","start_of_loop = time.time() while True: choice = input(leading_char + \"Do you really want","higher than specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) >","for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\"No allowed pipelines to","return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a list of pipelines","as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML object at file","{_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size,","in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective]","Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed:","all pipelines for this problem type are allowed. Setting this field will cause","== \"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif choice == \"n\": # So that","def _should_continue(self): \"\"\"Given the original stopping criterion and current state, should the search","search. For example, in time series problems, values should be passed in for","if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch =","set of parameters used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self):","score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError:","learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective","return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id (int):","and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able to","from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results from each pipeline,","to a dictionary of scores. Note that the any pipelines that error out","Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with","variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may not perform as","X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None,","= cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores =","name that maps to the fitted pipeline. Note that the any pipelines that","also accepts kwargs, so AutoMLSearch is able to pass along other appropriate parameters","if pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline {} is not compatible with problem_type","pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\"","if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed in AutoML! \" \"Use","_check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and logs a warning if variance","return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file path Arguments:","Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary` to `multiclass` or `regression` depending on","self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and","type(None))): raise TypeError(f\"Parameter max_time must be a float, int, string or None. Received","isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch","before selecting `best_pipeline`.\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at","get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline )","optimize for. Used to propose and rank pipelines, but not for optimizing each","+ num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches -","early \"\"\" leading_char = \"\\n\" start_of_loop = time.time() while True: choice = input(leading_char","after the first one. The first batch will train a baseline pipline +","evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types","for o in obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else:","or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows an","in the log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score","problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback","search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str,","objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective)","= self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved =","None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\"","(tolerance > 1.0 or tolerance < 0.0): raise ValueError(\"tolerance value must be a","allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if","the current AutoML batch produced a score of np.nan on the primary objective","is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches","each pipeline training iteration. Callback function takes three positional parameters: The pipeline class,",">= {threshold}) within cross validation scores. {pipeline_name} may not perform as estimated on","self.problem_type: raise ValueError(\"Given pipeline {} is not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def","will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold.","objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results'])","elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline =","class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40 # Necessary for \"Plotting\" documentation,","chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification problems,","the user asking if they want to stop the search. Returns: bool: If","may not perform as estimated on unseen data.\") return high_variance_cv def get_pipeline(self, pipeline_id):","[ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration","the traceback, and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is","in the rankings If self._best_pipeline already exists, check to make sure it is","PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family ==","self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None","except ImportError: logger.warning(\"Unable to import plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines","y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter,","for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is not","threshold) if high_variance_cv: logger.warning(f\"High coefficient of variation (cv >= {threshold}) within cross validation","really want to exit search (y/n)? \").strip().lower() if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\")","a separate batch after every allowed pipeline class has been iterated over. If","time can be specified as seconds, minutes, or hours. patience (int): Number of","return True elif choice == \"n\": # So that the time in this","specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if","max_iterations >= {first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling will run at the {first_ensembling_iteration}","vs. score plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \"\"\"","PipelineNotFoundError(\"Pipeline class or parameters not found in automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def","if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm =","return # don't show iteration plot outside of a jupyter notebook if show_iteration_plot:","None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration =","and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\"Using default limit of","(cv >= {threshold}) within cross validation scores. {pipeline_name} may not perform as estimated","if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is not compatible with a {}","(int, float, str, type(None))): raise TypeError(f\"Parameter max_time must be a float, int, string","parameters used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns a",") from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data","included in the dictionary but the exception and stacktrace will be displayed in","or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the user asking if they","from each pipeline, and `search_order`: a list describing the order the pipelines were","pipelines. This will not start a new pipeline search after the duration has","\"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids =","logger.info(f\"Using {self._engine.__class__.__name__} to train and score pipelines.\") if self.max_batches is not None: logger.info(f\"Searching","+ 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0:","will train a baseline pipline + one of each pipeline family allowed in","Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance:","already been run and will not run again on the same instance. Re-initialize","name that maps to a dictionary of scores. Note that the any pipelines","with scoring results from the highest-scoring set of parameters used with each pipeline.\"\"\"","early stopping. Only applicable if patience is not None. Defaults to None. allowed_pipelines","to -1, all CPUs are used. For n_jobs below -1, (n_cpus + 1","self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id]","\" % self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__}","valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is not compatible","\\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start","== ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else:","= input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score","f\"\\nSearch finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not","1: logger.warning(\"Ensembling is set to True, but the number of unique pipelines is","Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random","before returning it. Defaults to True. pipeline_parameters (dict): A dict of the parameters","self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds =","self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families","= rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \"\"\"Returns","optimize binary classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training time (including CV): %.1f","any pipelines that error out during scoring will not be included in the","self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping search early...\".format(self.patience)) return False return True def","* self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is","float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\") if isinstance(max_time, (int,","the log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a","of objectives to score on. Will override default objectives for problem type if","add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling","ascending = False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if","= 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score -","\"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a list of","pipeline instance. Returns: PipelineBase: A trained instance of the best pipeline and parameters","one of each pipeline family allowed in the search. \"\"\" if X_train is","Received {type(max_time)} with value {str(max_time)}..\") if isinstance(max_time, (int, float)) and max_time < 0:","a dictionary of scores. Note that the any pipelines that error out during","if not empty. random_seed (int): Seed for the random number generator. Defaults to","binary classification problems, - LogLossMulticlass for multiclass classification problems, and - R2 for","run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\"Ensembling split size must","ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\"Ensembling is set to True,","of the parameters used to initalize a pipeline with. _ensembling_split_size (float): The amount","elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc)","associated with the provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None:","that allows access to a copy of the results from `automl_search`. Returns: dict","and evaluates a given pipeline then adds the results to the automl rankings","(including CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"]","to {[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters)","pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\"# Training\", \"# Validation\"]: all_objective_scores[c]","for fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c","logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"]","objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o in obj_list]) return","returns an untrained instance of the specified pipeline initialized with the parameters used","self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import plotly; skipping pipeline search plotting\\n\")","self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total","or `regression` depending on the problem type. Note that if allowed_pipelines is provided,","the search. Returns: bool: If True, search should terminate early \"\"\" leading_char =","get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from","A list of PipelineBase subclasses indicating the pipelines allowed in the search. The","model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary` to `multiclass` or `regression`","the provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline","percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id","== ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif","find file to load Returns: AutoSearchBase object \"\"\" with open(file_path, 'rb') as f:","a pipeline training result, returns an untrained instance of the specified pipeline initialized","of parameters used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns","int, string or None. Received {type(max_time)} with value {str(max_time)}..\") if isinstance(max_time, (int, float))","self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else:","ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type","first one. The first batch will train a baseline pipline + one of","parameters is None: raise PipelineNotFoundError(\"Pipeline class or parameters not found in automl results\")","over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\")","scores and logs a warning if variance is higher than specified threshhold.\"\"\" pipeline_name","allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) ==","ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type]","\"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data)","for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline","use. Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner.","don't show iteration plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython","\"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original stopping criterion and current state,","called before search _baseline_cv_scores will be empty so we will return # nan","ImportError: logger.warning(\"Unable to import plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families","be displayed in the log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout,","with value {str(max_time)}..\") if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\"Parameter","every allowed pipeline class has been iterated over. If the number of unique","_, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine =","during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline","return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from","be in seconds. For strings, time can be specified as seconds, minutes, or","scores[field] += value return {objective: float(score) / n_folds for objective, score in scores.items()}","is too small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch +","the gap and max_delay \" f\"parameters. Received {problem_configuration}.\") return problem_configuration or {} def","= objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id =","containing the parameters used during training, and the AutoMLSearch object. error_callback (callable): Function","the pickle data stream format. Returns: None \"\"\" with open(file_path, 'wb') as f:","set aside for training ensemble metalearners. Only used when ensembling is True. Must","X_train argument') if y_train is None: raise ValueError('Must specify training data target values","best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self):","max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations","rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df","self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid data","self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train,","np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] =","received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type,","list. objective (str, ObjectiveBase): The objective to optimize for. Used to propose and","= time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines =","self.max_batches = 1 logger.info(\"Using default limit of max_batches=1.\\n\") if patience and (not isinstance(patience,","{} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in","X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if","AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training iteration. Callback function","the AutoMLSearch object. error_callback (callable): Function called when `search()` errors and raises an","high_variance_cv: logger.warning(f\"High coefficient of variation (cv >= {threshold}) within cross validation scores. {pipeline_name}","= get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines =","= abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score > best_score","\"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\":","self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search _baseline_cv_scores will be","or not to train the best pipeline before returning it. Defaults to True.","self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time,","1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number","{best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self): \"\"\"Finds the best pipeline in the rankings","which have been made Returns: int: the number of pipeline evaluations made in","data we'll set aside for training ensemble metalearners. Only used when ensembling is","for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o","pipeline family allowed in the search. \"\"\" if X_train is None: raise ValueError('Must","without improvement. Stopping search early...\".format(self.patience)) return False return True def _validate_problem_type(self): for obj","= [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines ==","run and will not run again on the same instance. Re-initialize AutoMLSearch to","or hours. patience (int): Number of iterations without improvement to stop search early.","(boolean, True): Shows an iteration vs. score plot in Jupyter notebook. Disabled by","search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance. Returns:","log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective ==","training data of length [n_samples]. Required for supervised learning tasks. problem_type (str or","ensemble metalearners. Only used when ensembling is True. Must be between 0 and","< first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_iterations is","and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type):","= time_elapsed(self._start) desc = f\"\\nSearch finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline()","pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\"Pipeline class or","if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification pipeline thresholds","for the new pipeline, an untrained_pipeline containing the parameters used during training, and","self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {}","> self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score < best_score","False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx","batches of pipelines to search. Parameters max_time, and max_iterations have precedence over stopping","numpy as np import pandas as pd import woodwork as ww from sklearn.model_selection","\"\"\"Fits and evaluates a given pipeline then adds the results to the automl","run before selecting `best_pipeline`.\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object","if obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives","or tolerance < 0.0): raise ValueError(\"tolerance value must be a float between 0.0","default. Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom set of","has been iterated over. If the number of unique pipelines to search over","a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger,","return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If","kwargs, so AutoMLSearch is able to pass along other appropriate parameters by default.","allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True,","options. Change `binary` to `multiclass` or `regression` depending on the problem type. Note","Only used when ensembling is True. Must be between 0 and 1, exclusive.","best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score =","= self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids","to False, returns an untrained pipeline instance. \"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl","self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type):","get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None","(pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features]. Required. y_train (pd.Series,","- 3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is","evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from","A dictionary containing the training results for the new pipeline, an untrained_pipeline containing","set of objectives for problem_type, remove it existing_main_objective = next((obj for obj in","search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original stopping criterion and current","(float): Minimum percentage difference to qualify as score improvement for early stopping. Only","self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in","them since # they are not scores if field in objective_name_to_class: scores[field] +=","pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay =","set. Arguments: feature_types (list, optional): list of feature types, either numerical or categorical.","parameters used during training, and the AutoMLSearch object. error_callback (callable): Function called when","raise ValueError(\"tolerance value must be a float between 0.0 and 1.0 inclusive. Received","{first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling will run at the {first_ensembling_iteration} iteration and","random_seed self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError:","= [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if not","split size must be between 0 and 1 exclusive, received {_ensembling_split_size}\") X_shape =","it is different from the current best pipeline before training and thresholding\"\"\" if","PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective )","to the fitted pipeline. Note that the any pipelines that error out during","random_seed (int): Seed for the random number generator. Defaults to 0. n_jobs (int","to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \") elif self.max_iterations is","Dict[str, float]]: Dictionary keyed by pipeline name that maps to a dictionary of","len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline ==","True, but max_batches is too small, so ensembling will not run. Set max_batches","is not compatible with a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives","existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives]","new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines","a pandas.DataFrame with scoring results from all pipelines searched\"\"\" ascending = True if","least one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines == 0:","self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot = None","be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class:","= ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION:","method to use. Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults","self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\"Using default limit of max_batches=1.\\n\") if","then max_iterations will default to max_iterations of 5. max_time (int, str): Maximum time","add_to_rankings if self._searched: return True # Run at least one pipeline for every","pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id (int): pipeline","1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch","ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\")","f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\"","= self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set to True,","in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters}","validation scores. {pipeline_name} may not perform as estimated on unseen data.\") return high_variance_cv","evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters","self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines","is None or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score =","axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None):","the number of pipeline evaluations which have been made Returns: int: the number","access to a copy of the results from `automl_search`. Returns: dict containing `pipeline_results`:","displayed in the log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives):","not None: logger.info(f\"Searching up to {self.max_batches} batches for a total of {self.max_iterations} pipelines.","and current state, should the search continue? Returns: bool: True if yes, False","def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o in obj_list]) return '\\n'.join(lines) def _get_funct_name(function):","\"\"\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads","problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to the data. This","self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches)","True elif choice == \"n\": # So that the time in this loop","type if not empty. random_seed (int): Seed for the random number generator. Defaults","pipelines.\") if self.max_batches is not None: logger.info(f\"Searching up to {self.max_batches} batches for a","= [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols)","runs ensembling in a separate batch after every allowed pipeline class has been","evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines","self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class =","each pipeline, and `search_order`: a list describing the order the pipelines were searched.","- self._start if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and","used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns a pandas.DataFrame","so AutoMLSearch is able to pass along other appropriate parameters by default. Defaults","problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or","inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \"\"\"Returns a trained instance of","self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline,","the time budget (if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop","max_delay}, \"Time Series Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data,","self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if","[\"# Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std =","for the data set. Arguments: feature_types (list, optional): list of feature types, either","X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train)","= self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"):","one, so ensembling will not run.\") run_ensembling = False if run_ensembling and self.max_iterations","and self.max_iterations is not None: # Baseline + first batch + each pipeline","in time series problems, values should be passed in for the gap and","_ensembling_split_size (float): The amount of the training data we'll set aside for training","the training results for the new pipeline, an untrained_pipeline containing the parameters used","(list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline","= self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train","for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data","format. Returns: None \"\"\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod","instance. Returns: PipelineBase: A trained instance of the best pipeline and parameters found","cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0]","o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives","not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is not compatible with a {} problem.\".format(self.objective.name,","pipeline name that maps to the fitted pipeline. Note that the any pipelines","of variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may not perform","cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML object at file path Arguments:","leading_char = \"\" def search(self, show_iteration_plot=True): \"\"\"Find the best pipeline for the data","NameError: show_iteration_plot = False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for %s. \" %","training and thresholding\"\"\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not","ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted =","the number of unique pipelines is one, so ensembling will not run.\") run_ensembling","= data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective","or parameters is None: raise PipelineNotFoundError(\"Pipeline class or parameters not found in automl","called when `search()` errors and raises an Exception. Callback function takes three positional","@property def results(self): \"\"\"Class that allows access to a copy of the results","False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']:","file pickle_protocol (int): the pickle data stream format. Returns: None \"\"\" with open(file_path,","ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc =","Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives","pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and","'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter","current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines","for every batch after the first one. The first batch will train a","Maximum time to search for pipelines. This will not start a new pipeline","and max_iterations < 0: raise ValueError(f\"Parameter max_iterations must be None or non-negative. Received","scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps to","search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is","= get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and not","in the search. \"\"\" if X_train is None: raise ValueError('Must specify training data","automl.\") return objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for","early...\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in self.additional_objectives: if not","max_iterations must be None or non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time,","is not allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all objective","for every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True #","\"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\":","pipeline instance associated with the provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results","self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return","== 0: return True # check max_time and max_iterations elapsed = time.time() -","if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot:","False # for add_to_rankings if self._searched: return True # Run at least one","ValueError(\"Given pipeline {} is not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits","not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\"Parameter max_time must be a float,","if score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement +=","to exclude them since # they are not scores if field in objective_name_to_class:","pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to","False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train,","if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed in","self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import plotly;","self._interrupted: return False # for add_to_rankings if self._searched: return True # Run at","1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\"Ensembling is set","parameters by default. Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom","of None indicates all pipelines for this problem type are allowed. Setting this","search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated","problems, - LogLossMulticlass for multiclass classification problems, and - R2 for regression problems.","use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary","Whether or not to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback","is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]:","variables. train_best_pipeline (boolean): Whether or not to train the best pipeline before returning","\" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all objective names allowed in automl.\") return","current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try:","list of PipelineBase subclasses indicating the pipelines allowed in the search. The default","default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options.","for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize =","train_best_pipeline (boolean): Whether or not to train the best pipeline before returning it.","protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML object at file path Arguments: file_path (str):","search continue? Returns: bool: True if yes, False if no. \"\"\" if self._interrupted:","value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives #","score is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and score","pipelines. \" % self.max_iterations) if self.max_time is not None: logger.info(\"Will stop searching for","isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed in AutoML!","training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training","a trained instance of the best pipeline and parameters found during automl search.","is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid data splitter\")","self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score = curr_score num_without_improvement","max_time and max_iterations elapsed = time.time() - self._start if self.max_time and elapsed >=","ValueError(\"Given objective {} is not compatible with a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives","@staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in","if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks","results(self): \"\"\"Class that allows access to a copy of the results from `automl_search`.","best pipeline and parameters found during automl search. If `train_best_pipeline` is set to","describe return_dict (bool): If True, return dictionary of information about pipeline. Defaults to","max_iterations have precedence over stopping the search. problem_configuration (dict, None): Additional parameters needed","documentation, since Sphinx does not work well with instance attributes. plot = PipelineSearchPlots","dict containing values for at least the gap and max_delay \" f\"parameters. Received","\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY)","f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if","since # they are not scores if field in objective_name_to_class: scores[field] += value","pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch","\"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\",","\"\"\"Train a list of pipelines on the training data. This can be helpful","[])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\"","= pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\"Pipeline class","self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted","Returns: dict containing `pipeline_results`: a dict with results from each pipeline, and `search_order`:","compatible with a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type)","parameters used to initalize a pipeline with. _ensembling_split_size (float): The amount of the","y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices]","None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid data splitter\") if not","pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True","\"\"\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id)","problems. max_iterations (int): Maximum number of iterations to search. If max_iterations and max_time","== self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o)","self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if","max_iterations is not None and max_iterations < 0: raise ValueError(f\"Parameter max_iterations must be","Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches:","@property def best_pipeline(self): \"\"\"Returns a trained instance of the best pipeline and parameters","self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train =","'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML object at","cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score =","False if self.allowed_pipelines is None: logger.info(\"Generating pipelines to search over...\") allowed_estimators = get_estimators(self.problem_type,","each pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary for binary","max_iterations and max_time is not set, then max_iterations will default to max_iterations of","validation, etc. \"\"\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline","{} is not compatible with a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines","convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger","ValueError('Must specify training data as a 2d array using the X_train argument') if","the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results from","ww.DataTable): The input training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn):","they are not scores if field in objective_name_to_class: scores[field] += value return {objective:","with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None,","open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a list of","pipeline. Defaults to False. Returns: Description of specified pipeline. Includes information such as","evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective,","True if yes, False if no. \"\"\" if self._interrupted: return False # for","self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score =","each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring","if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a given pipeline","== ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if","{self.objective}.\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch finished","initalize a pipeline with. _ensembling_split_size (float): The amount of the training data we'll","for pipelines. This will not start a new pipeline search after the duration","on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train.","choice == \"n\": # So that the time in this loop does not","self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation","@property def full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from all pipelines searched\"\"\"","if yes, False if no. \"\"\" if self._interrupted: return False # for add_to_rankings","problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None,","not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives =","max_delay variables. train_best_pipeline (boolean): Whether or not to train the best pipeline before","for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline':","elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False","pipeline_class is None or parameters is None: raise PipelineNotFoundError(\"Pipeline class or parameters not","(int or None): Non-negative integer describing level of parallelism used for pipelines. None","+ len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else:","a given pipeline then adds the results to the automl rankings with the","skipping pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None","must be run before selecting `best_pipeline`.\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves","the best pipeline for the data set. Arguments: feature_types (list, optional): list of","Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps to a dictionary","if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid","but not for optimizing each pipeline during fit-time. When set to 'auto', chooses:","train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input","if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline,","ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline,","or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback","to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The default","\"\\n\" start_of_loop = time.time() while True: choice = input(leading_char + \"Do you really","from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from","+ 1 + n_jobs) are used. ensembling (boolean): If True, runs ensembling in","array using the X_train argument') if y_train is None: raise ValueError('Must specify training","every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True # check","'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs","so we will return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name],","pipeline search after the duration has elapsed. If it is an integer, then","AutoMLSearch object. error_callback (callable): Function called when `search()` errors and raises an Exception.","to optimize binary classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training time (including CV):","= time.time() while True: choice = input(leading_char + \"Do you really want to","want to exit search (y/n)? \").strip().lower() if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return","None, which will call `log_error_callback`. additional_objectives (list): Custom set of objectives to score","supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The","Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary pipeline","recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted","asking if they want to stop the search. Returns: bool: If True, search","return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original stopping criterion and current state, should","a copy of the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict","str): Maximum time to search for pipelines. This will not start a new","0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score)","training data. This can be helpful for training pipelines once the search is","current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the current","return None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time:","ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if","None: logger.info(\"Searching up to %s pipelines. \" % self.max_iterations) if self.max_time is not","= pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High coefficient","pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits","pandas.DataFrame with scoring results from all pipelines searched\"\"\" ascending = True if self.objective.greater_is_better:","TypeError(f\"Parameter max_time must be a float, int, string or None. Received {type(max_time)} with","\"# Testing\", so we want to exclude them since # they are not","object. Must also accepts kwargs, so AutoMLSearch is able to pass along other","list of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of","pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines with IDs: \"","of batches of pipelines to search. Parameters max_time, and max_iterations have precedence over","which will call `log_error_callback`. additional_objectives (list): Custom set of objectives to score on.","add_to_rankings is called before search _baseline_cv_scores will be empty so we will return","if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc","error_callback (callable): Function called when `search()` errors and raises an Exception. Callback function","# So that the time in this loop does not count towards the","PipelineBase subclasses indicating the pipelines allowed in the search. The default of None","must be a float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\")","\"validation_score\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in","pipeline name that maps to a dictionary of scores. Note that the any","search early. Must be positive. If None, early stopping is disabled. Defaults to","will call `log_error_callback`. additional_objectives (list): Custom set of objectives to score on. Will","to retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided ID \"\"\"","be positive. If None, early stopping is disabled. Defaults to None. tolerance (float):","that the any pipelines that error out during training will not be included","as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback","[self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize","and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) +","to search again.\") return # don't show iteration plot outside of a jupyter","pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self,","\"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df =","{self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\"","True # check max_time and max_iterations elapsed = time.time() - self._start if self.max_time","if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score = curr_score","(0 < _ensembling_split_size < 1): raise ValueError(f\"Ensembling split size must be between 0","objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0,","f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs:","{self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\"","with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to the data.","n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters,","Defaults to None. tolerance (float): Minimum percentage difference to qualify as score improvement","compatible with a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if","for new pipelines after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed model families: %s\\n\" %","# In the event add_to_rankings is called before search _baseline_cv_scores will be empty","= self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return","criterion and current state, should the search continue? Returns: bool: True if yes,","// ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\"Ensembling is set to","(bool): Whether or not to optimize the binary pipeline threshold. Defaults to True.","and 1 exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ =","= pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time Series Baseline Estimator\": {\"gap\": gap, \"max_delay\":","classification problems, and - R2 for regression problems. max_iterations (int): Maximum number of","it existing_main_objective = next((obj for obj in additional_objectives if obj.name == self.objective.name), None)","set to True, but max_iterations is too small, so ensembling will not run.","disabled. Defaults to None. tolerance (float): Minimum percentage difference to qualify as score","new pipeline, an untrained_pipeline containing the parameters used during training, and the AutoMLSearch","None and max_batches < 0: raise ValueError(f\"Parameter max_batches must be None or non-negative.","objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()}","self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective","max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None,","to optimize for. Used to propose and rank pipelines, but not for optimizing","duration has elapsed. If it is an integer, then the time will be","True, search should terminate early \"\"\" leading_char = \"\\n\" start_of_loop = time.time() while","and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\"tolerance value must be","data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner","this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults","number of unique pipelines to search over per batch is one, ensembling will","ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline","can be helpful for training pipelines once the search is complete. Arguments: pipelines","run again on the same instance. Re-initialize AutoMLSearch to search again.\") return #","allowed in the search. \"\"\" if X_train is None: raise ValueError('Must specify training","False logger.warning(f\"Ensembling is set to True, but max_iterations is too small, so ensembling","= curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and","if not problem_configuration or not all(p in problem_configuration for p in required_parameters): raise","train the best pipeline before returning it. Defaults to True. pipeline_parameters (dict): A","desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name =","(int, str): Maximum time to search for pipelines. This will not start a","an untrained instance of the specified pipeline initialized with the parameters used to","pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True, return dictionary","= self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else","classification problems, - LogLossMulticlass for multiclass classification problems, and - R2 for regression","exception and stacktrace will be displayed in the log. \"\"\" return self._engine.train_batch(pipelines) def","pipeline evaluations made in the search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the","set to False, returns an untrained pipeline instance. Returns: PipelineBase: A trained instance","def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data:","(self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size <","in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for objective, score","found in automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is","= bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High coefficient of variation (cv","fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in","self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = []","a total of {self.max_iterations} pipelines. \") elif self.max_iterations is not None: logger.info(\"Searching up","= tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': []","* (self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size < 1):","time_elapsed(self._start) desc = f\"\\nSearch finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if","Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed","When set to 'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass","pipeline instance. \"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl search must be run before","default in non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already been run and","the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\") if self.max_batches","self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return the number of pipeline evaluations which","handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class","Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\"","cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High coefficient of variation (cv >= {threshold}) within","to search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for estimator","rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc","[] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch()","elapsed. If it is an integer, then the time will be in seconds.","same instance. Re-initialize AutoMLSearch to search again.\") return # don't show iteration plot","num_pipelines == 0: return True # check max_time and max_iterations elapsed = time.time()","List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame):","import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from","{ensemble_nth_batch} batches.\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1","= input(leading_char + \"Do you really want to exit search (y/n)? \").strip().lower() if","problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main","returning it. Defaults to True. pipeline_parameters (dict): A dict of the parameters used","pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a list of pipelines on the given holdout","all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results","pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\":","amount of the training data we'll set aside for training ensemble metalearners. Only","to exit search (y/n)? \").strip().lower() if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return True","max_iterations will default to max_iterations of 5. max_time (int, str): Maximum time to","pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class that","for early stopping. Only applicable if patience is not None. Defaults to None.","iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch +","in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe()","or non-negative. Received {max_batches}.\") if max_iterations is not None and max_iterations < 0:","best pipeline in the rankings If self._best_pipeline already exists, check to make sure","0: return True # check max_time and max_iterations elapsed = time.time() - self._start","} if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"]","[\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df","specified as seconds, minutes, or hours. patience (int): Number of iterations without improvement","max_iterations is too small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration}","Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True, return dictionary of","pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to","rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \"\"\"Returns a trained instance of the","while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm","= ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations:","raise ValueError(\"Given pipeline {} is not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self):","ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective","all_objective_scores = [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in","not scores if field in objective_name_to_class: scores[field] += value return {objective: float(score) /","of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of","< 0): raise ValueError(\"patience value must be a positive integer. Received {} instead\".format(patience))","if self.max_batches is not None: logger.info(f\"Searching up to {self.max_batches} batches for a total","self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores","update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives","small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\")","\"\"\"Score a list of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)):","parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to","with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML","allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {}","searching for new pipelines after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed model families: %s\\n\"","if not self._best_pipeline: raise PipelineNotFoundError(\"automl search must be run before selecting `best_pipeline`.\") return","len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original stopping criterion and current state, should the","> threshold) if high_variance_cv: logger.warning(f\"High coefficient of variation (cv >= {threshold}) within cross","import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils","{self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if not self.rankings.empty: rankings_str =","the best pipeline before returning it. Defaults to True. pipeline_parameters (dict): A dict","optional): list of feature types, either numerical or categorical. Categorical features will automatically","will default to max_iterations of 5. max_time (int, str): Maximum time to search","return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class that allows access to a copy","to None, which will call `log_error_callback`. additional_objectives (list): Custom set of objectives to","of pipelines to search. Parameters max_time, and max_iterations have precedence over stopping the","fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains","evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score","evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from","MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils","start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds","cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline),","to see options. Change `binary` to `multiclass` or `regression` depending on the problem","evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective to optimize for.","default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or","desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives =","and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the current AutoML batch produced a score","gap and max_delay variables. train_best_pipeline (boolean): Whether or not to train the best","If True, return dictionary of information about pipeline. Defaults to False. Returns: Description","raise PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family","does not work well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None,","additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for","= self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def","import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import","for %s. \" % self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower'))","np.nan on the primary objective {self.objective}.\") self.search_duration = time.time() - self._start elapsed_time =","pipeline then adds the results to the automl rankings with the requirement that","max_batches (int): The maximum number of batches of pipelines to search. Parameters max_time,","Disabled by default in non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already been","or not to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable):","batch + each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) +","if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train","in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before","IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and","The maximum number of batches of pipelines to search. Parameters max_time, and max_iterations","of length [n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes): type","p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set","`search()` errors and raises an Exception. Callback function takes three positional parameters: the","Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations","objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names()","targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str,","pipeline): \"\"\"Fits and evaluates a given pipeline then adds the results to the","additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in","to run ensembling.\") else: logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\") self.max_iterations = (1","series problems, values should be passed in for the gap and max_delay variables.","+ 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if","search. problem_configuration (dict, None): Additional parameters needed to configure the search. For example,","isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\"Parameter max_time must be a float, int,","try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for","Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string()","the best pipeline and parameters found during automl search. If `train_best_pipeline` is set","indicating the pipelines allowed in the search. The default of None indicates all","= { \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\":","+ each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines)","metalearners. Only used when ensembling is True. Must be between 0 and 1,","\"\"\" leading_char = \"\\n\" start_of_loop = time.time() while True: choice = input(leading_char +","if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if","self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output","will not run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\") else:","that error out during scoring will not be included in the dictionary but","and (not isinstance(patience, int) or patience < 0): raise ValueError(\"patience value must be","or None): Non-negative integer describing level of parallelism used for pipelines. None and","\"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\",","early stopping is disabled. Defaults to None. tolerance (float): Minimum percentage difference to","rankings with the requirement that automl search has been run. Arguments: pipeline (PipelineBase):","def best_pipeline(self): \"\"\"Returns a trained instance of the best pipeline and parameters found","False, returns an untrained pipeline instance. \"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl search","= self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling,","and rank pipelines, but not for optimizing each pipeline during fit-time. When set","pipeline before returning it. Defaults to True. pipeline_parameters (dict): A dict of the","pipeline.parameters, self) desc = f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN -","the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv =","or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed","search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for estimator in","automl rankings with the requirement that automl search has been run. Arguments: pipeline","callable(function): return function.__name__ else: return None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\"","`regression` depending on the problem type. Note that if allowed_pipelines is provided, this","(sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner class","the ID of a pipeline training result, returns an untrained instance of the","\"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not found in","\") elif self.max_iterations is not None: logger.info(\"Searching up to %s pipelines. \" %","X_train is None: raise ValueError('Must specify training data as a 2d array using","path Arguments: file_path (str): location to find file to load Returns: AutoSearchBase object","used. ensembling (boolean): If True, runs ensembling in a separate batch after every","= 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without","choice = input(leading_char + \"Do you really want to exit search (y/n)? \").strip().lower()","is an integer, then the time will be in seconds. For strings, time","dictionary but the exception and stacktrace will be displayed in the log. \"\"\"","self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2):","if no. \"\"\" if self._interrupted: return False # for add_to_rankings if self._searched: return","'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and score pipelines.\") if self.max_batches is not None:","searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary` to","a warning if variance is higher than specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv","p in required_parameters): raise ValueError(\"user_parameters must be a dict containing values for at","logger.warning(f\"High coefficient of variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may","{o.name: o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float,","will be empty so we will return # nan for the base score.","example, in time series problems, values should be passed in for the gap","tolerance (float): Minimum percentage difference to qualify as score improvement for early stopping.","cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info]","else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1],","if c in [\"# Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean =","is None: raise ValueError('Must specify training data target values as a 1d vector","{self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience:","loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except","check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily","self._pipelines_per_batch} iterations after that.\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True","Re-initialize AutoMLSearch to search again.\") return # don't show iteration plot outside of","sure it is different from the current best pipeline before training and thresholding\"\"\"","evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils","run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name']","strings, time can be specified as seconds, minutes, or hours. patience (int): Number","Defaults to False. Returns: Description of specified pipeline. Includes information such as type","a list of pipelines on the training data. This can be helpful for","from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils","data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective =","self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params","specify training data as a 2d array using the X_train argument') if y_train","not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline","ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling","= \"\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results:","# nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] =","count towards the time budget (if set) time_in_loop = time.time() - start_of_loop self._start","return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check for","or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective","= -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass","< 1): raise ValueError(f\"Ensembling split size must be between 0 and 1 exclusive,","self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out","PipelineNotFoundError(\"automl search must be run before selecting `best_pipeline`.\") return self._best_pipeline def save(self, file_path,","for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]:","base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline,","def _num_pipelines(self): \"\"\"Return the number of pipeline evaluations which have been made Returns:","file to load Returns: AutoSearchBase object \"\"\" with open(file_path, 'rb') as f: return","ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\",","Parameters max_time, and max_iterations have precedence over stopping the search. problem_configuration (dict, None):","pipeline_parameters (dict): A dict of the parameters used to initalize a pipeline with.","(not isinstance(patience, int) or patience < 0): raise ValueError(\"patience value must be a","= False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for %s. \" % self.objective.name) logger.info(\"{}","* (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 +","self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def _validate_problem_configuration(self,","\"\"\" if self._interrupted: return False # for add_to_rankings if self._searched: return True #","pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and logs a warning if variance is","cross-validation scores and logs a warning if variance is higher than specified threshhold.\"\"\"","TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from","file_path (str): location to find file to load Returns: AutoSearchBase object \"\"\" with","n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import","'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try:","loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings =","all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self,","If it is an integer, then the time will be in seconds. For","raise ValueError(f\"Ensembling split size must be between 0 and 1 exclusive, received {_ensembling_split_size}\")","c in all_objective_scores: if c in [\"# Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\")","raise ValueError(\"patience value must be a positive integer. Received {} instead\".format(patience)) if tolerance","self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p","stopping the search. problem_configuration (dict, None): Additional parameters needed to configure the search.","sorted(['\\t{}'.format(o.name) for o in obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__","feature types, either numerical or categorical. Categorical features will automatically be encoded show_iteration_plot","= ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION:","maximum number of batches of pipelines to search. Parameters max_time, and max_iterations have","self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)]))","terminate early \"\"\" leading_char = \"\\n\" start_of_loop = time.time() while True: choice =","to score on. Will override default objectives for problem type if not empty.","[], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None","import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle,","logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self): \"\"\"Finds","for the gap and max_delay variables. train_best_pipeline (boolean): Whether or not to train","the parameters used to initalize a pipeline with. _ensembling_split_size (float): The amount of","self._best_pipeline: raise PipelineNotFoundError(\"automl search must be run before selecting `best_pipeline`.\") return self._best_pipeline def","when `search()` errors and raises an Exception. Callback function takes three positional parameters:","baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class =","from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique,","first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations","0: raise ValueError(f\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\") self.max_time =","_baseline_cv_scores will be empty so we will return # nan for the base","after every allowed pipeline class has been iterated over. If the number of","def load(file_path): \"\"\"Loads AutoML object at file path Arguments: file_path (str): location to","= allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output =","max_batches is too small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch","None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter","self.tolerance is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement =","traceback, and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able","return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name]","self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback =","# Run at least one pipeline for every search num_pipelines = self._num_pipelines() if","pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and logs a warning","= tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback","stopping is disabled. Defaults to None. tolerance (float): Minimum percentage difference to qualify","Callback function takes three positional parameters: The pipeline class, the pipeline parameters, and","return objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o","if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices]","search must be run before selecting `best_pipeline`.\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL):","plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None,","used for pipelines. None and 1 are equivalent. If set to -1, all","\", \".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot =","were searched. \"\"\" return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a pandas.DataFrame with scoring","f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives","target values as a 1d vector using the y_train argument') try: self.problem_type =","thresholds for: {}\".format(self.objective)) logger.info(\"Total training time (including CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger,","will be in seconds. For strings, time can be specified as seconds, minutes,","1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping search early...\".format(self.patience)) return","False logger.warning(f\"Ensembling is set to True, but max_batches is too small, so ensembling","logger.info(f\"Searching up to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \") elif","self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches =","pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a given pipeline then adds the","def full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from all pipelines searched\"\"\" ascending","ValueError(f\"Parameter max_time must be None or non-negative. Received {max_time}.\") if max_batches is not","if self.objective.greater_is_better: ascending = False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\",","ensembling will not run.\") run_ensembling = False if run_ensembling and self.max_iterations is not","self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time Series Baseline Estimator\": {\"gap\":","= train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train =","training time (including CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores","logger.info(\"Total training time (including CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\")","from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from","in problem_configuration for p in required_parameters): raise ValueError(\"user_parameters must be a dict containing","convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch","LogLossMulticlass for multiclass classification problems, and - R2 for regression problems. max_iterations (int):","it. Defaults to True. pipeline_parameters (dict): A dict of the parameters used to","start a new pipeline search after the duration has elapsed. If it is","self.allowed_pipelines == []: raise ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling","(float): The amount of the training data we'll set aside for training ensemble","max_time must be a float, int, string or None. Received {type(max_time)} with value","search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features].","num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping search early...\".format(self.patience)) return False return","defaultdict import cloudpickle import numpy as np import pandas as pd import woodwork","since Sphinx does not work well with instance attributes. plot = PipelineSearchPlots def","This will not start a new pipeline search after the duration has elapsed.","self.patience = patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {},","the duration has elapsed. If it is an integer, then the time will","input training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target","evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class","self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\"No allowed","( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import","for training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List of","current AutoML batch produced a score of np.nan on the primary objective {self.objective}.\")","current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if","file path Arguments: file_path (str): location to find file to load Returns: AutoSearchBase","self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines)","not empty. random_seed (int): Seed for the random number generator. Defaults to 0.","difference to qualify as score improvement for early stopping. Only applicable if patience","f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML object at file path","else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping","to True, but max_iterations is too small, so ensembling will not run. Set","y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return the number of pipeline","= 40 # Necessary for \"Plotting\" documentation, since Sphinx does not work well","regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback =","{ensemble_nth_batch + 1} to run ensembling.\") else: logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\")","not run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\") else: logger.info(f\"Ensembling","and stacktrace will be displayed in the log. \"\"\" return self._engine.score_batch(pipelines, X_holdout, y_holdout,","primary objective {self.objective}.\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc =","self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change =","cross validation scores. {pipeline_name} may not perform as estimated on unseen data.\") return","full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all():","== 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if","field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives","estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\"No allowed pipelines to search\")","= best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def","Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str,","have been made Returns: int: the number of pipeline evaluations made in the","to the automl rankings with the requirement that automl search has been run.","Change `binary` to `multiclass` or `regression` depending on the problem type. Note that","not to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function","self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not found in automl results\") pipeline_class","value must be a positive integer. Received {} instead\".format(patience)) if tolerance and (tolerance","time.time() - start_of_loop self._start += time_in_loop return False else: leading_char = \"\" def","well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto',","= False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings","for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))):","instance. \"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl search must be run before selecting","scoring will not be included in the dictionary but the exception and stacktrace","return True # check max_time and max_iterations elapsed = time.time() - self._start if","dict with results from each pipeline, and `search_order`: a list describing the order","of unique pipelines to search over per batch is one, ensembling will not","True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score']","not None: logger.info(\"Will stop searching for new pipelines after %d seconds.\\n\" % self.max_time)","None: logger.info(f\"Searching up to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \")","iterations without improvement. Stopping search early...\".format(self.patience)) return False return True def _validate_problem_type(self): for","score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores)","the gap and max_delay variables. train_best_pipeline (boolean): Whether or not to train the","`multiclass` or `regression` depending on the problem type. Note that if allowed_pipelines is","AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40 # Necessary for \"Plotting\" documentation, since","f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines:","again.\") return # don't show iteration plot outside of a jupyter notebook if","estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\"No","self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from all","(list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by","one. The first batch will train a baseline pipline + one of each","[] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try: self.plot","evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family","def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {}","int) or patience < 0): raise ValueError(\"patience value must be a positive integer.","\"\"\"Given the original stopping criterion and current state, should the search continue? Returns:","out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except","score of np.nan on the primary objective {self.objective}.\") self.search_duration = time.time() - self._start","gap and max_delay \" f\"parameters. Received {problem_configuration}.\") return problem_configuration or {} def _handle_keyboard_interrupt(self):","positional parameters: the Exception raised, the traceback, and the AutoMLSearch object. Must also","f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\"","in this loop does not count towards the time budget (if set) time_in_loop","a float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\") if isinstance(max_time,","import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import","type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str,","TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import","random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not","= (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) +","{self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback:","for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field","that pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase:","Includes information such as type of pipeline components, problem, training time, cross validation,","self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration)","with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds","all objectives # but also fields like \"# Training\" and \"# Testing\", so","self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training time","Seed for the random number generator. Defaults to 0. n_jobs (int or None):","run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_iterations is too small,","self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices","will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling will","evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger,","self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores =","_num_pipelines(self): \"\"\"Return the number of pipeline evaluations which have been made Returns: int:","to True. start_iteration_callback (callable): Function called before each pipeline training iteration. Callback function","all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] =","return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters =","input(leading_char + \"Do you really want to exit search (y/n)? \").strip().lower() if choice","pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not found in automl results\") pipeline_class = pipeline_results.get('pipeline_class')","the results to the automl rankings with the requirement that automl search has","split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import","untrained pipeline instance. Returns: PipelineBase: A trained instance of the best pipeline and","ValueError(\"patience value must be a positive integer. Received {} instead\".format(patience)) if tolerance and","logger.info(\"Optimizing for %s. \" % self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better else","be None or non-negative. Received {max_time}.\") if max_batches is not None and max_batches","pipeline. Includes information such as type of pipeline components, problem, training time, cross","pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter:","{}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to the data. This is",">= self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping search early...\".format(self.patience)) return False return True",") logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40 #","pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv,","not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc","or []: if pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline {} is not compatible","and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if","abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False):","set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params =","\"\"\" if X_train is None: raise ValueError('Must specify training data as a 2d","new pipelines after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed model families: %s\\n\" % \",","be a dict containing values for at least the gap and max_delay \"","ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine","data. This can be helpful for training pipelines once the search is complete.","pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary for binary classification","used when ensembling is True. Must be between 0 and 1, exclusive. Defaults","cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and logs a warning if variance is higher","{ 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs =","pipeline in the rankings If self._best_pipeline already exists, check to make sure it","\"\"\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline","None. tolerance (float): Minimum percentage difference to qualify as score improvement for early","not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\"Using default","pipeline and parameters found during automl search. If `train_best_pipeline` is set to False,","log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN","np import pandas as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator","to initalize a pipeline with. _ensembling_split_size (float): The amount of the training data","log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for %s. \" % self.objective.name) logger.info(\"{} score is","pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)):","describing level of parallelism used for pipelines. None and 1 are equivalent. If","ensembling (boolean): If True, runs ensembling in a separate batch after every allowed","allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The","logger.info(\"\\n\\n{} iterations without improvement. Stopping search early...\".format(self.patience)) return False return True def _validate_problem_type(self):","self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto':","rankings_desc = \"\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch","except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = []","pipeline evaluations which have been made Returns: int: the number of pipeline evaluations","ValueError(\"tolerance value must be a float between 0.0 and 1.0 inclusive. Received {}","(bool): If True, return dictionary of information about pipeline. Defaults to False. Returns:","[])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback:","0: raise ValueError(f\"Parameter max_batches must be None or non-negative. Received {max_batches}.\") if max_iterations","= self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train =","\"\"\" return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from","with scoring results from all pipelines searched\"\"\" ascending = True if self.objective.greater_is_better: ascending","PipelineBase: untrained pipeline instance associated with the provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id)","percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"]","retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided ID \"\"\" pipeline_results","= pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is None:","batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def","pipelines in the current AutoML batch produced a score of np.nan on the","automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance","the specified pipeline initialized with the parameters used to train that pipeline during","y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for","is the first pipeline fit during search. \"\"\" if self.problem_type == ProblemTypes.BINARY: baseline","data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout","string or None. Received {type(max_time)} with value {str(max_time)}..\") if isinstance(max_time, (int, float)) and","raise ValueError(f\"Parameter max_time must be None or non-negative. Received {max_time}.\") if max_batches is","self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline","bool: True if yes, False if no. \"\"\" if self._interrupted: return False #","over. If the number of unique pipelines to search over per batch is","than specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold)","the random number generator. Defaults to 0. n_jobs (int or None): Non-negative integer","be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number","So that the time in this loop does not count towards the time","def rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from the highest-scoring set of","= f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type","and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations,","gap, \"max_delay\": max_delay}, \"Time Series Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod","= self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break","None and max_iterations < 0: raise ValueError(f\"Parameter max_iterations must be None or non-negative.","= f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\"","time_in_loop return False else: leading_char = \"\" def search(self, show_iteration_plot=True): \"\"\"Find the best","self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and score pipelines.\") if self.max_batches is","= patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order':","object. error_callback (callable): Function called when `search()` errors and raises an Exception. Callback","return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the ID of a pipeline training result,","`log_error_callback`. additional_objectives (list): Custom set of objectives to score on. Will override default","is not None: logger.info(\"Will stop searching for new pipelines after %d seconds.\\n\" %","self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and","propose and rank pipelines, but not for optimizing each pipeline during fit-time. When","self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines","int: the number of pipeline evaluations made in the search \"\"\" return len(self._results['pipeline_results'])","search again.\") return # don't show iteration plot outside of a jupyter notebook","per batch is one, ensembling will not run. Defaults to False. max_batches (int):","from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from","raise ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and","of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change","significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score >","float, str, type(None))): raise TypeError(f\"Parameter max_time must be a float, int, string or","is not None: logger.info(\"Searching up to %s pipelines. \" % self.max_iterations) if self.max_time","PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def","pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots","`train_best_pipeline` is set to False, returns an untrained pipeline instance. Returns: PipelineBase: A","then the time will be in seconds. For strings, time can be specified","(int): pipeline to describe return_dict (bool): If True, return dictionary of information about","be None or non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else","float)) and max_time < 0: raise ValueError(f\"Parameter max_time must be None or non-negative.","every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\") if self.max_batches and self.max_iterations is None:","is None: raise PipelineNotFoundError(\"Pipeline not found in automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters","== 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])):","components, problem, training time, cross validation, etc. \"\"\" if pipeline_id not in self._results['pipeline_results']:","(self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family ==","is not None and max_iterations < 0: raise ValueError(f\"Parameter max_iterations must be None","0.0 and 1.0 inclusive. Received {} instead\".format(tolerance)) self.patience = patience self.tolerance = tolerance","self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters","max_delay \" f\"parameters. Received {problem_configuration}.\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a","maps to a dictionary of scores. Note that the any pipelines that error","return_dict: return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a given pipeline then","1)) if run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\"Ensembling split","update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40","set to True, but max_batches is too small, so ensembling will not run.","[]: raise ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling","= full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in","[n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes): type of supervised","float between 0.0 and 1.0 inclusive. Received {} instead\".format(tolerance)) self.patience = patience self.tolerance","= pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property","if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is not compatible with a {}","tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or","etc. \"\"\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline =","set, then max_iterations will default to max_iterations of 5. max_time (int, str): Maximum","The pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function","raise PipelineNotFoundError(\"Pipeline not found in automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters')","for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all","full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from all pipelines searched\"\"\" ascending =","helpful for training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List","called after each pipeline training iteration. Callback function takes three positional parameters: A","self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters =","pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time Series Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}})","Series Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores","pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return","SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter","mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of var\", c] = std / mean","first batch will train a baseline pipline + one of each pipeline family","with a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) #","of feature types, either numerical or categorical. Categorical features will automatically be encoded","X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed)","obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called","n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time']","must be None or non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str)","return False else: leading_char = \"\" def search(self, show_iteration_plot=True): \"\"\"Find the best pipeline","found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input","additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated","if self._searched: logger.info(\"AutoMLSearch.search() has already been run and will not run again on","== 0: run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_batches is","# they are not scores if field in objective_name_to_class: scores[field] += value return","run ensembling.\") else: logger.info(f\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines)","def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives:","time, cross validation, etc. \"\"\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not","results from all pipelines searched\"\"\" ascending = True if self.objective.greater_is_better: ascending = False","search\") logger.info(\"Optimizing for %s. \" % self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better","self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification pipeline thresholds for:","% pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]]","If None, early stopping is disabled. Defaults to None. tolerance (float): Minimum percentage","\"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl search must be run before selecting `best_pipeline`.\")","= self.X_train y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train","pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training","and the AutoMLSearch object. error_callback (callable): Function called when `search()` errors and raises","set of objectives to score on. Will override default objectives for problem type","max_time (int, str): Maximum time to search for pipelines. This will not start","additional_objectives (list): Custom set of objectives to score on. Will override default objectives","None. Received {type(max_time)} with value {str(max_time)}..\") if isinstance(max_time, (int, float)) and max_time <","in seconds. For strings, time can be specified as seconds, minutes, or hours.","{} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification","import time from collections import defaultdict import cloudpickle import numpy as np import","= None self._interrupted = False if self.allowed_pipelines is None: logger.info(\"Generating pipelines to search","# The 'all_objective_scores' field contains scores for all objectives # but also fields","None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt:","(1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration:","for: {}\".format(self.objective)) logger.info(\"Total training time (including CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross","\"\"\"Class that allows access to a copy of the results from `automl_search`. Returns:","class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize","(callable): Function called before each pipeline training iteration. Callback function takes three positional","\"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif choice == \"n\": # So that the","logger.warning(\"Ensembling is set to True, but the number of unique pipelines is one,","function takes three positional parameters: A dictionary containing the training results for the","of information about pipeline. Defaults to False. Returns: Description of specified pipeline. Includes","elif choice == \"n\": # So that the time in this loop does","self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train,","== self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train","Jupyter notebook. Disabled by default in non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has","{} instead\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results = {","percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family","baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap","rankings If self._best_pipeline already exists, check to make sure it is different from","from collections import defaultdict import cloudpickle import numpy as np import pandas as","is None or parameters is None: raise PipelineNotFoundError(\"Pipeline class or parameters not found","to save file pickle_protocol (int): the pickle data stream format. Returns: None \"\"\"","- R2 for regression problems. max_iterations (int): Maximum number of iterations to search.","if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\"Pipeline class or parameters","a list describing the order the pipelines were searched. \"\"\" return copy.deepcopy(self._results) @property","= ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False)","a baseline pipline + one of each pipeline family allowed in the search.","problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue,","to load Returns: AutoSearchBase object \"\"\" with open(file_path, 'rb') as f: return cloudpickle.load(f)","len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores'","(n_cpus + 1 + n_jobs) are used. ensembling (boolean): If True, runs ensembling","to search. The default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\")","def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc = ( f\"{handle_problem_types(self.problem_type).name}","_ensembling_split_size < 1): raise ValueError(f\"Ensembling split size must be between 0 and 1","{ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline","fit during search. \"\"\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type","SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary pipeline threshold. Defaults","Returns: bool: If True, search should terminate early \"\"\" leading_char = \"\\n\" start_of_loop","None indicates all pipelines for this problem type are allowed. Setting this field","dictionary of information about pipeline. Defaults to False. Returns: Description of specified pipeline.","Will override default objectives for problem type if not empty. random_seed (int): Seed","elapsed = time.time() - self._start if self.max_time and elapsed >= self.max_time: return False","self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False","field in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for objective,","separate batch after every allowed pipeline class has been iterated over. If the","= list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline in","ensembling will not run. Defaults to False. max_batches (int): The maximum number of","self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set to True, but","search.\"\"\" _MAX_NAME_LEN = 40 # Necessary for \"Plotting\" documentation, since Sphinx does not","objectives # but also fields like \"# Training\" and \"# Testing\", so we","= self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices]","(str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full","self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores)","an untrained_pipeline containing the parameters used during training, and the AutoMLSearch object. error_callback","with the parameters used to train that pipeline during automl search. Arguments: pipeline_id","families to search. The default of None searches over all model families. Run","the same instance. Re-initialize AutoMLSearch to search again.\") return # don't show iteration","def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a given pipeline then adds the results","run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\") else: logger.info(f\"Ensembling will","the search. problem_configuration (dict, None): Additional parameters needed to configure the search. For","the Exception raised, the traceback, and the AutoMLSearch object. Must also accepts kwargs,","{threshold}) within cross validation scores. {pipeline_name} may not perform as estimated on unseen","mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format,","return '\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc =","max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches:","like \"# Training\" and \"# Testing\", so we want to exclude them since","desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm","from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import (","self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches","run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set to True, but the number","self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and logs","best pipeline before returning it. Defaults to True. pipeline_parameters (dict): A dict of","automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or","- start_of_loop self._start += time_in_loop return False else: leading_char = \"\" def search(self,","not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True)","(int): the pickle data stream format. Returns: None \"\"\" with open(file_path, 'wb') as","Received {} instead\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results =","optimizing each pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary for","before training and thresholding\"\"\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if","but also fields like \"# Training\" and \"# Testing\", so we want to","= None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except","KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids","Received {} instead\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance < 0.0):","False if no. \"\"\" if self._interrupted: return False # for add_to_rankings if self._searched:","pipelines to search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for","== ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY:","want to stop the search. Returns: bool: If True, search should terminate early","Returns: AutoSearchBase object \"\"\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self,","keep=\"first\") @property def full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from all pipelines","_ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data","Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner class to","configure the search. For example, in time series problems, values should be passed","{} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None: logger.info(\"Generating pipelines","= pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data,","describing the order the pipelines were searched. \"\"\" return copy.deepcopy(self._results) @property def rankings(self):","the problem type. Note that if allowed_pipelines is provided, this parameter will be","for a total of {self.max_iterations} pipelines. \") elif self.max_iterations is not None: logger.info(\"Searching","can be specified as seconds, minutes, or hours. patience (int): Number of iterations","if max_batches is not None and max_batches < 0: raise ValueError(f\"Parameter max_batches must","stopping criterion and current state, should the search continue? Returns: bool: True if","is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In","else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if","# but also fields like \"# Training\" and \"# Testing\", so we want","automl search has been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate.","problems, and - R2 for regression problems. max_iterations (int): Maximum number of iterations","None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the","else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None: logger.info(\"Generating","[] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try:","None: raise PipelineNotFoundError(\"Pipeline not found in automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters =","problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The","are used. For n_jobs below -1, (n_cpus + 1 + n_jobs) are used.","True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective","cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores","pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\"# Training\",","self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try:","_validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise","@staticmethod def load(file_path): \"\"\"Loads AutoML object at file path Arguments: file_path (str): location","= ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices","self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc,","'\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc = (","the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training iteration. Callback","True): Shows an iteration vs. score plot in Jupyter notebook. Disabled by default","import cloudpickle import numpy as np import pandas as pd import woodwork as","self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN","\\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed","= IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def","is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and score pipelines.\")","== parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class that allows access to","self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance","from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from","or non-negative. Received {max_time}.\") if max_batches is not None and max_batches < 0:","max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time","else: return None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max","y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self):","batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective","X_holdout, y_holdout, objectives): \"\"\"Score a list of pipelines on the given holdout data.","train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring.","training iteration. Callback function takes three positional parameters: A dictionary containing the training","of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt:","(int): Seed for the random number generator. Defaults to 0. n_jobs (int or","exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to train for","test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback,","import numpy as np import pandas as pd import woodwork as ww from","Returns: Description of specified pipeline. Includes information such as type of pipeline components,","as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import","self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start,","pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples,","5. max_time (int, str): Maximum time to search for pipelines. This will not","patience is not None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase","up to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \") elif self.max_iterations","= self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched","1} to run ensembling.\") else: logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\") self.max_iterations =","best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices],","score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if score_improved","threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv:","from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types","one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback","gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay},","scores if field in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds","scores for all objectives # but also fields like \"# Training\" and \"#","baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type ==","is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results'])","run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_batches is too small,","pipeline training result, returns an untrained instance of the specified pipeline initialized with","class or parameters not found in automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self,","to the user asking if they want to stop the search. Returns: bool:","file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file path Arguments: file_path (str): location to","seconds, minutes, or hours. patience (int): Number of iterations without improvement to stop","ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class","function takes three positional parameters: The pipeline class, the pipeline parameters, and the","None: raise ValueError('Must specify training data target values as a 1d vector using","for optimizing each pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary","not None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating","class has been iterated over. If the number of unique pipelines to search","values should be passed in for the gap and max_delay variables. train_best_pipeline (boolean):","if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] =","PipelineBase]: Dictionary keyed by pipeline name that maps to the fitted pipeline. Note","(int): Number of iterations without improvement to stop search early. Must be positive.","The amount of the training data we'll set aside for training ensemble metalearners.","* self._pipelines_per_batch} iterations after that.\") if self.max_batches and self.max_iterations is None: self.show_batch_output =","if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if","regression problems. max_iterations (int): Maximum number of iterations to search. If max_iterations and","AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective","rank pipelines, but not for optimizing each pipeline during fit-time. When set to","and max_delay variables. train_best_pipeline (boolean): Whether or not to train the best pipeline","_get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\"","search over per batch is one, ensembling will not run. Defaults to False.","def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a list of pipelines on the","default objectives for problem type if not empty. random_seed (int): Seed for the","must be None or non-negative. Received {max_batches}.\") if max_iterations is not None and","numerical or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows","to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary pipeline threshold.","search _baseline_cv_scores will be empty so we will return # nan for the","evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import","problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification pipeline","for binary classification problems, - LogLossMulticlass for multiclass classification problems, and - R2","pipelines. \") elif self.max_iterations is not None: logger.info(\"Searching up to %s pipelines. \"","time.time() - self._start if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations","for all objectives # but also fields like \"# Training\" and \"# Testing\",","after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed model families: %s\\n\" % \", \".join([model.value for","if our main objective is part of default set of objectives for problem_type,","return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True)","we'll set aside for training ensemble metalearners. Only used when ensembling is True.","already exists, check to make sure it is different from the current best","list of feature types, either numerical or categorical. Categorical features will automatically be","self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train","= False if run_ensembling and self.max_iterations is not None: # Baseline + first","{first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\") if self.max_batches and","Additional parameters needed to configure the search. For example, in time series problems,","raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or","for supervised learning tasks. problem_type (str or ProblemTypes): type of supervised learning problem.","not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE:","f\"parameters. Received {problem_configuration}.\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to","Arguments: file_path (str): location to save file pickle_protocol (int): the pickle data stream","is set to True, but the number of unique pipelines is one, so","= self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if","again on the same instance. Re-initialize AutoMLSearch to search again.\") return # don't","in the search. The default of None indicates all pipelines for this problem","will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to","None: logger.info(\"Will stop searching for new pipelines after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed","self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class that allows access to a copy of","but max_batches is too small, so ensembling will not run. Set max_batches >=","from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed,","default of None indicates all pipelines for this problem type are allowed. Setting","ObjectiveBase): The objective to optimize for. Used to propose and rank pipelines, but","batch after every allowed pipeline class has been iterated over. If the number","{} problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type:","plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \"\"\" if self._searched:","pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self)","len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set to True, but the number of unique","+ self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\"Parameter max_time must","ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if","not found in automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe","raise ValueError('Must specify training data as a 2d array using the X_train argument')","loop does not count towards the time budget (if set) time_in_loop = time.time()","show iteration plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython except","train and score pipelines.\") if self.max_batches is not None: logger.info(f\"Searching up to {self.max_batches}","hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline","obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is not compatible","on. Will override default objectives for problem type if not empty. random_seed (int):","model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score","[get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives","self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores","if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\"Parameter max_time must be a","self.max_iterations is not None: logger.info(\"Searching up to %s pipelines. \" % self.max_iterations) if","batches.\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 -","random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback)","Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom set of objectives","run_ensembling = False if run_ensembling and self.max_iterations is not None: # Baseline +","cloudpickle import numpy as np import pandas as pd import woodwork as ww","= len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The","given pipeline then adds the results to the automl rankings with the requirement","**self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs,","AutoML object at file path Arguments: file_path (str): location to find file to","Dictionary keyed by pipeline name that maps to a dictionary of scores. Note","field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families","search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max","over per batch is one, ensembling will not run. Defaults to False. max_batches","pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1:","list describing the order the pipelines were searched. \"\"\" return copy.deepcopy(self._results) @property def","log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective","< _ensembling_split_size < 1): raise ValueError(f\"Ensembling split size must be between 0 and","Stopping search early...\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in self.additional_objectives:","self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch finished after","StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted","pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called","logs a warning if variance is higher than specified threshhold.\"\"\" pipeline_name = pipeline.name","= mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event","multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback","self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a list of pipelines on","logger.warning(\"Unable to import plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families =","time will be in seconds. For strings, time can be specified as seconds,","time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN =","\"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already been run and will not run again","number of pipelines to train for every batch after the first one. The","= start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter","data of length [n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes):","ValueError(f\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if","and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\") if self.max_batches and self.max_iterations is","= all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c] = std","and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training","%s pipelines. \" % self.max_iterations) if self.max_time is not None: logger.info(\"Will stop searching","pipeline to describe return_dict (bool): If True, return dictionary of information about pipeline.","return False # for add_to_rankings if self._searched: return True # Run at least","% self.max_iterations) if self.max_time is not None: logger.info(\"Will stop searching for new pipelines","'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification","override default objectives for problem type if not empty. random_seed (int): Seed for","None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for","train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train)","self.X_train y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train =","max_iterations (int): Maximum number of iterations to search. If max_iterations and max_time is","= False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration:","one, ensembling will not run. Defaults to False. max_batches (int): The maximum number","and max_iterations have precedence over stopping the search. problem_configuration (dict, None): Additional parameters","{problem_configuration}.\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the user","containing values for at least the gap and max_delay \" f\"parameters. Received {problem_configuration}.\")","= best_pipeline def _num_pipelines(self): \"\"\"Return the number of pipeline evaluations which have been","set to {[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration):","provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not","pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before each pipeline training","will not start a new pipeline search after the duration has elapsed. If","score on. Will override default objectives for problem type if not empty. random_seed","self.allowed_pipelines is None: logger.info(\"Generating pipelines to search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators","for. Used to propose and rank pipelines, but not for optimizing each pipeline","is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives","\"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape","and max_time < 0: raise ValueError(f\"Parameter max_time must be None or non-negative. Received","= (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations <","+ (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size","{self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc","not problem_configuration or not all(p in problem_configuration for p in required_parameters): raise ValueError(\"user_parameters","for add_to_rankings if self._searched: return True # Run at least one pipeline for","time from collections import defaultdict import cloudpickle import numpy as np import pandas","error out during training will not be included in the dictionary but the","from all pipelines searched\"\"\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols","coefficient of variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may not","def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the user asking if they want to","None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches =","self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines with IDs:","in the search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original stopping criterion","takes three positional parameters: the Exception raised, the traceback, and the AutoMLSearch object.","def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not","in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective)","max_iterations < 0: raise ValueError(f\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\")","stacktrace will be displayed in the log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines,","f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\"","num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches","patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [],","an untrained pipeline instance. \"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl search must be","self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable","Arguments: feature_types (list, optional): list of feature types, either numerical or categorical. Categorical","obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is not compatible with a {} problem.\".format(obj.name, self.problem_type.value))","pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if","made Returns: int: the number of pipeline evaluations made in the search \"\"\"","of pipelines to train for every batch after the first one. The first","with the requirement that automl search has been run. Arguments: pipeline (PipelineBase): pipeline","the training data. This can be helpful for training pipelines once the search","rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \"\"\"Returns a trained","default to max_iterations of 5. max_time (int, str): Maximum time to search for","is not set, then max_iterations will default to max_iterations of 5. max_time (int,","False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML","batch produced a score of np.nan on the primary objective {self.objective}.\") self.search_duration =","to True, but the number of unique pipelines is one, so ensembling will","if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline","objective {} is not compatible with a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline in","None: raise PipelineNotFoundError(\"Pipeline class or parameters not found in automl results\") return pipeline_class(parameters,","PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines =","( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from","\"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters ==","- 1)) if run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\"Ensembling","self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\"Parameter max_time must be","pipeline fit during search. \"\"\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif","len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\":","Pipeline search.\"\"\" _MAX_NAME_LEN = 40 # Necessary for \"Plotting\" documentation, since Sphinx does","len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches ==","None self._interrupted = False if self.allowed_pipelines is None: logger.info(\"Generating pipelines to search over...\")","False return True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise","to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling","= 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number =","= self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise","train and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']:","and stacktrace will be displayed in the log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self,","not None: logger.info(\"Searching up to %s pipelines. \" % self.max_iterations) if self.max_time is","in a separate batch after every allowed pipeline class has been iterated over.","random number generator. Defaults to 0. n_jobs (int or None): Non-negative integer describing","See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective to optimize","be a positive integer. Received {} instead\".format(patience)) if tolerance and (tolerance > 1.0","1.0 or tolerance < 0.0): raise ValueError(\"tolerance value must be a float between","not to train the best pipeline before returning it. Defaults to True. pipeline_parameters","= full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the current AutoML","in all_objective_scores: if c in [\"# Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue","of max_batches=1.\\n\") if patience and (not isinstance(patience, int) or patience < 0): raise","non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already been run and will not","show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing","self._searched: return True # Run at least one pipeline for every search num_pipelines","or parameters not found in automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id,","the current best pipeline before training and thresholding\"\"\" if len(self.rankings) == 0: return","search for pipelines. This will not start a new pipeline search after the","pipeline class has been iterated over. If the number of unique pipelines to","a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type !=","the search continue? Returns: bool: True if yes, False if no. \"\"\" if","Function called before each pipeline training iteration. Callback function takes three positional parameters:","Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if not","results from each pipeline, and `search_order`: a list describing the order the pipelines","jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \"Beginning","automl search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance.","desc = f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] +","is higher than specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean())","- LogLossMulticlass for multiclass classification problems, and - R2 for regression problems. max_iterations","call `log_error_callback`. additional_objectives (list): Custom set of objectives to score on. Will override","to train that pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve","size must be between 0 and 1 exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0]))","specified pipeline. Includes information such as type of pipeline components, problem, training time,","problem type. Note that if allowed_pipelines is provided, this parameter will be ignored.","= evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE","== pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline()","will not be included in the dictionary but the exception and stacktrace will","= infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True,","ensembling.\") else: logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\") self.max_iterations = (1 + len(self.allowed_pipelines)","Defaults to 0. n_jobs (int or None): Non-negative integer describing level of parallelism","= SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p","self._best_pipeline already exists, check to make sure it is different from the current","get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline","objectives for problem type if not empty. random_seed (int): Seed for the random","first batch + each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines)","- num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch *","in automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline","for pipelines. None and 1 are equivalent. If set to -1, all CPUs","evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline","= (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False","is set to True, but max_batches is too small, so ensembling will not","f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'],","problem_configuration or not all(p in problem_configuration for p in required_parameters): raise ValueError(\"user_parameters must","parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class that allows access to a","pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start","first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_iterations is too","None: logger.info(\"Generating pipelines to search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to","len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN)","= curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience:","binary classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training time (including CV): %.1f seconds\"","= all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of var\",","add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a given pipeline then adds the results to","+= 1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without improvement. Stopping search early...\".format(self.patience))","= True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted =","TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline =","def train_pipelines(self, pipelines): \"\"\"Train a list of pipelines on the training data. This","_, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist()","n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search","raise PipelineNotFoundError(\"automl search must be run before selecting `best_pipeline`.\") return self._best_pipeline def save(self,","= \"\\n\" start_of_loop = time.time() while True: choice = input(leading_char + \"Do you","pipelines that error out during scoring will not be included in the dictionary","If `train_best_pipeline` is set to False, returns an untrained pipeline instance. \"\"\" if","- LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification problems, and","model families to search. The default of None searches over all model families.","= desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name","pipeline {} is not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a","best_pipeline(self): \"\"\"Returns a trained instance of the best pipeline and parameters found during","not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\"Using default limit of max_batches=1.\\n\")","True, return dictionary of information about pipeline. Defaults to False. Returns: Description of","of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase):","evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary` to `multiclass` or `regression` depending on the","ValueError(f\"{objective.name.lower()} is not allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all","= {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline","search after the duration has elapsed. If it is an integer, then the","The objective to optimize for. Used to propose and rank pipelines, but not","None else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None:","-1, all CPUs are used. For n_jobs below -1, (n_cpus + 1 +","is called before search _baseline_cv_scores will be empty so we will return #","# don't show iteration plot outside of a jupyter notebook if show_iteration_plot: try:","high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the ID of a pipeline training result, returns","default set of objectives for problem_type, remove it existing_main_objective = next((obj for obj","Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines","during training, and the AutoMLSearch object. error_callback (callable): Function called when `search()` errors","( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline,","Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features.","ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline:","no. \"\"\" if self._interrupted: return False # for add_to_rankings if self._searched: return True","in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) >","If the number of unique pipelines to search over per batch is one,","problem, training time, cross validation, etc. \"\"\" if pipeline_id not in self._results['pipeline_results']: raise","self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True","None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}:","allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method",">= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False #","to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines allowed","the first one. The first batch will train a baseline pipline + one","iteration plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError:","self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train","three positional parameters: A dictionary containing the training results for the new pipeline,","num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling =","max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline):","%.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold","evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline","len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling","value return {objective: float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self,","complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary","pipeline training iteration. Callback function takes three positional parameters: The pipeline class, the","None \"\"\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path):","(PipelineBase): pipeline to train and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for","num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations","number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self)","Set max_iterations >= {first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling will run at the","in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time()","True. start_iteration_callback (callable): Function called before each pipeline training iteration. Callback function takes","import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline )","high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High coefficient of variation","self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train =","f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a list of pipelines on the","iterated over. If the number of unique pipelines to search over per batch","been iterated over. If the number of unique pipelines to search over per","allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if","stop searching for new pipelines after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed model families:","evalml.objectives.utils.get_core_objective_names() \" \"to get all objective names allowed in automl.\") return objective() return","of the training data we'll set aside for training ensemble metalearners. Only used","batch after the first one. The first batch will train a baseline pipline","objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is","by pipeline name that maps to the fitted pipeline. Note that the any","parameters found during automl search. If `train_best_pipeline` is set to False, returns an","number of pipeline evaluations made in the search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self):","self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return the number of pipeline evaluations which have","training data we'll set aside for training ensemble metalearners. Only used when ensembling","in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores","{[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for","(1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches)","import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners","pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc","pipelines, but not for optimizing each pipeline during fit-time. When set to 'auto',","Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to train for every","= _pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches =","return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id","self._start if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines","subclasses indicating the pipelines allowed in the search. The default of None indicates","not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid data splitter\") if","The target training data of length [n_samples]. Required for supervised learning tasks. problem_type","raise ValueError(f\"{objective.name.lower()} is not allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get","num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score","cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for","raise PipelineNotFoundError(\"Pipeline class or parameters not found in automl results\") return pipeline_class(parameters, random_seed=self.random_seed)","desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1","the automl rankings with the requirement that automl search has been run. Arguments:","add_result_callback (callable): Function called after each pipeline training iteration. Callback function takes three","A dict of the parameters used to initalize a pipeline with. _ensembling_split_size (float):","_find_best_pipeline(self): \"\"\"Finds the best pipeline in the rankings If self._best_pipeline already exists, check","for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary","+ first batch + each pipeline iteration + 1 first_ensembling_iteration = (1 +","(self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\"Ensembling","pipeline, and `search_order`: a list describing the order the pipelines were searched. \"\"\"","an integer, then the time will be in seconds. For strings, time can","to None. tolerance (float): Minimum percentage difference to qualify as score improvement for","+ len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False","o in obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return","be empty so we will return # nan for the base score. percent_better","object. add_result_callback (callable): Function called after each pipeline training iteration. Callback function takes","False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for %s. \" % self.objective.name) logger.info(\"{} score","to train and score pipelines.\") if self.max_batches is not None: logger.info(f\"Searching up to","if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc =","will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after","instead\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\"tolerance","to True. pipeline_parameters (dict): A dict of the parameters used to initalize a","data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None,","existing_main_objective = next((obj for obj in additional_objectives if obj.name == self.objective.name), None) if","work well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None,","all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef","each pipeline training iteration. Callback function takes three positional parameters: A dictionary containing","if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline,","f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\"","+ one of each pipeline family allowed in the search. \"\"\" if X_train","ValueError('Must specify training data target values as a 1d vector using the y_train","desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0]","training will not be included in the dictionary but the exception and stacktrace","return dictionary of information about pipeline. Defaults to False. Returns: Description of specified","= get_core_objectives(self.problem_type) # if our main objective is part of default set of","for c in all_objective_scores: if c in [\"# Training\", \"# Validation\"]: all_objective_scores[c] =","the primary objective {self.objective}.\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc","error out during scoring will not be included in the dictionary but the","= True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\",","not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to","1 exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape,","number of pipeline evaluations which have been made Returns: int: the number of","least the gap and max_delay \" f\"parameters. Received {problem_configuration}.\") return problem_configuration or {}","1 logger.info(\"Using default limit of max_batches=1.\\n\") if patience and (not isinstance(patience, int) or","= std all_objective_scores.loc[\"coef of var\", c] = std / mean if abs(mean) >","Required for supervised learning tasks. problem_type (str or ProblemTypes): type of supervised learning","cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search.","get_core_objectives(self.problem_type) # if our main objective is part of default set of objectives","self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file path Arguments: file_path","the parameters used to train that pipeline during automl search. Arguments: pipeline_id (int):","to find file to load Returns: AutoSearchBase object \"\"\" with open(file_path, 'rb') as","[]: if pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline {} is not compatible with","baseline pipeline to the data. This is the first pipeline fit during search.","to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called","has been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \"\"\" pipeline_rows","Note that the any pipelines that error out during scoring will not be","allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines","(list(class)): A list of PipelineBase subclasses indicating the pipelines allowed in the search.","obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives =","score pipelines.\") if self.max_batches is not None: logger.info(f\"Searching up to {self.max_batches} batches for","training data target values as a 1d vector using the y_train argument') try:","False, returns an untrained pipeline instance. Returns: PipelineBase: A trained instance of the","show_iteration_plot = False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for %s. \" % self.objective.name)","= [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline:","return problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the user asking","{_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc","lines = sorted(['\\t{}'.format(o.name) for o in obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if callable(function):","if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective =","0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines","input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try:","parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training","feature_types (list, optional): list of feature types, either numerical or categorical. Categorical features","For strings, time can be specified as seconds, minutes, or hours. patience (int):","location to find file to load Returns: AutoSearchBase object \"\"\" with open(file_path, 'rb')","Used to propose and rank pipelines, but not for optimizing each pipeline during","else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and score pipelines.\") if self.max_batches is not","self._searched: logger.info(\"AutoMLSearch.search() has already been run and will not run again on the","None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o","if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in","\\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters","in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline {} is","accepts kwargs, so AutoMLSearch is able to pass along other appropriate parameters by","for \"Plotting\" documentation, since Sphinx does not work well with instance attributes. plot","def _find_best_pipeline(self): \"\"\"Finds the best pipeline in the rankings If self._best_pipeline already exists,","'{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and","small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1} to","best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id'])","instead\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results':","None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary`","(dict): A dict of the parameters used to initalize a pipeline with. _ensembling_split_size","_handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the user asking if they want to stop","def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file path Arguments: file_path (str):","not found in automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class","< 0: raise ValueError(f\"Parameter max_time must be None or non-negative. Received {max_time}.\") if","y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds,","will not run.\") run_ensembling = False if run_ensembling and self.max_iterations is not None:","try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id],","that maps to a dictionary of scores. Note that the any pipelines that","from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__)","evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from","pipeline training iteration. Callback function takes three positional parameters: A dictionary containing the","all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\"# Training\", \"#","try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of","n_folds = len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): #","Maximum number of iterations to search. If max_iterations and max_time is not set,","takes three positional parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch","self.objective = self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise","additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class =","run every {ensemble_nth_batch} batches.\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches","mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c] =","self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter","= False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not","allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False","along other appropriate parameters by default. Defaults to None, which will call `log_error_callback`.","self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None: logger.info(\"Generating pipelines to","additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part of default set","[self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\"Parameter max_time","0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{} iterations without improvement.","parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable):","ensembling.\") else: logger.info(f\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) *","full list. objective (str, ObjectiveBase): The objective to optimize for. Used to propose","{best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self): \"\"\"Finds the best","( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated","curr_score < best_score if score_improved and significant_change: best_score = curr_score num_without_improvement = 0","pipelines): \"\"\"Train a list of pipelines on the training data. This can be","def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe","scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary","by pipeline name that maps to a dictionary of scores. Note that the","\"\" def search(self, show_iteration_plot=True): \"\"\"Find the best pipeline for the data set. Arguments:","been made Returns: int: the number of pipeline evaluations made in the search","too small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1}","= desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if","rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if","or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc =","self._num_pipelines() if num_pipelines == 0: return True # check max_time and max_iterations elapsed","to {[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params","R2 for regression problems. max_iterations (int): Maximum number of iterations to search. If","current state, should the search continue? Returns: bool: True if yes, False if","The input training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The","data.\") return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the ID of a pipeline training","return function.__name__ else: return None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective:","0: run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_batches is too","optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train","`train_best_pipeline` is set to False, returns an untrained pipeline instance. \"\"\" if not","for p in required_parameters): raise ValueError(\"user_parameters must be a dict containing values for","infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger =","of default set of objectives for problem_type, remove it existing_main_objective = next((obj for","if run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\"Ensembling split size","best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >=","type. Note that if allowed_pipelines is provided, this parameter will be ignored. data_splitter","objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o in","ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The","scoring results from all pipelines searched\"\"\" ascending = True if self.objective.greater_is_better: ascending =","not all(p in problem_configuration for p in required_parameters): raise ValueError(\"user_parameters must be a","num_pipelines = self._num_pipelines() if num_pipelines == 0: return True # check max_time and","(boolean): If True, runs ensembling in a separate batch after every allowed pipeline","logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if","been run and will not run again on the same instance. Re-initialize AutoMLSearch","search. \"\"\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS:","highest-scoring set of parameters used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def","check to make sure it is different from the current best pipeline before","load Returns: AutoSearchBase object \"\"\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def","argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass,","required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration for","= sorted(['\\t{}'.format(o.name) for o in obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if callable(function): return","self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are","f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data","classification pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training time (including CV): %.1f seconds\" %","argument') if y_train is None: raise ValueError('Must specify training data target values as","score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data']","training, and the AutoMLSearch object. error_callback (callable): Function called when `search()` errors and","in for the gap and max_delay variables. train_best_pipeline (boolean): Whether or not to","A trained instance of the best pipeline and parameters found during automl search.","is None: logger.info(\"Generating pipelines to search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set","0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline","self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\"Given","untrained instance of the specified pipeline initialized with the parameters used to train","leading_char = \"\\n\" start_of_loop = time.time() while True: choice = input(leading_char + \"Do","the parameters used during training, and the AutoMLSearch object. error_callback (callable): Function called","if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the current AutoML batch produced","of specified pipeline. Includes information such as type of pipeline components, problem, training","objective {} is not compatible with a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is","n_jobs below -1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean): If","all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of var\", c]","to %s pipelines. \" % self.max_iterations) if self.max_time is not None: logger.info(\"Will stop","\" f\"parameters. Received {problem_configuration}.\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt","objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field,","the exception and stacktrace will be displayed in the log. \"\"\" return self._engine.score_batch(pipelines,","of var\", c] = std / mean if abs(mean) > 0 else np.inf","stop the search. Returns: bool: If True, search should terminate early \"\"\" leading_char","import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline,","< 0: raise ValueError(f\"Parameter max_batches must be None or non-negative. Received {max_batches}.\") if","X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline =","False # check for early stopping if self.patience is None or self.tolerance is","best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train","from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner","if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline =","tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback:","untrained_pipeline containing the parameters used during training, and the AutoMLSearch object. error_callback (callable):","notebook. Disabled by default in non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already","This is the first pipeline fit during search. \"\"\" if self.problem_type == ProblemTypes.BINARY:","curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved","best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return","desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0:","AutoMLSearchException(f\"All pipelines in the current AutoML batch produced a score of np.nan on","pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results =","Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name","training time, cross validation, etc. \"\"\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline","original stopping criterion and current state, should the search continue? Returns: bool: True","ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The default of None","= evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {}","search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start =","{max_batches}.\") if max_iterations is not None and max_iterations < 0: raise ValueError(f\"Parameter max_iterations","a dict with results from each pipeline, and `search_order`: a list describing the","pipeline search\") logger.info(\"Optimizing for %s. \" % self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if","(ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring.","run_ensembling and self.max_iterations is not None: # Baseline + first batch + each","Whether or not to train the best pipeline before returning it. Defaults to","def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores']","% self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to","self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a","y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None,","False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched =","1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling","pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout","pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class)","between 0 and 1 exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _,","is one, so ensembling will not run.\") run_ensembling = False if run_ensembling and","pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps","all_objective_scores.loc[\"coef of var\", c] = std / mean if abs(mean) > 0 else","= additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives} if","evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives","problem_type (str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a","bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High coefficient of variation (cv >=","depending on the problem type. Note that if allowed_pipelines is provided, this parameter","limit of max_batches=1.\\n\") if patience and (not isinstance(patience, int) or patience < 0):","self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings","{'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration for p in","after each pipeline training iteration. Callback function takes three positional parameters: A dictionary","pipeline_id): \"\"\"Given the ID of a pipeline training result, returns an untrained instance","1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not","search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap',","then adds the results to the automl rankings with the requirement that automl","at file path Arguments: file_path (str): location to save file pickle_protocol (int): the","dictionary of scores. Note that the any pipelines that error out during scoring","(int): The maximum number of batches of pipelines to search. Parameters max_time, and","to propose and rank pipelines, but not for optimizing each pipeline during fit-time.","instance. Re-initialize AutoMLSearch to search again.\") return # don't show iteration plot outside","self.objective.greater_is_better: ascending = False full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"]","log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\"","ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not found","# if our main objective is part of default set of objectives for","is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part of","of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout","Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\"","return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from the","True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False","self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for","is not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline","non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is","pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline,","run. Defaults to False. max_batches (int): The maximum number of batches of pipelines","default limit of max_batches=1.\\n\") if patience and (not isinstance(patience, int) or patience <","num_pipelines >= self.max_iterations: return False # check for early stopping if self.patience is","warning if variance is higher than specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv =","given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable,","instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None,","+ self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations =","= time.time() - self._start if self.max_time and elapsed >= self.max_time: return False elif","infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration,","of the specified pipeline initialized with the parameters used to train that pipeline","train_pipelines(self, pipelines): \"\"\"Train a list of pipelines on the training data. This can","optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before","and 1 are equivalent. If set to -1, all CPUs are used. For","all CPUs are used. For n_jobs below -1, (n_cpus + 1 + n_jobs)","problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives if obj.name ==","= X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline","pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name],","types, either numerical or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean,","threshold=0.2): \"\"\"Checks cross-validation scores and logs a warning if variance is higher than","rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \"\"\"Returns a","logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch","MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap']","rankings_df @property def best_pipeline(self): \"\"\"Returns a trained instance of the best pipeline and","during scoring will not be included in the dictionary but the exception and","continue? Returns: bool: True if yes, False if no. \"\"\" if self._interrupted: return","desc = f\"\\nSearch finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline","type): if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed in AutoML! \"","time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = []","they want to stop the search. Returns: bool: If True, search should terminate","called before each pipeline training iteration. Callback function takes three positional parameters: The","to search. Parameters max_time, and max_iterations have precedence over stopping the search. problem_configuration","log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores =","on the problem type. Note that if allowed_pipelines is provided, this parameter will","== \"n\": # So that the time in this loop does not count","else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay","pipeline, an untrained_pipeline containing the parameters used during training, and the AutoMLSearch object.","during automl search. If `train_best_pipeline` is set to False, returns an untrained pipeline","\"to get all objective names allowed in automl.\") return objective() return objective def","self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train,","self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data","f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if not self.rankings.empty: rankings_str","returns an untrained pipeline instance. \"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl search must","parameters: A dictionary containing the training results for the new pipeline, an untrained_pipeline","f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\"","split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices,","Arguments: file_path (str): location to find file to load Returns: AutoSearchBase object \"\"\"","the training data we'll set aside for training ensemble metalearners. Only used when","self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train","is not None and max_batches < 0: raise ValueError(f\"Parameter max_batches must be None","as a 1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except","get_pipeline(self, pipeline_id): \"\"\"Given the ID of a pipeline training result, returns an untrained","log. \"\"\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a list","else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if","provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use.","initialized with the parameters used to train that pipeline during automl search. Arguments:","if self._searched: return True # Run at least one pipeline for every search","__init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None,","objectives for problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives if","None: # Baseline + first batch + each pipeline iteration + 1 first_ensembling_iteration","takes three positional parameters: A dictionary containing the training results for the new","elif self.max_iterations is not None: logger.info(\"Searching up to %s pipelines. \" % self.max_iterations)","self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is not compatible with a","scores. Note that the any pipelines that error out during scoring will not","pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self): \"\"\"Finds the","if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train","for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def","iterations to search. If max_iterations and max_time is not set, then max_iterations will","run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling will run at","type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data,","pipelines were searched. \"\"\" return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a pandas.DataFrame with","the any pipelines that error out during training will not be included in","search has been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \"\"\"","\"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline =","number of unique pipelines is one, so ensembling will not run.\") run_ensembling =","is different from the current best pipeline before training and thresholding\"\"\" if len(self.rankings)","be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The default of","at least the gap and max_delay \" f\"parameters. Received {problem_configuration}.\") return problem_configuration or","= None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices =","if num_pipelines == 0: return True # check max_time and max_iterations elapsed =","1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise","self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter",">= {first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling will run at the {first_ensembling_iteration} iteration","not (0 < _ensembling_split_size < 1): raise ValueError(f\"Ensembling split size must be between","f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result","score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1","The 'all_objective_scores' field contains scores for all objectives # but also fields like","cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline =","set to True, but the number of unique pipelines is one, so ensembling",">= {ensemble_nth_batch + 1} to run ensembling.\") else: logger.info(f\"Ensembling will run every {ensemble_nth_batch}","Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0)","default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot =","train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline,","so we want to exclude them since # they are not scores if","generator. Defaults to 0. n_jobs (int or None): Non-negative integer describing level of","the first pipeline fit during search. \"\"\" if self.problem_type == ProblemTypes.BINARY: baseline =","as estimated on unseen data.\") return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the ID","of unique pipelines is one, so ensembling will not run.\") run_ensembling = False","self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params )","= random_seed self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except","is part of default set of objectives for problem_type, remove it existing_main_objective =","all(p in problem_configuration for p in required_parameters): raise ValueError(\"user_parameters must be a dict","pipeline to train and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter","integer describing level of parallelism used for pipelines. None and 1 are equivalent.","import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter","should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to","best_score) / best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else","problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the user asking if","if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\"Using","if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set to True, but the","self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self,","'max_delay'} if not problem_configuration or not all(p in problem_configuration for p in required_parameters):","model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start =","all pipelines searched\"\"\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols =","self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids =","if self.max_time is not None: logger.info(\"Will stop searching for new pipelines after %d","of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False","tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments:","\"\"\"Find the best pipeline for the data set. Arguments: feature_types (list, optional): list","self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize,","= max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not","should be passed in for the gap and max_delay variables. train_best_pipeline (boolean): Whether","that the time in this loop does not count towards the time budget","splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner class to use.","logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and","threshold. Defaults to True. start_iteration_callback (callable): Function called before each pipeline training iteration.","pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration},","\"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\":","length [n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes): type of","mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search","Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int)","between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of","is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best","< 0: raise ValueError(f\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\") self.max_time","in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class","is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except","+ 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is set to","0. n_jobs (int or None): Non-negative integer describing level of parallelism used for","> 0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores)","BaseCrossValidator): raise ValueError(\"Not a valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective","(str): location to find file to load Returns: AutoSearchBase object \"\"\" with open(file_path,","for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps","sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks","n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters","> 1.0 or tolerance < 0.0): raise ValueError(\"tolerance value must be a float","Necessary for \"Plotting\" documentation, since Sphinx does not work well with instance attributes.","so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\") else:","pipelines searched\"\"\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\"id\",","to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the","Received {problem_configuration}.\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the","= False logger.warning(f\"Ensembling is set to True, but max_iterations is too small, so","if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >=","{elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline =","set to {[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type,","self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update()","+ str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and","percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for","self.max_iterations) if self.max_time is not None: logger.info(\"Will stop searching for new pipelines after","If max_iterations and max_time is not set, then max_iterations will default to max_iterations","if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids =","self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline])","self.max_iterations and num_pipelines >= self.max_iterations: return False # check for early stopping if","problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective to","for early stopping if self.patience is None or self.tolerance is None: return True","pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline","\" % self.max_iterations) if self.max_time is not None: logger.info(\"Will stop searching for new","is True. Must be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch","Number of iterations without improvement to stop search early. Must be positive. If","self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is set","to pass along other appropriate parameters by default. Defaults to None, which will","True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in","\"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE:","Sphinx does not work well with instance attributes. plot = PipelineSearchPlots def __init__(self,","raises an Exception. Callback function takes three positional parameters: the Exception raised, the","pipelines. None and 1 are equivalent. If set to -1, all CPUs are","search (y/n)? \").strip().lower() if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif choice","keyed by pipeline name that maps to a dictionary of scores. Note that","function.__name__ else: return None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\"","evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives,","from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline,","in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all objective names allowed in","# check for early stopping if self.patience is None or self.tolerance is None:","= pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted","{_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\"","= PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None,","run.\") run_ensembling = False if run_ensembling and self.max_iterations is not None: # Baseline","pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise","break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and","Dictionary keyed by pipeline name that maps to the fitted pipeline. Note that","pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file path Arguments: file_path (str): location to save","is too small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to","pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch,","None, early stopping is disabled. Defaults to None. tolerance (float): Minimum percentage difference","add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2,","= self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self,","as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import","import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family","_validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is","applicable if patience is not None. Defaults to None. allowed_pipelines (list(class)): A list","about pipeline. Defaults to False. Returns: Description of specified pipeline. Includes information such","should terminate early \"\"\" leading_char = \"\\n\" start_of_loop = time.time() while True: choice","{self._engine.__class__.__name__} to train and score pipelines.\") if self.max_batches is not None: logger.info(f\"Searching up","error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5):","self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling =","the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after each","{[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params =","raised, the traceback, and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch","non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to","the event add_to_rankings is called before search _baseline_cv_scores will be empty so we","SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in","pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High coefficient of","iterations after that.\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True if","exit search (y/n)? \").strip().lower() if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif","to max_iterations of 5. max_time (int, str): Maximum time to search for pipelines.","if run_ensembling and self.max_iterations is not None: # Baseline + first batch +","not run. Defaults to False. max_batches (int): The maximum number of batches of","we will return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name,","not for optimizing each pipeline during fit-time. When set to 'auto', chooses: -","y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required for supervised","Callback function takes three positional parameters: the Exception raised, the traceback, and the","else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj","score plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \"\"\" if","in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class","+ \"Do you really want to exit search (y/n)? \").strip().lower() if choice ==","contains scores for all objectives # but also fields like \"# Training\" and","% self.max_time) logger.info(\"Allowed model families: %s\\n\" % \", \".join([model.value for model in self.allowed_model_families]))","KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids)","= next((obj for obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective","\"\"\"Checks cross-validation scores and logs a warning if variance is higher than specified","three positional parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch object.","else: logger.info(f\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch}","with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline):","significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement","raise ValueError(\"Additional objective {} is not compatible with a {} problem.\".format(obj.name, self.problem_type.value)) for","containing the training results for the new pipeline, an untrained_pipeline containing the parameters","= convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches","keyed by pipeline name that maps to the fitted pipeline. Note that the","input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not","PipelineBase: A trained instance of the best pipeline and parameters found during automl","adds the results to the automl rankings with the requirement that automl search","woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm","made in the search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original stopping","`search_order`: a list describing the order the pipelines were searched. \"\"\" return copy.deepcopy(self._results)","\"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40 # Necessary for \"Plotting\" documentation, since Sphinx","logger.info(f\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations","\"# Training\" and \"# Testing\", so we want to exclude them since #","results to the automl rankings with the requirement that automl search has been","after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline","Function called after each pipeline training iteration. Callback function takes three positional parameters:","get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed","= False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched","optimize_thresholds (bool): Whether or not to optimize the binary pipeline threshold. Defaults to","1.0 inclusive. Received {} instead\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0","If True, search should terminate early \"\"\" leading_char = \"\\n\" start_of_loop = time.time()","pipelines on the training data. This can be helpful for training pipelines once","parallelism used for pipelines. None and 1 are equivalent. If set to -1,","allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling","len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling =","training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines","y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline","= n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to","import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import","improvement for early stopping. Only applicable if patience is not None. Defaults to","self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3,","seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold in","1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective,","not set, then max_iterations will default to max_iterations of 5. max_time (int, str):","or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement","to the data. This is the first pipeline fit during search. \"\"\" if","max_time is not set, then max_iterations will default to max_iterations of 5. max_time","set to False, returns an untrained pipeline instance. \"\"\" if not self._best_pipeline: raise","\"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to","in the current AutoML batch produced a score of np.nan on the primary","self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines","if not (0 < _ensembling_split_size < 1): raise ValueError(f\"Ensembling split size must be","in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p in","{} instead\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance < 0.0): raise","\"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and","True: choice = input(leading_char + \"Do you really want to exit search (y/n)?","the rankings If self._best_pipeline already exists, check to make sure it is different","for estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator","underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c","been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \"\"\" pipeline_rows =","an untrained pipeline instance. Returns: PipelineBase: A trained instance of the best pipeline","file_path (str): location to save file pickle_protocol (int): the pickle data stream format.","max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time Series Baseline","num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1))","in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives # but","self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed,","a dict containing values for at least the gap and max_delay \" f\"parameters.","std all_objective_scores.loc[\"coef of var\", c] = std / mean if abs(mean) > 0","pandas.DataFrame with scoring results from the highest-scoring set of parameters used with each",">= self.max_iterations: return False # check for early stopping if self.patience is None","curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\"\\n\\n{}","for problem type if not empty. random_seed (int): Seed for the random number","percentage difference to qualify as score improvement for early stopping. Only applicable if","must be a float between 0.0 and 1.0 inclusive. Received {} instead\".format(tolerance)) self.patience","if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline:","be passed in for the gap and max_delay variables. train_best_pipeline (boolean): Whether or","exists, check to make sure it is different from the current best pipeline","ValueError(\"Not a valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is","fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives # but also","self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train,","the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of","of pipeline evaluations which have been made Returns: int: the number of pipeline","that error out during training will not be included in the dictionary but","None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1,","all objective names allowed in automl.\") return objective() return objective def __str__(self): def","type of pipeline components, problem, training time, cross validation, etc. \"\"\" if pipeline_id","the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before each",") from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import","evaluates a given pipeline then adds the results to the automl rankings with","number of batches of pipelines to search. Parameters max_time, and max_iterations have precedence","pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns:","first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]:","try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if","is None: raise PipelineNotFoundError(\"Pipeline class or parameters not found in automl results\") return","empty. random_seed (int): Seed for the random number generator. Defaults to 0. n_jobs","= False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter =","AutoMLSearch.\") return True elif choice == \"n\": # So that the time in","raise ValueError(\"user_parameters must be a dict containing values for at least the gap","Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for","Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False","= True def _find_best_pipeline(self): \"\"\"Finds the best pipeline in the rankings If self._best_pipeline","tolerance < 0.0): raise ValueError(\"tolerance value must be a float between 0.0 and","self) desc = f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3]","SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException,","if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using {self._engine.__class__.__name__} to train and score pipelines.\") if self.max_batches","best_pipeline def _num_pipelines(self): \"\"\"Return the number of pipeline evaluations which have been made","else curr_score < best_score if score_improved and significant_change: best_score = curr_score num_without_improvement =","as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a list of pipelines on","pipelines is one, so ensembling will not run.\") run_ensembling = False if run_ensembling","1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is set to True,","= [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives]","\"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\":","to search over per batch is one, ensembling will not run. Defaults to","if self._interrupted: return False # for add_to_rankings if self._searched: return True # Run","desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number","def search(self, show_iteration_plot=True): \"\"\"Find the best pipeline for the data set. Arguments: feature_types","mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives:","high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if","pipeline with. _ensembling_split_size (float): The amount of the training data we'll set aside","def results(self): \"\"\"Class that allows access to a copy of the results from","objectives): \"\"\"Score a list of pipelines on the given holdout data. Arguments: pipelines","pipelines for this problem type are allowed. Setting this field will cause allowed_model_families","if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\"Ensembling is set to True, but","\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot)","return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a given pipeline then adds","Defaults to True. pipeline_parameters (dict): A dict of the parameters used to initalize","batches for a total of {self.max_iterations} pipelines. \") elif self.max_iterations is not None:","False if run_ensembling and self.max_iterations is not None: # Baseline + first batch","= self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train,","if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is set to True, but","as type of pipeline components, problem, training time, cross validation, etc. \"\"\" if","ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame,","self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train)","= self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted =","selecting `best_pipeline`.\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file","self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else:","pipeline. Note that the any pipelines that error out during training will not","= {o.name: o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int,","will not run. Defaults to False. max_batches (int): The maximum number of batches","get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__,","all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of var\", c] =",".pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine","score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \"\"\"Score a list of pipelines on the given","parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self):","evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions","For n_jobs below -1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean):","{len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\") if self.max_batches and self.max_iterations is None: self.show_batch_output","the time in this loop does not count towards the time budget (if","and the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training iteration.","containing `pipeline_results`: a dict with results from each pipeline, and `search_order`: a list","if self.patience is None or self.tolerance is None: return True first_id = self._results['search_order'][0]","scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores =","import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives,","for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives =","False else: leading_char = \"\" def search(self, show_iteration_plot=True): \"\"\"Find the best pipeline for","current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the current AutoML batch produced a score of","finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None:","For example, in time series problems, values should be passed in for the","raise ValueError(f\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time)","1 + n_jobs) are used. ensembling (boolean): If True, runs ensembling in a","the best pipeline in the rankings If self._best_pipeline already exists, check to make","random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id (int): pipeline to","in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search()","self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted","training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family ==","too small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run","run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\")","len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives()","return_dict (bool): If True, return dictionary of information about pipeline. Defaults to False.","The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not","None and 1 are equivalent. If set to -1, all CPUs are used.","if field in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for","Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to the fitted pipeline.","Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\"","be between 0 and 1 exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices,","of 5. max_time (int, str): Maximum time to search for pipelines. This will","self.max_batches: self.max_batches = 1 logger.info(\"Using default limit of max_batches=1.\\n\") if patience and (not","names allowed in automl.\") return objective() return objective def __str__(self): def _print_list(obj_list): lines","X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return the number of","AutoSearchBase object \"\"\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines):","compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to the","objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data =","exception and stacktrace will be displayed in the log. \"\"\" return self._engine.score_batch(pipelines, X_holdout,","not run.\") run_ensembling = False if run_ensembling and self.max_iterations is not None: #","PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None,","\"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\":","0.2 _pipelines_per_batch (int): The number of pipelines to train for every batch after","+ len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0","Must be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The","of pipelines on the training data. This can be helpful for training pipelines","ModelFamily)): The model families to search. The default of None searches over all","{type(max_time)} with value {str(max_time)}..\") if isinstance(max_time, (int, float)) and max_time < 0: raise","logger.info(\"Generating pipelines to search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name","ValueError(f\"Parameter max_batches must be None or non-negative. Received {max_batches}.\") if max_iterations is not","X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective)","_should_continue(self): \"\"\"Given the original stopping criterion and current state, should the search continue?","features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used","is set to True, but max_iterations is too small, so ensembling will not","with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns a pandas.DataFrame with","+ len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling","max_time, and max_iterations have precedence over stopping the search. problem_configuration (dict, None): Additional","== []: raise ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if","and max_delay \" f\"parameters. Received {problem_configuration}.\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \"\"\"Presents","results from the highest-scoring set of parameters used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\",","type are allowed. Setting this field will cause allowed_model_families to be ignored. allowed_model_families","self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter =","untrained pipeline instance associated with the provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if","None or parameters is None: raise PipelineNotFoundError(\"Pipeline class or parameters not found in","return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline =","in non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already been run and will","Non-negative integer describing level of parallelism used for pipelines. None and 1 are","%s\\n\" % \", \".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot:","Must also accepts kwargs, so AutoMLSearch is able to pass along other appropriate","self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations:","Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps to a","= max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and not","except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class =","indicates all pipelines for this problem type are allowed. Setting this field will","problems, values should be passed in for the gap and max_delay variables. train_best_pipeline","return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o in obj_list])","== ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train =","must be a dict containing values for at least the gap and max_delay","self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1","to 0. n_jobs (int or None): Non-negative integer describing level of parallelism used","(int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided","not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\") else: logger.info(f\"Ensembling will run","+ rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'}","choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif choice == \"n\": # So","get all objective names allowed in automl.\") return objective() return objective def __str__(self):","self.max_time is not None: logger.info(\"Will stop searching for new pipelines after %d seconds.\\n\"","integer, then the time will be in seconds. For strings, time can be","= None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import plotly; skipping","ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family","self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines()","pipelines allowed in the search. The default of None indicates all pipelines for","of pipeline evaluations made in the search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given","= None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter =","search. Returns: bool: If True, search should terminate early \"\"\" leading_char = \"\\n\"","logger.info(\"AutoMLSearch.search() has already been run and will not run again on the same","obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not None:","or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list.","are equivalent. If set to -1, all CPUs are used. For n_jobs below","False elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check for early","this field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model","AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number =","(str): location to save file pickle_protocol (int): the pickle data stream format. Returns:","_validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration","ID of a pipeline training result, returns an untrained instance of the specified","and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return","parameters used to train that pipeline during automl search. Arguments: pipeline_id (int): pipeline","this problem type are allowed. Setting this field will cause allowed_model_families to be","Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs.","and \"# Testing\", so we want to exclude them since # they are","40 # Necessary for \"Plotting\" documentation, since Sphinx does not work well with","the pipelines allowed in the search. The default of None indicates all pipelines","raise TypeError(f\"Parameter max_time must be a float, int, string or None. Received {type(max_time)}","and score pipelines.\") if self.max_batches is not None: logger.info(f\"Searching up to {self.max_batches} batches","pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class that allows access","save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file path Arguments: file_path (str): location","and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE:","for problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives if obj.name","best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\")","the data set. Arguments: feature_types (list, optional): list of feature types, either numerical","every {ensemble_nth_batch} batches.\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches -","{} is not compatible with a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None:","None): Non-negative integer describing level of parallelism used for pipelines. None and 1","be helpful for training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)):","logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a given","to search for pipelines. This will not start a new pipeline search after","for this problem type are allowed. Setting this field will cause allowed_model_families to","in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators]","num_ensemble_batches == 0: run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_batches","= get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not","_post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline","are not scores if field in objective_name_to_class: scores[field] += value return {objective: float(score)","training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family ==","Returns: bool: True if yes, False if no. \"\"\" if self._interrupted: return False","the new pipeline, an untrained_pipeline containing the parameters used during training, and the","without improvement to stop search early. Must be positive. If None, early stopping","or non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations","self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else","additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name:","non-negative. Received {max_batches}.\") if max_iterations is not None and max_iterations < 0: raise","self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All","function takes three positional parameters: the Exception raised, the traceback, and the AutoMLSearch","(self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size < 1): raise","\"\"\"Finds the best pipeline in the rankings If self._best_pipeline already exists, check to","estimated on unseen data.\") return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the ID of","import defaultdict import cloudpickle import numpy as np import pandas as pd import","/ mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format',","for ensembler are pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for","and num_pipelines >= self.max_iterations: return False # check for early stopping if self.patience","ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger","self.max_iterations is not None: # Baseline + first batch + each pipeline iteration","if allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting","and max_iterations elapsed = time.time() - self._start if self.max_time and elapsed >= self.max_time:","logger.debug(f\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator,","= time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch finished after {elapsed_time}\"","and not self.max_batches: self.max_batches = 1 logger.info(\"Using default limit of max_batches=1.\\n\") if patience","to train for every batch after the first one. The first batch will","so ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1} to run","not None and max_batches < 0: raise ValueError(f\"Parameter max_batches must be None or","for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is","== ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id)","from the highest-scoring set of parameters used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\")","pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted = False if","self._interrupted = False if self.allowed_pipelines is None: logger.info(\"Generating pipelines to search over...\") allowed_estimators","= True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches -","score improvement for early stopping. Only applicable if patience is not None. Defaults","the pipelines were searched. \"\"\" return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a pandas.DataFrame","time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch finished after {elapsed_time}\" desc","elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS:","and max_time is not set, then max_iterations will default to max_iterations of 5.","aside for training ensemble metalearners. Only used when ensembling is True. Must be","and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize binary classification pipeline thresholds for: {}\".format(self.objective))","{\"gap\": gap, \"max_delay\": max_delay}, \"Time Series Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline])","out during training will not be included in the dictionary but the exception","+= value return {objective: float(score) / n_folds for objective, score in scores.items()} def","patience (int): Number of iterations without improvement to stop search early. Must be","if self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch =","with the provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise","that the any pipelines that error out during scoring will not be included","def __str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o in obj_list]) return '\\n'.join(lines)","Note that the any pipelines that error out during training will not be","{pipeline_name} may not perform as estimated on unseen data.\") return high_variance_cv def get_pipeline(self,","if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return","(int): Maximum number of iterations to search. If max_iterations and max_time is not","notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \"Beginning pipeline","# for add_to_rankings if self._searched: return True # Run at least one pipeline","vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose","True def _find_best_pipeline(self): \"\"\"Finds the best pipeline in the rankings If self._best_pipeline already","should the search continue? Returns: bool: True if yes, False if no. \"\"\"","used during training, and the AutoMLSearch object. error_callback (callable): Function called when `search()`","self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the","show_iteration_plot=True): \"\"\"Find the best pipeline for the data set. Arguments: feature_types (list, optional):","pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids","= { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs","object at file path Arguments: file_path (str): location to find file to load","- self._start elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch finished after {elapsed_time}\" desc =","exclude them since # they are not scores if field in objective_name_to_class: scores[field]","{get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines","try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression)","c] = std all_objective_scores.loc[\"coef of var\", c] = std / mean if abs(mean)","maps to the fitted pipeline. Note that the any pipelines that error out","plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0","required_parameters): raise ValueError(\"user_parameters must be a dict containing values for at least the","raise ValueError(\"Given objective {} is not compatible with a {} problem.\".format(self.objective.name, self.problem_type.value)) if","objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search _baseline_cv_scores","be a float between 0.0 and 1.0 inclusive. Received {} instead\".format(tolerance)) self.patience =","1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if","random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline","and thresholding\"\"\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline","[self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for","as a 2d array using the X_train argument') if y_train is None: raise","AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all objective names allowed in automl.\")","full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the","pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self): \"\"\"Finds the best pipeline in","the time will be in seconds. For strings, time can be specified as","def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\" if len(desc)","in the dictionary but the exception and stacktrace will be displayed in the","a positive integer. Received {} instead\".format(patience)) if tolerance and (tolerance > 1.0 or","== 1: logger.warning(\"Ensembling is set to True, but the number of unique pipelines","- 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\"Ensembling is","pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline {} is not compatible with problem_type {}.\".format(pipeline.name,","are pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type))","full_rankings_cols = [\"id\", \"pipeline_name\", \"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return","of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback =","\"\"\"Saves AutoML object at file path Arguments: file_path (str): location to save file","a prompt to the user asking if they want to stop the search.","allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None,","= PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines","Callback function takes three positional parameters: A dictionary containing the training results for","the data. This is the first pipeline fit during search. \"\"\" if self.problem_type","AutoMLSearch to search again.\") return # don't show iteration plot outside of a","0: raise ValueError(f\"Parameter max_time must be None or non-negative. Received {max_time}.\") if max_batches","except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id","\"\"\"Return the number of pipeline evaluations which have been made Returns: int: the","pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline =","if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending,","the requirement that automl search has been run. Arguments: pipeline (PipelineBase): pipeline to","objective to optimize for. Used to propose and rank pipelines, but not for","objective names allowed in automl.\") return objective() return objective def __str__(self): def _print_list(obj_list):","new pipeline search after the duration has elapsed. If it is an integer,","len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations","in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores","Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to the fitted","training iteration. Callback function takes three positional parameters: The pipeline class, the pipeline","file path Arguments: file_path (str): location to save file pickle_protocol (int): the pickle","The model families to search. The default of None searches over all model","in automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None","str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold:","and will not run again on the same instance. Re-initialize AutoMLSearch to search","(pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required for supervised learning","parameters: the Exception raised, the traceback, and the AutoMLSearch object. Must also accepts","_add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to the data. This is the first pipeline","0.0): raise ValueError(\"tolerance value must be a float between 0.0 and 1.0 inclusive.","open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML object","attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None,","the search. \"\"\" if X_train is None: raise ValueError('Must specify training data as","self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective,","each pipeline family allowed in the search. \"\"\" if X_train is None: raise","used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that","self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if","= self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search _baseline_cv_scores will","the fitted pipeline. Note that the any pipelines that error out during training","_get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for","values as a 1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type)","are used. ensembling (boolean): If True, runs ensembling in a separate batch after","= len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches","tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] }","copy import time from collections import defaultdict import cloudpickle import numpy as np","get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train,","\"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores)","logger.warning(f\"Ensembling is set to True, but max_iterations is too small, so ensembling will","before each pipeline training iteration. Callback function takes three positional parameters: The pipeline","pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def","scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field, value","to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for","check max_time and max_iterations elapsed = time.time() - self._start if self.max_time and elapsed","stopping. Only applicable if patience is not None. Defaults to None. allowed_pipelines (list(class)):","loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores","{} def _handle_keyboard_interrupt(self): \"\"\"Presents a prompt to the user asking if they want","max_batches is not None and max_batches < 0: raise ValueError(f\"Parameter max_batches must be","self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type()","self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc +","self.max_time) logger.info(\"Allowed model families: %s\\n\" % \", \".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot","of iterations without improvement to stop search early. Must be positive. If None,","remove it existing_main_objective = next((obj for obj in additional_objectives if obj.name == self.objective.name),","self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective):","if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif choice == \"n\": #","information about pipeline. Defaults to False. Returns: Description of specified pipeline. Includes information","None or non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time","None or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score']","from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError","for obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not","that automl search has been run. Arguments: pipeline (PipelineBase): pipeline to train and",") def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\" if","of objectives for problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives","make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters","None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj)","best pipeline for the data set. Arguments: feature_types (list, optional): list of feature","in non_core_objectives: raise ValueError(f\"{objective.name.lower()} is not allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \"","_MAX_NAME_LEN = 40 # Necessary for \"Plotting\" documentation, since Sphinx does not work","ensembler are pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\") logger.info(\"Training for {}","to import plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families","ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import (","best_score if score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement","we want to exclude them since # they are not scores if field","(str, ObjectiveBase): The objective to optimize for. Used to propose and rank pipelines,","is not None else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines","id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score)","will return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan))","= self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return the","tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False,","except NameError: show_iteration_plot = False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for %s. \"","time (including CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores =","(list): Custom set of objectives to score on. Will override default objectives for","(self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines)","to False. max_batches (int): The maximum number of batches of pipelines to search.","pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High","BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback","fitted pipeline. Note that the any pipelines that error out during training will","{} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our","of scores. Note that the any pipelines that error out during scoring will","target training data of length [n_samples]. Required for supervised learning tasks. problem_type (str","must be a positive integer. Received {} instead\".format(patience)) if tolerance and (tolerance >","be None or non-negative. Received {max_batches}.\") if max_iterations is not None and max_iterations","\"Plotting\" documentation, since Sphinx does not work well with instance attributes. plot =","\"\"\"Given the ID of a pipeline training result, returns an untrained instance of","True # Run at least one pipeline for every search num_pipelines = self._num_pipelines()","Testing\", so we want to exclude them since # they are not scores","binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before each pipeline","f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max","copy of the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with","= [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue():","data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data","logger.info(\"Searching up to %s pipelines. \" % self.max_iterations) if self.max_time is not None:","ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm","= {'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration for p","over stopping the search. problem_configuration (dict, None): Additional parameters needed to configure the","or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot","(self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\")","search. The default of None indicates all pipelines for this problem type are","y_holdout, objectives): \"\"\"Score a list of pipelines on the given holdout data. Arguments:","objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed","get_ipython except NameError: show_iteration_plot = False log_title(logger, \"Beginning pipeline search\") logger.info(\"Optimizing for %s.","between 0.0 and 1.0 inclusive. Received {} instead\".format(tolerance)) self.patience = patience self.tolerance =","in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines)","list of pipelines on the training data. This can be helpful for training","import copy import time from collections import defaultdict import cloudpickle import numpy as","error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if","field contains scores for all objectives # but also fields like \"# Training\"","of the best pipeline and parameters found during automl search. If `train_best_pipeline` is","= all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean","is set to False, returns an untrained pipeline instance. Returns: PipelineBase: A trained","iteration. Callback function takes three positional parameters: The pipeline class, the pipeline parameters,","not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Additional objective {} is not compatible with a {} problem.\".format(obj.name,","Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize","pickle_protocol (int): the pickle data stream format. Returns: None \"\"\" with open(file_path, 'wb')","self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling:","ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline","Baseline + first batch + each pipeline iteration + 1 first_ensembling_iteration = (1","budget (if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False","the any pipelines that error out during scoring will not be included in","If set to -1, all CPUs are used. For n_jobs below -1, (n_cpus","self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt():","not None and max_iterations < 0: raise ValueError(f\"Parameter max_iterations must be None or","not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise","max_iterations of 5. max_time (int, str): Maximum time to search for pipelines. This","n_jobs) are used. ensembling (boolean): If True, runs ensembling in a separate batch","problem type if not empty. random_seed (int): Seed for the random number generator.","time to search for pipelines. This will not start a new pipeline search","every batch after the first one. The first batch will train a baseline","able to pass along other appropriate parameters by default. Defaults to None, which","at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\") if","produced a score of np.nan on the primary objective {self.objective}.\") self.search_duration = time.time()","= self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class","the exception and stacktrace will be displayed in the log. \"\"\" return self._engine.train_batch(pipelines)","Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]:","objective {self.objective}.\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch","else: logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\") self.max_iterations = (1 + len(self.allowed_pipelines) +","{self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds: {self.optimize_thresholds}\\n\" ) rankings_desc = \"\" if not self.rankings.empty:","cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a list of pipelines on the training data.","{self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add","self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices =","\"max_delay\": max_delay}, \"Time Series Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def","using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one","except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx =","`best_pipeline`.\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML object at file path","collections import defaultdict import cloudpickle import numpy as np import pandas as pd","patience < 0): raise ValueError(\"patience value must be a positive integer. Received {}","search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True # check max_time","optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective =","len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\"All pipelines in the current AutoML batch produced a","ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap,","-cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if","stream format. Returns: None \"\"\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol)","current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while","List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name","pandas as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots","to False. Returns: Description of specified pipeline. Includes information such as type of","by default. Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom set","self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better","first pipeline fit during search. \"\"\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={})","not perform as estimated on unseen data.\") return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given","pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines","__str__(self): def _print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o in obj_list]) return '\\n'.join(lines) def","in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\"Parameter","if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr',","if self.allowed_pipelines is None: logger.info(\"Generating pipelines to search over...\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families)","if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \"Beginning pipeline search\")","not allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all objective names","AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able to pass along","is not None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses","below -1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean): If True,","= MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap =","std / mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with","location to save file pickle_protocol (int): the pickle data stream format. Returns: None","allowed in AutoML! \" \"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all objective names allowed","instance associated with the provided ID \"\"\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is","(ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str),","additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else:","outside of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot =","ensembling is True. Must be between 0 and 1, exclusive. Defaults to 0.2","and raises an Exception. Callback function takes three positional parameters: the Exception raised,","std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of","for regression problems. max_iterations (int): Maximum number of iterations to search. If max_iterations","not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family","dict of the parameters used to initalize a pipeline with. _ensembling_split_size (float): The","you really want to exit search (y/n)? \").strip().lower() if choice == \"y\": logger.info(\"Exiting","baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time Series Baseline Estimator\": {\"gap\": gap,","a list of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List","all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary` to `multiclass` or","continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c] = mean all_objective_scores.loc[\"std\", c]","ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\":","batch is one, ensembling will not run. Defaults to False. max_batches (int): The","that.\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch","self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices","if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if","non-negative. Received {max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations =","is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0","with. _ensembling_split_size (float): The amount of the training data we'll set aside for","self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return the number","logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted =","return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator):","< 0.0): raise ValueError(\"tolerance value must be a float between 0.0 and 1.0","our main objective is part of default set of objectives for problem_type, remove","holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame):","max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1,","pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time,","all_objective_scores: if c in [\"# Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean","pickle data stream format. Returns: None \"\"\" with open(file_path, 'wb') as f: cloudpickle.dump(self,","self._start += time_in_loop return False else: leading_char = \"\" def search(self, show_iteration_plot=True): \"\"\"Find",") rankings_desc = \"\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc =","run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch","= {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines,","+= time_in_loop return False else: leading_char = \"\" def search(self, show_iteration_plot=True): \"\"\"Find the","on the training data. This can be helpful for training pipelines once the","def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to the data. This is the first","early. Must be positive. If None, early stopping is disabled. Defaults to None.","make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import","problem type are allowed. Setting this field will cause allowed_model_families to be ignored.","LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification problems, and -","scores. {pipeline_name} may not perform as estimated on unseen data.\") return high_variance_cv def","positive integer. Received {} instead\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance","set to -1, all CPUs are used. For n_jobs below -1, (n_cpus +","batch will train a baseline pipline + one of each pipeline family allowed","+ 1} to run ensembling.\") else: logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\") self.max_iterations","get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline,","dictionary containing the training results for the new pipeline, an untrained_pipeline containing the","/ cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\"High coefficient of variation (cv >= {threshold})","c] = mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of var\", c] = std","'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot =","\"n\": # So that the time in this loop does not count towards","max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable):","precedence over stopping the search. problem_configuration (dict, None): Additional parameters needed to configure","pipeline initialized with the parameters used to train that pipeline during automl search.","The number of pipelines to train for every batch after the first one.","if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\"Parameter max_time must be","the dictionary but the exception and stacktrace will be displayed in the log.","\"\"\"Presents a prompt to the user asking if they want to stop the","= 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline","'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a list of pipelines","hours. patience (int): Number of iterations without improvement to stop search early. Must","\"\"\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a","max_batches must be None or non-negative. Received {max_batches}.\") if max_iterations is not None","= self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time","= evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean()","and logs a warning if variance is higher than specified threshhold.\"\"\" pipeline_name =","If True, runs ensembling in a separate batch after every allowed pipeline class","not None else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is","get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes,","= optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective","self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\")","to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters","( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters: \\n{'='*20}\\n\" f\"Objective: {get_objective(self.objective).name}\\n\" f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\"","if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] #","other appropriate parameters by default. Defaults to None, which will call `log_error_callback`. additional_objectives","search should terminate early \"\"\" leading_char = \"\\n\" start_of_loop = time.time() while True:","loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break","a full list. objective (str, ObjectiveBase): The objective to optimize for. Used to","\"\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\"","ww.DataColumn): The target training data of length [n_samples]. Required for supervised learning tasks.","else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \"transform_sample\"): train_indices = self.data_splitter.transform_sample(X_train,","equivalent. If set to -1, all CPUs are used. For n_jobs below -1,","import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import","0 and 1 exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _","= percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\":","a float between 0.0 and 1.0 inclusive. Received {} instead\".format(tolerance)) self.patience = patience","self._start elapsed_time = time_elapsed(self._start) desc = f\"\\nSearch finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN)","number generator. Defaults to 0. n_jobs (int or None): Non-negative integer describing level","self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches","families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary` to `multiclass` or `regression` depending","= 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if","= all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def","high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\":","(callable): Function called after each pipeline training iteration. Callback function takes three positional","current best pipeline before training and thresholding\"\"\" if len(self.rankings) == 0: return best_pipeline","defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items():","{self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm","None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import plotly; skipping pipeline","after the duration has elapsed. If it is an integer, then the time","= std / mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\"-\")","ValueError(f\"Ensembling split size must be between 0 and 1 exclusive, received {_ensembling_split_size}\") X_shape","import plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm","the original stopping criterion and current state, should the search continue? Returns: bool:","percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = {","but the exception and stacktrace will be displayed in the log. \"\"\" return","improvement to stop search early. Must be positive. If None, early stopping is","an Exception. Callback function takes three positional parameters: the Exception raised, the traceback,","not None: # Baseline + first batch + each pipeline iteration + 1","towards the time budget (if set) time_in_loop = time.time() - start_of_loop self._start +=","= len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary,","True. Must be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int):","ValueError(\"Additional objective {} is not compatible with a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline","search early...\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in self.additional_objectives: if","and `search_order`: a list describing the order the pipelines were searched. \"\"\" return","X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives","of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines","tasks. problem_type (str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for","f, protocol=pickle_protocol) @staticmethod def load(file_path): \"\"\"Loads AutoML object at file path Arguments: file_path","logger.info(\"Will stop searching for new pipelines after %d seconds.\\n\" % self.max_time) logger.info(\"Allowed model","of PipelineBase subclasses indicating the pipelines allowed in the search. The default of","{max_iterations}.\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches","during search. \"\"\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type ==","for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families","Returns: None \"\"\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def","iteration. Callback function takes three positional parameters: A dictionary containing the training results","CPUs are used. For n_jobs below -1, (n_cpus + 1 + n_jobs) are","= allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores =","load(file_path): \"\"\"Loads AutoML object at file path Arguments: file_path (str): location to find","inplace=True) return rankings_df @property def best_pipeline(self): \"\"\"Returns a trained instance of the best","= [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o","ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class","return False return True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type):","has already been run and will not run again on the same instance.","at file path Arguments: file_path (str): location to find file to load Returns:","the search. For example, in time series problems, values should be passed in","exclusive, received {_ensembling_split_size}\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train,","`binary` to `multiclass` or `regression` depending on the problem type. Note that if","> 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number,","is None: raise ValueError('Must specify training data as a 2d array using the","ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \"\"\"Returns a trained instance","max_time < 0: raise ValueError(f\"Parameter max_time must be None or non-negative. Received {max_time}.\")","if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not","if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \"\"\"Class that allows","Returns: PipelineBase: A trained instance of the best pipeline and parameters found during","import pandas as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from","`automl_search`. Returns: dict containing `pipeline_results`: a dict with results from each pipeline, and","+ 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1)","iterations without improvement to stop search early. Must be positive. If None, early","a valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is not","Must be positive. If None, early stopping is disabled. Defaults to None. tolerance","or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration","\"Use evalml.objectives.utils.get_core_objective_names() \" \"to get all objective names allowed in automl.\") return objective()","= desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number >","on unseen data.\") return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the ID of a","_print_list(obj_list): lines = sorted(['\\t{}'.format(o.name) for o in obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if","Only applicable if patience is not None. Defaults to None. allowed_pipelines (list(class)): A","True, runs ensembling in a separate batch after every allowed pipeline class has","pipline + one of each pipeline family allowed in the search. \"\"\" if","and max_batches < 0: raise ValueError(f\"Parameter max_batches must be None or non-negative. Received","not run again on the same instance. Re-initialize AutoMLSearch to search again.\") return","run ensembling.\") else: logger.info(f\"Ensembling will run every {ensemble_nth_batch} batches.\") self.max_iterations = (1 +","mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings","import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch:","plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot","self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline,","to 'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass","from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import (","best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score']","result, returns an untrained instance of the specified pipeline initialized with the parameters","{max_time}.\") if max_batches is not None and max_batches < 0: raise ValueError(f\"Parameter max_batches","time.time() while True: choice = input(leading_char + \"Do you really want to exit","self.rankings.iloc[0] best_pipeline_name = best_pipeline[\"pipeline_name\"] logger.info(f\"Best pipeline: {best_pipeline_name}\") logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched =","handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import","start_of_loop self._start += time_in_loop return False else: leading_char = \"\" def search(self, show_iteration_plot=True):","evaluations made in the search \"\"\" return len(self._results['pipeline_results']) def _should_continue(self): \"\"\"Given the original","cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives","< best_score if score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else:","training result, returns an untrained instance of the specified pipeline initialized with the","Custom set of objectives to score on. Will override default objectives for problem","to 0.2 _pipelines_per_batch (int): The number of pipelines to train for every batch","batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number","to StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool):","supervised learning tasks. problem_type (str or ProblemTypes): type of supervised learning problem. See","f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\"","(list, optional): list of feature types, either numerical or categorical. Categorical features will","as np import pandas as pd import woodwork as ww from sklearn.model_selection import","def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \"\"\"Checks cross-validation scores and logs a warning if","pipeline before training and thresholding\"\"\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0]","if they want to stop the search. Returns: bool: If True, search should","not count towards the time budget (if set) time_in_loop = time.time() - start_of_loop","\"score\": cv_score, \"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\":","additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part","object \"\"\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train","fields like \"# Training\" and \"# Testing\", so we want to exclude them","{ \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters, \"score\": cv_score,","check for early stopping if self.patience is None or self.tolerance is None: return","pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided ID","get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40 # Necessary for \"Plotting\"","= ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families =","time in this loop does not count towards the time budget (if set)","data. This is the first pipeline fit during search. \"\"\" if self.problem_type ==","= False logger.warning(f\"Ensembling is set to True, but max_batches is too small, so","if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations,","trained instance of the best pipeline and parameters found during automl search. If","Returns: int: the number of pipeline evaluations made in the search \"\"\" return","a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True, return","allowed_model_families (list(str, ModelFamily)): The model families to search. The default of None searches","while True: choice = input(leading_char + \"Do you really want to exit search","scoring results from the highest-scoring set of parameters used with each pipeline.\"\"\" return","patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True,","max_batches=1.\\n\") if patience and (not isinstance(patience, int) or patience < 0): raise ValueError(\"patience","if self.allowed_pipelines == []: raise ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling =","True. pipeline_parameters (dict): A dict of the parameters used to initalize a pipeline","seconds. For strings, time can be specified as seconds, minutes, or hours. patience","an iteration vs. score plot in Jupyter notebook. Disabled by default in non-Jupyter","encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot in Jupyter notebook.","!= self.problem_type: raise ValueError(\"Given pipeline {} is not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value))","a new pipeline search after the duration has elapsed. If it is an","a pandas.DataFrame with scoring results from the highest-scoring set of parameters used with","= make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter","want to exclude them since # they are not scores if field in","logger.info(\"Training for {} problems.\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\"Objective to optimize","self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None","\"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols]","None or non-negative. Received {max_time}.\") if max_batches is not None and max_batches <","model families: %s\\n\" % \", \".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None","cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids","# Baseline + first batch + each pipeline iteration + 1 first_ensembling_iteration =","Returns: PipelineBase: untrained pipeline instance associated with the provided ID \"\"\" pipeline_results =","in [\"# Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std","True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1)","Required. y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required for","tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or","in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\"input_pipeline_ids\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if","+ \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None","return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a list of pipelines on the training","f\"\\nSearch Results: \\n{'='*20}\\n{rankings_str}\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in","the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns:","If self._best_pipeline already exists, check to make sure it is different from the","rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if","class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after","self.patience is None or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score","qualify as score improvement for early stopping. Only applicable if patience is not","main objective is part of default set of objectives for problem_type, remove it","pipeline thresholds for: {}\".format(self.objective)) logger.info(\"Total training time (including CV): %.1f seconds\" % pipeline_results[\"training_time\"])","results\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters","instance of the best pipeline and parameters found during automl search. If `train_best_pipeline`","and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to","\"\"\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True,","1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to train","data set. Arguments: feature_types (list, optional): list of feature types, either numerical or","- best_score) / best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better","{self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self): \"\"\"Finds the best pipeline in the","pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets","during fit-time. When set to 'auto', chooses: - LogLossBinary for binary classification problems,","self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if","raise ValueError('Must specify training data target values as a 1d vector using the","issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\"Not a valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given","str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if","self.max_batches is not None: logger.info(f\"Searching up to {self.max_batches} batches for a total of","else: leading_char = \"\" def search(self, show_iteration_plot=True): \"\"\"Find the best pipeline for the","time series problems, values should be passed in for the gap and max_delay","\"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\"id\"] for","logger.info(f\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\") self._searched = True def _find_best_pipeline(self): \"\"\"Finds the best pipeline","= self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\": max_delay}, \"Time Series Baseline Estimator\":","TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing","pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not found in automl","next((obj for obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is","_pipelines_per_batch=5): \"\"\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of","Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting:","if high_variance_cv: logger.warning(f\"High coefficient of variation (cv >= {threshold}) within cross validation scores.","- 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) +","self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\"Using default limit","is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives =","if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not found in automl results\") pipeline_class =","else np.inf all_objective_scores = all_objective_scores.fillna(\"-\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict:","be run before selecting `best_pipeline`.\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \"\"\"Saves AutoML","to run ensembling.\") else: logger.info(f\"Ensembling will run at the {first_ensembling_iteration} iteration and every","Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\"mean\", c]","_pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\"{pipeline.name}\" if len(desc) >","number of iterations to search. If max_iterations and max_time is not set, then","self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\": type(pipeline), \"pipeline_summary\": pipeline.summary, \"parameters\": pipeline.parameters,","if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\") pipeline = self.get_pipeline(pipeline_id) pipeline_results","in required_parameters): raise ValueError(\"user_parameters must be a dict containing values for at least","automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments:","\"\"\"Loads AutoML object at file path Arguments: file_path (str): location to find file","max_time must be None or non-negative. Received {max_time}.\") if max_batches is not None","= None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration","1): raise ValueError(f\"Ensembling split size must be between 0 and 1 exclusive, received","check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set","search. \"\"\" if X_train is None: raise ValueError('Must specify training data as a","results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results from each","self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores,","allowed in automl.\") return objective() return objective def __str__(self): def _print_list(obj_list): lines =","patience and (not isinstance(patience, int) or patience < 0): raise ValueError(\"patience value must","@property def rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from the highest-scoring set","\"score\", \"validation_score\", \"percent_better_than_baseline\", \"high_variance_cv\", \"parameters\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values())","stopping if self.patience is None or self.tolerance is None: return True first_id =","of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that","def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None,","c] = std / mean if abs(mean) > 0 else np.inf all_objective_scores =","to False, returns an untrained pipeline instance. Returns: PipelineBase: A trained instance of","PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine","over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see options. Change `binary` to `multiclass`","such as type of pipeline components, problem, training time, cross validation, etc. \"\"\"","= defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field, value in","search(self, show_iteration_plot=True): \"\"\"Find the best pipeline for the data set. Arguments: feature_types (list,","dict containing `pipeline_results`: a dict with results from each pipeline, and `search_order`: a","True, but max_iterations is too small, so ensembling will not run. Set max_iterations","is one, ensembling will not run. Defaults to False. max_batches (int): The maximum","level of parallelism used for pipelines. None and 1 are equivalent. If set","estimator in allowed_estimators]}\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in","so ensembling will not run.\") run_ensembling = False if run_ensembling and self.max_iterations is","{\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds","searched\"\"\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\"id\", \"pipeline_name\",","will run every {ensemble_nth_batch} batches.\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch *","{self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives:","a 1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError:","to `multiclass` or `regression` depending on the problem type. Note that if allowed_pipelines","{} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline =","= self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler","is disabled. Defaults to None. tolerance (float): Minimum percentage difference to qualify as","pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property def full_rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results","in pipeline_results[\"cv_data\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\"#","shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of length","and parameters found during automl search. If `train_best_pipeline` is set to False, returns","Splitting: {self.data_splitter}\\n\" f\"Tuner: {self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional","plotly; skipping pipeline search plotting\\n\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm =","_pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1","integer. Received {} instead\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance <","part of default set of objectives for problem_type, remove it existing_main_objective = next((obj","returns an untrained pipeline instance. Returns: PipelineBase: A trained instance of the best","nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better","(int, float)) and max_time < 0: raise ValueError(f\"Parameter max_time must be None or","self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \"\"\"Return the number of pipeline evaluations","= mean all_objective_scores.loc[\"std\", c] = std all_objective_scores.loc[\"coef of var\", c] = std /","ValueError(\"No allowed pipelines to search\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines)","not work well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None,","float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time","must be None or non-negative. Received {max_time}.\") if max_batches is not None and","using the X_train argument') if y_train is None: raise ValueError('Must specify training data","will not run again on the same instance. Re-initialize AutoMLSearch to search again.\")","self.max_iterations: return False # check for early stopping if self.patience is None or","self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={})","Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds","= 1 logger.info(\"Using default limit of max_batches=1.\\n\") if patience and (not isinstance(patience, int)","None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part of default","self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={})","not self.max_batches: self.max_batches = 1 logger.info(\"Using default limit of max_batches=1.\\n\") if patience and","show_iteration_plot (boolean, True): Shows an iteration vs. score plot in Jupyter notebook. Disabled","set to 'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for","of a pipeline training result, returns an untrained instance of the specified pipeline","each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) *","and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set to True, but the number of","{_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\n\" f\"Random Seed: {self.random_seed}\\n\" f\"n_jobs: {self.n_jobs}\\n\" f\"Optimize Thresholds:","= error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling","problem_configuration (dict, None): Additional parameters needed to configure the search. For example, in","self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\"Ensembling is set to True, but max_iterations","for at least the gap and max_delay \" f\"parameters. Received {problem_configuration}.\") return problem_configuration","except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines)","pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property","after that.\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling:","elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check for early stopping","f\"Max Time: {self.max_time}\\n\" f\"Max Iterations: {self.max_iterations}\\n\" f\"Max Batches: {self.max_batches}\\n\" f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or","either numerical or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True):","0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline =","not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending')","else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not","not start a new pipeline search after the duration has elapsed. If it","start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None,","y_train is None: raise ValueError('Must specify training data target values as a 1d","IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import (","Training\" and \"# Testing\", so we want to exclude them since # they","\"high_variance_cv\": high_variance_cv, \"training_time\": training_time, \"cv_data\": cv_data, \"percent_better_than_baseline_all_objectives\": percent_better_than_baseline, \"percent_better_than_baseline\": percent_better_than_baseline[self.objective.name], \"validation_score\": cv_scores[0] }","PipelineNotFoundError(\"Pipeline not found in automl results\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if","but the number of unique pipelines is one, so ensembling will not run.\")","return True # Run at least one pipeline for every search num_pipelines =","is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to","fit-time. When set to 'auto', chooses: - LogLossBinary for binary classification problems, -","for the random number generator. Defaults to 0. n_jobs (int or None): Non-negative","{self.tuner_class.__name__}\\n\" f\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\n\" f\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\n\" f\"Additional Objectives: {_print_list(self.additional_objectives or","import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import","with a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type","{} is not compatible with problem_type {}.\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline","logger.debug(f\"allowed_model_families set to {self.allowed_model_families}\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params","= get_logger(__file__) class AutoMLSearch: \"\"\"Automated Pipeline search.\"\"\" _MAX_NAME_LEN = 40 # Necessary for","raise ValueError(f\"Parameter max_batches must be None or non-negative. Received {max_batches}.\") if max_iterations is","iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\") if self.max_batches and self.max_iterations","The default of None indicates all pipelines for this problem type are allowed.","% \", \".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot","ValueError(\"user_parameters must be a dict containing values for at least the gap and","training ensemble metalearners. Only used when ensembling is True. Must be between 0","and 1.0 inclusive. Received {} instead\".format(tolerance)) self.patience = patience self.tolerance = tolerance or","\"\"\"Returns a pandas.DataFrame with scoring results from the highest-scoring set of parameters used","set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False else: leading_char","if patience is not None. Defaults to None. allowed_pipelines (list(class)): A list of","time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False else: leading_char =","pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline {}","pipeline for the data set. Arguments: feature_types (list, optional): list of feature types,","In the event add_to_rankings is called before search _baseline_cv_scores will be empty so","if variance is higher than specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std()","ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices =","Defaults to True. start_iteration_callback (callable): Function called before each pipeline training iteration. Callback","values for at least the gap and max_delay \" f\"parameters. Received {problem_configuration}.\") return","%s. \" % self.objective.name) logger.info(\"{} score is better.\\n\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\"Using","from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from","new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt():","once the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train.","pipeline (PipelineBase): pipeline to train and evaluate. \"\"\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name]","or None. Received {type(max_time)} with value {str(max_time)}..\") if isinstance(max_time, (int, float)) and max_time","path Arguments: file_path (str): location to save file pickle_protocol (int): the pickle data","one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return","= f\"\\nSearch finished after {elapsed_time}\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is","# check max_time and max_iterations elapsed = time.time() - self._start if self.max_time and","by default in non-Jupyter enviroments. \"\"\" if self._searched: logger.info(\"AutoMLSearch.search() has already been run","of {self.max_iterations} pipelines. \") elif self.max_iterations is not None: logger.info(\"Searching up to %s","for multiclass classification problems, and - R2 for regression problems. max_iterations (int): Maximum","learning tasks. problem_type (str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types","len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm(","to qualify as score improvement for early stopping. Only applicable if patience is","= True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores =","AutoMLSearch is able to pass along other appropriate parameters by default. Defaults to","self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\": pipeline.name, \"pipeline_class\":","0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output)","results for the new pipeline, an untrained_pipeline containing the parameters used during training,","return {objective: float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline,","[make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []:","not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id])","-1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean): If True, runs","baseline pipline + one of each pipeline family allowed in the search. \"\"\"","the search. The default of None indicates all pipelines for this problem type","in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\") logger.debug(f\"allowed_model_families set to","= self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change","a pipeline with. _ensembling_split_size (float): The amount of the training data we'll set","AutoML batch produced a score of np.nan on the primary objective {self.objective}.\") self.search_duration","improvement. Stopping search early...\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in","self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\"Given pipeline {} is not","cross validation, etc. \"\"\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\"Pipeline not found\")","objective is part of default set of objectives for problem_type, remove it existing_main_objective","unseen data.\") return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the ID of a pipeline","self.problem_type.value)) def _add_baseline_pipelines(self): \"\"\"Fits a baseline pipeline to the data. This is the","ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \"Training\")","= self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score","objectives to score on. Will override default objectives for problem type if not","= add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds","to a copy of the results from `automl_search`. Returns: dict containing `pipeline_results`: a","self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train =","= self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\"Pipeline not found in automl results\")","stacktrace will be displayed in the log. \"\"\" return self._engine.score_batch(pipelines, X_holdout, y_holdout, objectives)","None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None","data stream format. Returns: None \"\"\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f,","search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance. \"\"\"","of iterations to search. If max_iterations and max_time is not set, then max_iterations","{objective: float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results):","unique pipelines to search over per batch is one, ensembling will not run.","empty so we will return # nan for the base score. percent_better =","variance is higher than specified threshhold.\"\"\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() /","(binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback","= handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type')","it is an integer, then the time will be in seconds. For strings,","to describe return_dict (bool): If True, return dictionary of information about pipeline. Defaults","= False if self.allowed_pipelines is None: logger.info(\"Generating pipelines to search over...\") allowed_estimators =","self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self)","if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines with IDs: \" +","ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective","passed in for the gap and max_delay variables. train_best_pipeline (boolean): Whether or not","used to initalize a pipeline with. _ensembling_split_size (float): The amount of the training","def get_pipeline(self, pipeline_id): \"\"\"Given the ID of a pipeline training result, returns an","rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\"score\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self):","value must be a float between 0.0 and 1.0 inclusive. Received {} instead\".format(tolerance))","Minimum percentage difference to qualify as score improvement for early stopping. Only applicable","inclusive. Received {} instead\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results","rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results from the highest-scoring set of parameters","ensembling in a separate batch after every allowed pipeline class has been iterated","order the pipelines were searched. \"\"\" return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a","is able to pass along other appropriate parameters by default. Defaults to None,","unique pipelines is one, so ensembling will not run.\") run_ensembling = False if","the number of pipeline evaluations made in the search \"\"\" return len(self._results['pipeline_results']) def","as seconds, minutes, or hours. patience (int): Number of iterations without improvement to","(y/n)? \").strip().lower() if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif choice ==","a score of np.nan on the primary objective {self.objective}.\") self.search_duration = time.time() -","searched. \"\"\" return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns a pandas.DataFrame with scoring results","var\", c] = std / mean if abs(mean) > 0 else np.inf all_objective_scores","n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required","If `train_best_pipeline` is set to False, returns an untrained pipeline instance. Returns: PipelineBase:","(boolean): Whether or not to train the best pipeline before returning it. Defaults","problem_configuration for p in required_parameters): raise ValueError(\"user_parameters must be a dict containing values","the number of unique pipelines to search over per batch is one, ensembling","allowed. Setting this field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)):","does not count towards the time budget (if set) time_in_loop = time.time() -","early stopping if self.patience is None or self.tolerance is None: return True first_id","a {} problem.\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if","/ best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score","total of {self.max_iterations} pipelines. \") elif self.max_iterations is not None: logger.info(\"Searching up to","evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline,","Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\") else: logger.info(f\"Ensembling will run","stop search early. Must be positive. If None, early stopping is disabled. Defaults","to search. If max_iterations and max_time is not set, then max_iterations will default","ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train","and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1 if","Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features]. Required.","IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self,","3] + \"...\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not","pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for fold in pipeline_results[\"cv_data\"]] all_objective_scores","to True, but max_batches is too small, so ensembling will not run. Set","self._find_best_pipeline() @property def results(self): \"\"\"Class that allows access to a copy of the","{}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot","additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives} if not","max_batches < 0: raise ValueError(f\"Parameter max_batches must be None or non-negative. Received {max_batches}.\")","train a baseline pipline + one of each pipeline family allowed in the","try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\"Unable to import plotly; skipping pipeline search","self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors':","to train the best pipeline before returning it. Defaults to True. pipeline_parameters (dict):","different from the current best pipeline before training and thresholding\"\"\" if len(self.rankings) ==","raise ValueError(\"Not a valid data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {}","value {str(max_time)}..\") if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\"Parameter max_time","allows access to a copy of the results from `automl_search`. Returns: dict containing","train for every batch after the first one. The first batch will train","self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set","= {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay']","state, should the search continue? Returns: bool: True if yes, False if no.","best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score <","run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\"Ensembling is set to","logger.info(\"Exiting AutoMLSearch.\") return True elif choice == \"n\": # So that the time","self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for","untrained pipeline instance. \"\"\" if not self._best_pipeline: raise PipelineNotFoundError(\"automl search must be run","bool: If True, search should terminate early \"\"\" leading_char = \"\\n\" start_of_loop =","if callable(function): return function.__name__ else: return None search_desc = ( f\"{handle_problem_types(self.problem_type).name} Search\\n\\n\" f\"Parameters:","of pipeline components, problem, training time, cross validation, etc. \"\"\" if pipeline_id not","1 are equivalent. If set to -1, all CPUs are used. For n_jobs","objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is not compatible with a {} problem.\".format(self.objective.name, self.problem_type.value))","get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline","search. If max_iterations and max_time is not set, then max_iterations will default to","problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise","the order the pipelines were searched. \"\"\" return copy.deepcopy(self._results) @property def rankings(self): \"\"\"Returns","self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check","= self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines with","make sure it is different from the current best pipeline before training and","if pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted = False","to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to","be a float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\") if","%d seconds.\\n\" % self.max_time) logger.info(\"Allowed model families: %s\\n\" % \", \".join([model.value for model","(if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False else:","allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines allowed in the","_ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train,","TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\"pipeline\": {\"gap\": gap, \"max_delay\":","SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title,","pipeline_id (int): pipeline to describe return_dict (bool): If True, return dictionary of information","# Necessary for \"Plotting\" documentation, since Sphinx does not work well with instance","automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot in","when ensembling is True. Must be between 0 and 1, exclusive. Defaults to","data splitter\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\"Given objective {} is not compatible with","from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from","if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is","return rankings_df @property def best_pipeline(self): \"\"\"Returns a trained instance of the best pipeline","found during automl search. If `train_best_pipeline` is set to False, returns an untrained","positive. If None, early stopping is disabled. Defaults to None. tolerance (float): Minimum","a 2d array using the X_train argument') if y_train is None: raise ValueError('Must","Shows an iteration vs. score plot in Jupyter notebook. Disabled by default in","import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from","f\"Allowed Pipelines: \\n{_print_list(self.allowed_pipelines or [])}\\n\" f\"Patience: {self.patience}\\n\" f\"Tolerance: {self.tolerance}\\n\" f\"Data Splitting: {self.data_splitter}\\n\" f\"Tuner:","not compatible with a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []:","False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \"\"\"Fits and evaluates a","self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed","positional parameters: A dictionary containing the training results for the new pipeline, an","self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise","up to %s pipelines. \" % self.max_iterations) if self.max_time is not None: logger.info(\"Will","max_iterations elapsed = time.time() - self._start if self.max_time and elapsed >= self.max_time: return","is not None: # Baseline + first batch + each pipeline iteration +","o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str,","if y_train is None: raise ValueError('Must specify training data target values as a","max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for","allowed in the search. The default of None indicates all pipelines for this","allowed pipeline class has been iterated over. If the number of unique pipelines","y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary,","is not None: logger.info(f\"Searching up to {self.max_batches} batches for a total of {self.max_iterations}","before search _baseline_cv_scores will be empty so we will return # nan for","results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \"\"\"Describe a pipeline Arguments: pipeline_id","`pipeline_results`: a dict with results from each pipeline, and `search_order`: a list describing","pipelines to search over per batch is one, ensembling will not run. Defaults","= self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \"id\": pipeline_id, \"pipeline_name\": pipeline.name,","with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \"\"\"Train a list","logger.info(\"Using default limit of max_batches=1.\\n\") if patience and (not isinstance(patience, int) or patience","needed to configure the search. For example, in time series problems, values should","of each pipeline family allowed in the search. \"\"\" if X_train is None:","self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives}","self._searched = True def _find_best_pipeline(self): \"\"\"Finds the best pipeline in the rankings If","has elapsed. If it is an integer, then the time will be in","Baseline Estimator\": {\"gap\": gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores =","iteration vs. score plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments.","Run at least one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines","isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\"Parameter max_time must be None","True, but the number of unique pipelines is one, so ensembling will not","self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name]","abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score > best_score if","parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\"Pipeline","c in [\"# Training\", \"# Validation\"]: all_objective_scores[c] = all_objective_scores[c].astype(\"object\") continue mean = all_objective_scores[c].mean(axis=0)","with results from each pipeline, and `search_order`: a list describing the order the","not self._best_pipeline: raise PipelineNotFoundError(\"automl search must be run before selecting `best_pipeline`.\") return self._best_pipeline","are allowed. Setting this field will cause allowed_model_families to be ignored. allowed_model_families (list(str,","to stop search early. Must be positive. If None, early stopping is disabled.","pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained","The default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\"binary\") to see",") from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from","in obj_list]) return '\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None","\").strip().lower() if choice == \"y\": logger.info(\"Exiting AutoMLSearch.\") return True elif choice == \"n\":","isinstance(patience, int) or patience < 0): raise ValueError(\"patience value must be a positive","[n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples].","categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration","pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\"Input for ensembler are pipelines with IDs: \" + str(pipeline_results['input_pipeline_ids']))","CV): %.1f seconds\" % pipeline_results[\"training_time\"]) log_subtitle(logger, \"Cross Validation\", underline=\"-\") all_objective_scores = [fold[\"all_objective_scores\"] for","Received {max_batches}.\") if max_iterations is not None and max_iterations < 0: raise ValueError(f\"Parameter","shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is","the highest-scoring set of parameters used with each pipeline.\"\"\" return self.full_rankings.drop_duplicates(subset=\"pipeline_name\", keep=\"first\") @property","training data as a 2d array using the X_train argument') if y_train is","pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is","0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed =","list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\")","three positional parameters: the Exception raised, the traceback, and the AutoMLSearch object. Must","is not compatible with a {} problem.\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or","pipelines to search. Parameters max_time, and max_iterations have precedence over stopping the search.","see options. Change `binary` to `multiclass` or `regression` depending on the problem type.","f\"{pipeline.name}\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \"...\" desc","on the same instance. Re-initialize AutoMLSearch to search again.\") return # don't show","parameters not found in automl results\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False):","to stop the search. Returns: bool: If True, search should terminate early \"\"\"","gap, \"max_delay\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds =","This can be helpful for training pipelines once the search is complete. Arguments:","train that pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns:","= \"\" def search(self, show_iteration_plot=True): \"\"\"Find the best pipeline for the data set.","instance of the specified pipeline initialized with the parameters used to train that","but max_iterations is too small, so ensembling will not run. Set max_iterations >=","log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter )","to configure the search. For example, in time series problems, values should be","\"\"\"Returns a trained instance of the best pipeline and parameters found during automl","of parallelism used for pipelines. None and 1 are equivalent. If set to","logger.warning(f\"Ensembling is set to True, but max_batches is too small, so ensembling will","perform as estimated on unseen data.\") return high_variance_cv def get_pipeline(self, pipeline_id): \"\"\"Given the","Exception. Callback function takes three positional parameters: the Exception raised, the traceback, and"],"string":"[\n \"[] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted:\",\n \"try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores\",\n \"percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id,\",\n \"not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) +\",\n \"requirement that automl search has been run. Arguments: pipeline (PipelineBase): pipeline to train\",\n \"> best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score\",\n \"Function called when `search()` errors and raises an Exception. Callback function takes three\",\n \"Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" )\",\n \"the X_train argument') if y_train is None: raise ValueError('Must specify training data target\",\n \"pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN:\",\n \"max_batches >= {ensemble_nth_batch + 1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run every\",\n \"objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and\",\n \"return False # check for early stopping if self.patience is None or self.tolerance\",\n \"{**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class,\",\n \"event add_to_rankings is called before search _baseline_cv_scores will be empty so we will\",\n \"self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value)\",\n \"o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise\",\n \"obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc\",\n \"or patience < 0): raise ValueError(\\\"patience value must be a positive integer. Received\",\n \"(list(str, ModelFamily)): The model families to search. The default of None searches over\",\n \"[fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if\",\n \"for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) /\",\n \"be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot in Jupyter\",\n \"= [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores:\",\n \"data target values as a 1d vector using the y_train argument') try: self.problem_type\",\n \"splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is not compatible with a\",\n \"post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for\",\n \"full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the current AutoML batch\",\n \"self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as\",\n \"logger.info(\\\"Input for ensembler are pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training\",\n \"better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score pipelines.\\\") if\",\n \"of the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results\",\n \"if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search early...\\\".format(self.patience)) return False\",\n \"None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines allowed in\",\n \"object at file path Arguments: file_path (str): location to save file pickle_protocol (int):\",\n \"(callable): Function called when `search()` errors and raises an Exception. Callback function takes\",\n \"if patience and (not isinstance(patience, int) or patience < 0): raise ValueError(\\\"patience value\",\n \"pipelines to train for every batch after the first one. The first batch\",\n \"time budget (if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return\",\n \"families: %s\\\\n\\\" % \\\", \\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if\",\n \"raise AutoMLSearchException(f\\\"All pipelines in the current AutoML batch produced a score of np.nan\",\n \"parameters needed to configure the search. For example, in time series problems, values\",\n \"self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline:\",\n \"data as a 2d array using the X_train argument') if y_train is None:\",\n \"and - R2 for regression problems. max_iterations (int): Maximum number of iterations to\",\n \"information such as type of pipeline components, problem, training time, cross validation, etc.\",\n \"Exception raised, the traceback, and the AutoMLSearch object. Must also accepts kwargs, so\",\n \"+ n_jobs) are used. ensembling (boolean): If True, runs ensembling in a separate\",\n \"self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1\",\n \"= y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def\",\n \"\\\"\\\"\\\"Fits a baseline pipeline to the data. This is the first pipeline fit\",\n \"'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates\",\n \"be specified as seconds, minutes, or hours. patience (int): Number of iterations without\",\n \"training results for the new pipeline, an untrained_pipeline containing the parameters used during\",\n \"False. max_batches (int): The maximum number of batches of pipelines to search. Parameters\",\n \"np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return\",\n \"{}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\",\",\n \"self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score\",\n \"errors and raises an Exception. Callback function takes three positional parameters: the Exception\",\n \"X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features]. Required. y_train\",\n \"seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\", \\\".join([model.value for model in\",\n \"False. Returns: Description of specified pipeline. Includes information such as type of pipeline\",\n \"also fields like \\\"# Training\\\" and \\\"# Testing\\\", so we want to exclude\",\n \"will be displayed in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout,\",\n \"(list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable,\",\n \"0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if\",\n \"\\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from all pipelines searched\\\"\\\"\\\" ascending = True\",\n \"= split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train,\",\n \"self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series\",\n \"curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change:\",\n \"have precedence over stopping the search. problem_configuration (dict, None): Additional parameters needed to\",\n \"additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] +\",\n \"'all_objective_scores' field contains scores for all objectives # but also fields like \\\"#\",\n \"in automl.\\\") return objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name)\",\n \"Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] ==\",\n \"return True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional\",\n \"break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True\",\n \"any pipelines that error out during training will not be included in the\",\n \"self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best\",\n \"best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True\",\n \"Setting this field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The\",\n \"2d array using the X_train argument') if y_train is None: raise ValueError('Must specify\",\n \"user asking if they want to stop the search. Returns: bool: If True,\",\n \"all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of var\\\", c] = std / mean if\",\n \"will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot\",\n \"obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in\",\n \"search. The default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to\",\n \"self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline\",\n \"= self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not\",\n \"Received {max_time}.\\\") if max_batches is not None and max_batches < 0: raise ValueError(f\\\"Parameter\",\n \"not be included in the dictionary but the exception and stacktrace will be\",\n \"features will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score\",\n \"if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger,\",\n \"minutes, or hours. patience (int): Number of iterations without improvement to stop search\",\n \"n_jobs (int or None): Non-negative integer describing level of parallelism used for pipelines.\",\n \"from the current best pipeline before training and thresholding\\\"\\\"\\\" if len(self.rankings) == 0:\",\n \"if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) //\",\n \"start_iteration_callback (callable): Function called before each pipeline training iteration. Callback function takes three\",\n \"appropriate parameters by default. Defaults to None, which will call `log_error_callback`. additional_objectives (list):\",\n \"if X_train is None: raise ValueError('Must specify training data as a 2d array\",\n \"the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original stopping criterion and\",\n \"None): Additional parameters needed to configure the search. For example, in time series\",\n \"the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able to pass\",\n \"None or non-negative. Received {max_batches}.\\\") if max_iterations is not None and max_iterations <\",\n \"{self.max_iterations} pipelines. \\\") elif self.max_iterations is not None: logger.info(\\\"Searching up to %s pipelines.\",\n \"(dict, None): Additional parameters needed to configure the search. For example, in time\",\n \"if max_iterations is not None and max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations must\",\n \"as score improvement for early stopping. Only applicable if patience is not None.\",\n \"import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators\",\n \"The first batch will train a baseline pipline + one of each pipeline\",\n \"= pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\\\"# Training\\\", \\\"# Validation\\\"]:\",\n \"thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and\",\n \"StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether\",\n \"to train and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in\",\n \"0): raise ValueError(\\\"patience value must be a positive integer. Received {} instead\\\".format(patience)) if\",\n \"copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from the highest-scoring\",\n \"non-negative. Received {max_time}.\\\") if max_batches is not None and max_batches < 0: raise\",\n \"return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs a\",\n \"= {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name\",\n \"best pipeline before training and thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return best_pipeline =\",\n \"make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives\",\n \"describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict\",\n \"found in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a\",\n \"/ n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time =\",\n \"at least one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines ==\",\n \"AutoML object at file path Arguments: file_path (str): location to save file pickle_protocol\",\n \"self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True,\",\n \"pipelines after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\", \\\".join([model.value\",\n \"logger.info(\\\"Objective to optimize binary classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including\",\n \"\\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family\",\n \"objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for objective, score in\",\n \"logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary\",\n \"be included in the dictionary but the exception and stacktrace will be displayed\",\n \"custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\\\"No allowed pipelines\",\n \"= self._num_pipelines() if num_pipelines == 0: return True # check max_time and max_iterations\",\n \"multiclass classification problems, and - R2 for regression problems. max_iterations (int): Maximum number\",\n \"random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__,\",\n \"to make sure it is different from the current best pipeline before training\",\n \"this loop does not count towards the time budget (if set) time_in_loop =\",\n \"= time.time() - start_of_loop self._start += time_in_loop return False else: leading_char = \\\"\\\"\",\n \"of np.nan on the primary objective {self.objective}.\\\") self.search_duration = time.time() - self._start elapsed_time\",\n \"pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for\",\n \"pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations,\",\n \"evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types\",\n \"if tolerance and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\\\"tolerance value\",\n \"pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name\",\n \"\\\"Do you really want to exit search (y/n)? \\\").strip().lower() if choice == \\\"y\\\":\",\n \"float]]: Dictionary keyed by pipeline name that maps to a dictionary of scores.\",\n \"(int): The number of pipelines to train for every batch after the first\",\n \"} self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try: self.plot =\",\n \"family allowed in the search. \\\"\\\"\\\" if X_train is None: raise ValueError('Must specify\",\n \"save file pickle_protocol (int): the pickle data stream format. Returns: None \\\"\\\"\\\" with\",\n \"specify training data target values as a 1d vector using the y_train argument')\",\n \"pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by\",\n \"pipelines that error out during training will not be included in the dictionary\",\n \"self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc =\",\n \"None: raise ValueError('Must specify training data as a 2d array using the X_train\",\n \"prompt to the user asking if they want to stop the search. Returns:\",\n \"to use. Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults to\",\n \"= infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type,\",\n \"ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import\",\n \"for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start\",\n \"a baseline pipeline to the data. This is the first pipeline fit during\",\n \"Note that if allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator):\",\n \"for training ensemble metalearners. Only used when ensembling is True. Must be between\",\n \"> AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number\",\n \"all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\",\",\n \"used. For n_jobs below -1, (n_cpus + 1 + n_jobs) are used. ensembling\",\n \"ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner\",\n \"= [] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines =\",\n \"cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback:\",\n \"that if allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data\",\n \"enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already been run and will not run\",\n \"or not all(p in problem_configuration for p in required_parameters): raise ValueError(\\\"user_parameters must be\",\n \"num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search\",\n \"not compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives =\",\n \"TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\":\",\n \"\\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df =\",\n \"import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import\",\n \"for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv\",\n \"pipeline components, problem, training time, cross validation, etc. \\\"\\\"\\\" if pipeline_id not in\",\n \"train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to the\",\n \"is set to False, returns an untrained pipeline instance. \\\"\\\"\\\" if not self._best_pipeline:\",\n \"the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout\",\n \"in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is not compatible with\",\n \"tolerance and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\\\"tolerance value must\",\n \"pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with the\",\n \"Defaults to False. max_batches (int): The maximum number of batches of pipelines to\",\n \"pass along other appropriate parameters by default. Defaults to None, which will call\",\n \"infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed)\",\n \"str, type(None))): raise TypeError(f\\\"Parameter max_time must be a float, int, string or None.\",\n \"{str(max_time)}..\\\") if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\\\"Parameter max_time must\",\n \"pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted =\",\n \"specified pipeline initialized with the parameters used to train that pipeline during automl\",\n \"that maps to the fitted pipeline. Note that the any pipelines that error\",\n \"len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0 <\",\n \"used to train that pipeline during automl search. Arguments: pipeline_id (int): pipeline to\",\n \"objective (str, ObjectiveBase): The objective to optimize for. Used to propose and rank\",\n \"\\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class):\",\n \"\\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df\",\n \"True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\",\",\n \"within cross validation scores. {pipeline_name} may not perform as estimated on unseen data.\\\")\",\n \"not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id]\",\n \"list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline\",\n \"= get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not\",\n \"evaluations which have been made Returns: int: the number of pipeline evaluations made\",\n \"yes, False if no. \\\"\\\"\\\" if self._interrupted: return False # for add_to_rankings if\",\n \"\\\" \\\"to get all objective names allowed in automl.\\\") return objective() return objective\",\n \"Description of specified pipeline. Includes information such as type of pipeline components, problem,\",\n \"must be between 0 and 1 exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _,\",\n \"{self.max_batches} batches for a total of {self.max_iterations} pipelines. \\\") elif self.max_iterations is not\",\n \"on the primary objective {self.objective}.\\\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start)\",\n \"during training will not be included in the dictionary but the exception and\",\n \"search. Parameters max_time, and max_iterations have precedence over stopping the search. problem_configuration (dict,\",\n \"pipeline to the data. This is the first pipeline fit during search. \\\"\\\"\\\"\",\n \"_pipelines_per_batch (int): The number of pipelines to train for every batch after the\",\n \"from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds,\",\n \"pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool):\",\n \"out during scoring will not be included in the dictionary but the exception\",\n \"tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to\",\n \"\\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\" % self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater'\",\n \"== ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger,\",\n \"positional parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback\",\n \"logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\", \\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot =\",\n \"for a full list. objective (str, ObjectiveBase): The objective to optimize for. Used\",\n \"start_of_loop = time.time() while True: choice = input(leading_char + \\\"Do you really want\",\n \"higher than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) >\",\n \"for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\\\"No allowed pipelines to\",\n \"return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list of pipelines\",\n \"as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object at file\",\n \"{_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size,\",\n \"in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective]\",\n \"Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed:\",\n \"all pipelines for this problem type are allowed. Setting this field will cause\",\n \"== \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice == \\\"n\\\": # So that\",\n \"def _should_continue(self): \\\"\\\"\\\"Given the original stopping criterion and current state, should the search\",\n \"search. For example, in time series problems, values should be passed in for\",\n \"if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch =\",\n \"set of parameters used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self):\",\n \"score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError:\",\n \"learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective\",\n \"return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int):\",\n \"and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able to\",\n \"from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results from each pipeline,\",\n \"to a dictionary of scores. Note that the any pipelines that error out\",\n \"Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with\",\n \"variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may not perform as\",\n \"X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None,\",\n \"= cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores =\",\n \"name that maps to the fitted pipeline. Note that the any pipelines that\",\n \"also accepts kwargs, so AutoMLSearch is able to pass along other appropriate parameters\",\n \"if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline {} is not compatible with problem_type\",\n \"pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\"\",\n \"if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML! \\\" \\\"Use\",\n \"_check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs a warning if variance\",\n \"return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file path Arguments:\",\n \"Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary` to `multiclass` or `regression` depending on\",\n \"self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and\",\n \"type(None))): raise TypeError(f\\\"Parameter max_time must be a float, int, string or None. Received\",\n \"isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch\",\n \"before selecting `best_pipeline`.\\\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at\",\n \"get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline )\",\n \"optimize for. Used to propose and rank pipelines, but not for optimizing each\",\n \"+ num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches -\",\n \"early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop = time.time() while True: choice = input(leading_char\",\n \"after the first one. The first batch will train a baseline pipline +\",\n \"evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types\",\n \"for o in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else:\",\n \"or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows an\",\n \"in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score\",\n \"problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback\",\n \"search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str,\",\n \"objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective)\",\n \"= self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved =\",\n \"None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\"\",\n \"(tolerance > 1.0 or tolerance < 0.0): raise ValueError(\\\"tolerance value must be a\",\n \"allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if\",\n \"the current AutoML batch produced a score of np.nan on the primary objective\",\n \"is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches\",\n \"each pipeline training iteration. Callback function takes three positional parameters: The pipeline class,\",\n \">= {threshold}) within cross validation scores. {pipeline_name} may not perform as estimated on\",\n \"self.problem_type: raise ValueError(\\\"Given pipeline {} is not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def\",\n \"will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold.\",\n \"objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results'])\",\n \"elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline =\",\n \"class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary for \\\"Plotting\\\" documentation,\",\n \"chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification problems,\",\n \"the user asking if they want to stop the search. Returns: bool: If\",\n \"may not perform as estimated on unseen data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id):\",\n \"[ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration\",\n \"the traceback, and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is\",\n \"in the rankings If self._best_pipeline already exists, check to make sure it is\",\n \"PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family ==\",\n \"self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None\",\n \"except ImportError: logger.warning(\\\"Unable to import plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines\",\n \"y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter,\",\n \"for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is not\",\n \"threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv >= {threshold}) within cross validation\",\n \"really want to exit search (y/n)? \\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\")\",\n \"a separate batch after every allowed pipeline class has been iterated over. If\",\n \"time can be specified as seconds, minutes, or hours. patience (int): Number of\",\n \"return True elif choice == \\\"n\\\": # So that the time in this\",\n \"specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if\",\n \"max_iterations >= {first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration}\",\n \"vs. score plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \\\"\\\"\\\"\",\n \"PipelineNotFoundError(\\\"Pipeline class or parameters not found in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def\",\n \"if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm =\",\n \"return # don't show iteration plot outside of a jupyter notebook if show_iteration_plot:\",\n \"None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration =\",\n \"and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\\\"Using default limit of\",\n \"(cv >= {threshold}) within cross validation scores. {pipeline_name} may not perform as estimated\",\n \"if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is not compatible with a {}\",\n \"(int, float, str, type(None))): raise TypeError(f\\\"Parameter max_time must be a float, int, string\",\n \"parameters used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a\",\n \") from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data\",\n \"included in the dictionary but the exception and stacktrace will be displayed in\",\n \"or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the user asking if they\",\n \"from each pipeline, and `search_order`: a list describing the order the pipelines were\",\n \"pipelines. This will not start a new pipeline search after the duration has\",\n \"\\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids =\",\n \"logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score pipelines.\\\") if self.max_batches is not None: logger.info(f\\\"Searching\",\n \"+ 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0:\",\n \"will train a baseline pipline + one of each pipeline family allowed in\",\n \"Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance:\",\n \"already been run and will not run again on the same instance. Re-initialize\",\n \"name that maps to a dictionary of scores. Note that the any pipelines\",\n \"with scoring results from the highest-scoring set of parameters used with each pipeline.\\\"\\\"\\\"\",\n \"early stopping. Only applicable if patience is not None. Defaults to None. allowed_pipelines\",\n \"to -1, all CPUs are used. For n_jobs below -1, (n_cpus + 1\",\n \"self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id]\",\n \"\\\" % self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__}\",\n \"valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is not compatible\",\n \"\\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start\",\n \"== ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else:\",\n \"= input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score\",\n \"f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not\",\n \"1: logger.warning(\\\"Ensembling is set to True, but the number of unique pipelines is\",\n \"Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random\",\n \"before returning it. Defaults to True. pipeline_parameters (dict): A dict of the parameters\",\n \"self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds =\",\n \"self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families\",\n \"= rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns\",\n \"optimize binary classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV): %.1f\",\n \"any pipelines that error out during scoring will not be included in the\",\n \"self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search early...\\\".format(self.patience)) return False return True def\",\n \"* self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is\",\n \"float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time, (int,\",\n \"the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a\",\n \"of objectives to score on. Will override default objectives for problem type if\",\n \"add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling\",\n \"ascending = False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if\",\n \"= 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score -\",\n \"\\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list of\",\n \"pipeline instance. Returns: PipelineBase: A trained instance of the best pipeline and parameters\",\n \"one of each pipeline family allowed in the search. \\\"\\\"\\\" if X_train is\",\n \"Received {type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time, (int, float)) and max_time < 0:\",\n \"a dictionary of scores. Note that the any pipelines that error out during\",\n \"if not empty. random_seed (int): Seed for the random number generator. Defaults to\",\n \"binary classification problems, - LogLossMulticlass for multiclass classification problems, and - R2 for\",\n \"run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split size must\",\n \"ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\\\"Ensembling is set to True,\",\n \"of the parameters used to initalize a pipeline with. _ensembling_split_size (float): The amount\",\n \"elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc)\",\n \"associated with the provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None:\",\n \"that allows access to a copy of the results from `automl_search`. Returns: dict\",\n \"and evaluates a given pipeline then adds the results to the automl rankings\",\n \"(including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"]\",\n \"to {[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters)\",\n \"pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c]\",\n \"for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c\",\n \"logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"]\",\n \"objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return\",\n \"returns an untrained instance of the specified pipeline initialized with the parameters used\",\n \"self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import plotly; skipping pipeline search plotting\\\\n\\\")\",\n \"self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total\",\n \"or `regression` depending on the problem type. Note that if allowed_pipelines is provided,\",\n \"the search. Returns: bool: If True, search should terminate early \\\"\\\"\\\" leading_char =\",\n \"get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from\",\n \"A list of PipelineBase subclasses indicating the pipelines allowed in the search. The\",\n \"model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary` to `multiclass` or `regression`\",\n \"the provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline\",\n \"percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id\",\n \"== ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif\",\n \"find file to load Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb') as f:\",\n \"a pipeline training result, returns an untrained instance of the specified pipeline initialized\",\n \"of parameters used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns\",\n \"int, string or None. Received {type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time, (int, float))\",\n \"self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else:\",\n \"ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type\",\n \"first one. The first batch will train a baseline pipline + one of\",\n \"parameters is None: raise PipelineNotFoundError(\\\"Pipeline class or parameters not found in automl results\\\")\",\n \"over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\\\")\",\n \"scores and logs a warning if variance is higher than specified threshhold.\\\"\\\"\\\" pipeline_name\",\n \"allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) ==\",\n \"ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type]\",\n \"\\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data)\",\n \"for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline\",\n \"use. Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner.\",\n \"don't show iteration plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython\",\n \"\\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original stopping criterion and current state,\",\n \"called before search _baseline_cv_scores will be empty so we will return # nan\",\n \"ImportError: logger.warning(\\\"Unable to import plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families\",\n \"be displayed in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout,\",\n \"with value {str(max_time)}..\\\") if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\\\"Parameter\",\n \"every allowed pipeline class has been iterated over. If the number of unique\",\n \"_, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine =\",\n \"during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline\",\n \"return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from\",\n \"be in seconds. For strings, time can be specified as seconds, minutes, or\",\n \"scores[field] += value return {objective: float(score) / n_folds for objective, score in scores.items()}\",\n \"is too small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch +\",\n \"the gap and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or {} def\",\n \"= objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id =\",\n \"containing the parameters used during training, and the AutoMLSearch object. error_callback (callable): Function\",\n \"the pickle data stream format. Returns: None \\\"\\\"\\\" with open(file_path, 'wb') as f:\",\n \"set aside for training ensemble metalearners. Only used when ensembling is True. Must\",\n \"X_train argument') if y_train is None: raise ValueError('Must specify training data target values\",\n \"best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self):\",\n \"max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations\",\n \"rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df\",\n \"self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid data\",\n \"self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train,\",\n \"np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] =\",\n \"received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type,\",\n \"list. objective (str, ObjectiveBase): The objective to optimize for. Used to propose and\",\n \"= time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines =\",\n \"self.max_batches = 1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if patience and (not isinstance(patience,\",\n \"{} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in\",\n \"X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if\",\n \"AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training iteration. Callback function\",\n \"the AutoMLSearch object. error_callback (callable): Function called when `search()` errors and raises an\",\n \"high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv >= {threshold}) within cross validation scores. {pipeline_name}\",\n \"= get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines =\",\n \"= abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score > best_score\",\n \"\\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\":\",\n \"self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search _baseline_cv_scores will be\",\n \"or not to train the best pipeline before returning it. Defaults to True.\",\n \"self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time,\",\n \"1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number\",\n \"{best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in the rankings\",\n \"which have been made Returns: int: the number of pipeline evaluations made in\",\n \"data we'll set aside for training ensemble metalearners. Only used when ensembling is\",\n \"for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o\",\n \"pipeline family allowed in the search. \\\"\\\"\\\" if X_train is None: raise ValueError('Must\",\n \"without improvement. Stopping search early...\\\".format(self.patience)) return False return True def _validate_problem_type(self): for obj\",\n \"= [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines ==\",\n \"run and will not run again on the same instance. Re-initialize AutoMLSearch to\",\n \"or hours. patience (int): Number of iterations without improvement to stop search early.\",\n \"(boolean, True): Shows an iteration vs. score plot in Jupyter notebook. Disabled by\",\n \"search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance. Returns:\",\n \"log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective ==\",\n \"training data of length [n_samples]. Required for supervised learning tasks. problem_type (str or\",\n \"ensemble metalearners. Only used when ensembling is True. Must be between 0 and\",\n \"< first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_iterations is\",\n \"and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type):\",\n \"= time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline()\",\n \"pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\\\"Pipeline class or\",\n \"if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification pipeline thresholds\",\n \"for the new pipeline, an untrained_pipeline containing the parameters used during training, and\",\n \"self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {}\",\n \"> self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score < best_score\",\n \"False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx\",\n \"batches of pipelines to search. Parameters max_time, and max_iterations have precedence over stopping\",\n \"numpy as np import pandas as pd import woodwork as ww from sklearn.model_selection\",\n \"\\\"\\\"\\\"Fits and evaluates a given pipeline then adds the results to the automl\",\n \"run before selecting `best_pipeline`.\\\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object\",\n \"if obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives\",\n \"or tolerance < 0.0): raise ValueError(\\\"tolerance value must be a float between 0.0\",\n \"default. Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom set of\",\n \"has been iterated over. If the number of unique pipelines to search over\",\n \"a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger,\",\n \"return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If\",\n \"kwargs, so AutoMLSearch is able to pass along other appropriate parameters by default.\",\n \"allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True,\",\n \"options. Change `binary` to `multiclass` or `regression` depending on the problem type. Note\",\n \"Only used when ensembling is True. Must be between 0 and 1, exclusive.\",\n \"best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score =\",\n \"= self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids\",\n \"to False, returns an untrained pipeline instance. \\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl\",\n \"self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type):\",\n \"get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None\",\n \"(pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features]. Required. y_train (pd.Series,\",\n \"- 3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is\",\n \"evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from\",\n \"A dictionary containing the training results for the new pipeline, an untrained_pipeline containing\",\n \"set of objectives for problem_type, remove it existing_main_objective = next((obj for obj in\",\n \"search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original stopping criterion and current\",\n \"(float): Minimum percentage difference to qualify as score improvement for early stopping. Only\",\n \"self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in\",\n \"them since # they are not scores if field in objective_name_to_class: scores[field] +=\",\n \"pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay =\",\n \"set. Arguments: feature_types (list, optional): list of feature types, either numerical or categorical.\",\n \"parameters used during training, and the AutoMLSearch object. error_callback (callable): Function called when\",\n \"raise ValueError(\\\"tolerance value must be a float between 0.0 and 1.0 inclusive. Received\",\n \"{first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration} iteration and\",\n \"random_seed self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError:\",\n \"= [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if not\",\n \"split size must be between 0 and 1 exclusive, received {_ensembling_split_size}\\\") X_shape =\",\n \"it is different from the current best pipeline before training and thresholding\\\"\\\"\\\" if\",\n \"PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective )\",\n \"to the fitted pipeline. Note that the any pipelines that error out during\",\n \"random_seed (int): Seed for the random number generator. Defaults to 0. n_jobs (int\",\n \"to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \\\") elif self.max_iterations is\",\n \"Dict[str, float]]: Dictionary keyed by pipeline name that maps to a dictionary of\",\n \"len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline ==\",\n \"True, but max_batches is too small, so ensembling will not run. Set max_batches\",\n \"is not compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives\",\n \"existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives]\",\n \"new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines\",\n \"a pandas.DataFrame with scoring results from all pipelines searched\\\"\\\"\\\" ascending = True if\",\n \"least one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines == 0:\",\n \"self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot = None\",\n \"be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class:\",\n \"= ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION:\",\n \"method to use. Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults\",\n \"self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if\",\n \"then max_iterations will default to max_iterations of 5. max_time (int, str): Maximum time\",\n \"add_to_rankings if self._searched: return True # Run at least one pipeline for every\",\n \"pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int): pipeline\",\n \"1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch\",\n \"ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\\\")\",\n \"f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\"\",\n \"= self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set to True,\",\n \"in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters}\",\n \"validation scores. {pipeline_name} may not perform as estimated on unseen data.\\\") return high_variance_cv\",\n \"evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters\",\n \"self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines\",\n \"is None or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score =\",\n \"axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None):\",\n \"the number of pipeline evaluations which have been made Returns: int: the number\",\n \"access to a copy of the results from `automl_search`. Returns: dict containing `pipeline_results`:\",\n \"displayed in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives):\",\n \"not None: logger.info(f\\\"Searching up to {self.max_batches} batches for a total of {self.max_iterations} pipelines.\",\n \"and current state, should the search continue? Returns: bool: True if yes, False\",\n \"def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function):\",\n \"\\\"\\\"\\\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads\",\n \"problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to the data. This\",\n \"self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches)\",\n \"True elif choice == \\\"n\\\": # So that the time in this loop\",\n \"type if not empty. random_seed (int): Seed for the random number generator. Defaults\",\n \"pipelines.\\\") if self.max_batches is not None: logger.info(f\\\"Searching up to {self.max_batches} batches for a\",\n \"= [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols)\",\n \"runs ensembling in a separate batch after every allowed pipeline class has been\",\n \"evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines\",\n \"self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class =\",\n \"each pipeline, and `search_order`: a list describing the order the pipelines were searched.\",\n \"- self._start if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and\",\n \"used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame\",\n \"so AutoMLSearch is able to pass along other appropriate parameters by default. Defaults\",\n \"problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or\",\n \"inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of\",\n \"self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline,\",\n \"the time budget (if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop\",\n \"max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data,\",\n \"self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if\",\n \"[\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std =\",\n \"for the data set. Arguments: feature_types (list, optional): list of feature types, either\",\n \"X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train)\",\n \"= self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"):\",\n \"one, so ensembling will not run.\\\") run_ensembling = False if run_ensembling and self.max_iterations\",\n \"and self.max_iterations is not None: # Baseline + first batch + each pipeline\",\n \"in time series problems, values should be passed in for the gap and\",\n \"_ensembling_split_size (float): The amount of the training data we'll set aside for training\",\n \"the training results for the new pipeline, an untrained_pipeline containing the parameters used\",\n \"(list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline\",\n \"= self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train\",\n \"for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data\",\n \"format. Returns: None \\\"\\\"\\\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod\",\n \"instance. Returns: PipelineBase: A trained instance of the best pipeline and parameters found\",\n \"cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0]\",\n \"o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives\",\n \"not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is not compatible with a {} problem.\\\".format(self.objective.name,\",\n \"pipeline name that maps to the fitted pipeline. Note that the any pipelines\",\n \"of variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may not perform\",\n \"cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object at file path Arguments:\",\n \"leading_char = \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for the data\",\n \"NameError: show_iteration_plot = False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\" %\",\n \"training and thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not\",\n \"ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted =\",\n \"the number of unique pipelines is one, so ensembling will not run.\\\") run_ensembling\",\n \"= data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective\",\n \"or parameters is None: raise PipelineNotFoundError(\\\"Pipeline class or parameters not found in automl\",\n \"called when `search()` errors and raises an Exception. Callback function takes three positional\",\n \"@property def results(self): \\\"\\\"\\\"Class that allows access to a copy of the results\",\n \"False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']:\",\n \"file pickle_protocol (int): the pickle data stream format. Returns: None \\\"\\\"\\\" with open(file_path,\",\n \"ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc =\",\n \"Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives\",\n \"pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and\",\n \"'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter\",\n \"current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines\",\n \"for every batch after the first one. The first batch will train a\",\n \"Maximum time to search for pipelines. This will not start a new pipeline\",\n \"and max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations must be None or non-negative. Received\",\n \"scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps to\",\n \"search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is\",\n \"= get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and not\",\n \"in the search. \\\"\\\"\\\" if X_train is None: raise ValueError('Must specify training data\",\n \"automl.\\\") return objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for\",\n \"early...\\\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in self.additional_objectives: if not\",\n \"max_iterations must be None or non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time,\",\n \"is not allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective\",\n \"for every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True #\",\n \"\\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\":\",\n \"pipeline instance associated with the provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results\",\n \"self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return\",\n \"== 0: return True # check max_time and max_iterations elapsed = time.time() -\",\n \"if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot:\",\n \"False # for add_to_rankings if self._searched: return True # Run at least one\",\n \"ValueError(\\\"Given pipeline {} is not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits\",\n \"not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\\\"Parameter max_time must be a float,\",\n \"if score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement +=\",\n \"to exclude them since # they are not scores if field in objective_name_to_class:\",\n \"pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to\",\n \"False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train,\",\n \"if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed in\",\n \"self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import plotly;\",\n \"self._interrupted: return False # for add_to_rankings if self._searched: return True # Run at\",\n \"1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\\\"Ensembling is set\",\n \"parameters by default. Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom\",\n \"of None indicates all pipelines for this problem type are allowed. Setting this\",\n \"search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated\",\n \"problems, - LogLossMulticlass for multiclass classification problems, and - R2 for regression problems.\",\n \"use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary\",\n \"Whether or not to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback\",\n \"is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]:\",\n \"variables. train_best_pipeline (boolean): Whether or not to train the best pipeline before returning\",\n \"\\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective names allowed in automl.\\\") return\",\n \"current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try:\",\n \"list of PipelineBase subclasses indicating the pipelines allowed in the search. The default\",\n \"default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options.\",\n \"for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize =\",\n \"train_best_pipeline (boolean): Whether or not to train the best pipeline before returning it.\",\n \"protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object at file path Arguments: file_path (str):\",\n \"search continue? Returns: bool: True if yes, False if no. \\\"\\\"\\\" if self._interrupted:\",\n \"value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives #\",\n \"score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score\",\n \"pipelines. \\\" % self.max_iterations) if self.max_time is not None: logger.info(\\\"Will stop searching for\",\n \"isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML!\",\n \"training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training\",\n \"a trained instance of the best pipeline and parameters found during automl search.\",\n \"is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid data splitter\\\")\",\n \"self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score = curr_score num_without_improvement\",\n \"max_time and max_iterations elapsed = time.time() - self._start if self.max_time and elapsed >=\",\n \"ValueError(\\\"Given objective {} is not compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives\",\n \"@staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in\",\n \"if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks\",\n \"results(self): \\\"\\\"\\\"Class that allows access to a copy of the results from `automl_search`.\",\n \"best pipeline and parameters found during automl search. If `train_best_pipeline` is set to\",\n \"describe return_dict (bool): If True, return dictionary of information about pipeline. Defaults to\",\n \"max_iterations have precedence over stopping the search. problem_configuration (dict, None): Additional parameters needed\",\n \"documentation, since Sphinx does not work well with instance attributes. plot = PipelineSearchPlots\",\n \"dict containing values for at least the gap and max_delay \\\" f\\\"parameters. Received\",\n \"\\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY)\",\n \"f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if\",\n \"since # they are not scores if field in objective_name_to_class: scores[field] += value\",\n \"pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch\",\n \"\\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\",\",\n \"\\\"\\\"\\\"Train a list of pipelines on the training data. This can be helpful\",\n \"[])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\"\",\n \"= pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\\\"Pipeline class\",\n \"self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted\",\n \"Returns: dict containing `pipeline_results`: a dict with results from each pipeline, and `search_order`:\",\n \"compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type)\",\n \"parameters used to initalize a pipeline with. _ensembling_split_size (float): The amount of the\",\n \"y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices]\",\n \"None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid data splitter\\\") if not\",\n \"pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True\",\n \"\\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id)\",\n \"problems. max_iterations (int): Maximum number of iterations to search. If max_iterations and max_time\",\n \"== self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o)\",\n \"self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if\",\n \"max_iterations is not None and max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations must be\",\n \"Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches:\",\n \"@property def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of the best pipeline and parameters\",\n \"self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train =\",\n \"'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object at\",\n \"cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score =\",\n \"False if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators = get_estimators(self.problem_type,\",\n \"each pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary for binary\",\n \"max_iterations and max_time is not set, then max_iterations will default to max_iterations of\",\n \"validation, etc. \\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline\",\n \"{} is not compatible with a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines\",\n \"convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger\",\n \"ValueError('Must specify training data as a 2d array using the X_train argument') if\",\n \"the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results from\",\n \"ww.DataTable): The input training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn):\",\n \"they are not scores if field in objective_name_to_class: scores[field] += value return {objective:\",\n \"with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None,\",\n \"open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of\",\n \"pipeline. Defaults to False. Returns: Description of specified pipeline. Includes information such as\",\n \"evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective,\",\n \"True if yes, False if no. \\\"\\\"\\\" if self._interrupted: return False # for\",\n \"self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score =\",\n \"each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring\",\n \"if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a given pipeline\",\n \"== ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if\",\n \"{self.objective}.\\\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch finished\",\n \"initalize a pipeline with. _ensembling_split_size (float): The amount of the training data we'll\",\n \"for pipelines. This will not start a new pipeline search after the duration\",\n \"on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train.\",\n \"choice == \\\"n\\\": # So that the time in this loop does not\",\n \"self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation\",\n \"@property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from all pipelines searched\\\"\\\"\\\"\",\n \"if yes, False if no. \\\"\\\"\\\" if self._interrupted: return False # for add_to_rankings\",\n \"problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None,\",\n \"not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives =\",\n \"max_delay variables. train_best_pipeline (boolean): Whether or not to train the best pipeline before\",\n \"for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline':\",\n \"elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False\",\n \"pipeline_class is None or parameters is None: raise PipelineNotFoundError(\\\"Pipeline class or parameters not\",\n \"(int or None): Non-negative integer describing level of parallelism used for pipelines. None\",\n \"+ len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else:\",\n \"a given pipeline then adds the results to the automl rankings with the\",\n \"skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None\",\n \"must be run before selecting `best_pipeline`.\\\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves\",\n \"the best pipeline for the data set. Arguments: feature_types (list, optional): list of\",\n \"Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps to a dictionary\",\n \"if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid\",\n \"but not for optimizing each pipeline during fit-time. When set to 'auto', chooses:\",\n \"train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input\",\n \"if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline,\",\n \"ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline,\",\n \"or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback\",\n \"to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The default\",\n \"\\\"\\\\n\\\" start_of_loop = time.time() while True: choice = input(leading_char + \\\"Do you really\",\n \"from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from\",\n \"+ 1 + n_jobs) are used. ensembling (boolean): If True, runs ensembling in\",\n \"array using the X_train argument') if y_train is None: raise ValueError('Must specify training\",\n \"every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True # check\",\n \"'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs\",\n \"so we will return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name],\",\n \"pipeline search after the duration has elapsed. If it is an integer, then\",\n \"AutoMLSearch object. error_callback (callable): Function called when `search()` errors and raises an Exception.\",\n \"to optimize binary classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV):\",\n \"= time.time() while True: choice = input(leading_char + \\\"Do you really want to\",\n \"want to exit search (y/n)? \\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return\",\n \"None, which will call `log_error_callback`. additional_objectives (list): Custom set of objectives to score\",\n \"supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The\",\n \"Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary pipeline\",\n \"recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted\",\n \"asking if they want to stop the search. Returns: bool: If True, search\",\n \"return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original stopping criterion and current state, should\",\n \"a copy of the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict\",\n \"str): Maximum time to search for pipelines. This will not start a new\",\n \"0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score)\",\n \"training data. This can be helpful for training pipelines once the search is\",\n \"current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the current\",\n \"return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time:\",\n \"ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if\",\n \"None: logger.info(\\\"Searching up to %s pipelines. \\\" % self.max_iterations) if self.max_time is not\",\n \"= pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High coefficient\",\n \"pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits\",\n \"pandas.DataFrame with scoring results from all pipelines searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better:\",\n \"TypeError(f\\\"Parameter max_time must be a float, int, string or None. Received {type(max_time)} with\",\n \"\\\"# Testing\\\", so we want to exclude them since # they are not\",\n \"object. Must also accepts kwargs, so AutoMLSearch is able to pass along other\",\n \"list of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of\",\n \"pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines with IDs: \\\"\",\n \"of batches of pipelines to search. Parameters max_time, and max_iterations have precedence over\",\n \"which will call `log_error_callback`. additional_objectives (list): Custom set of objectives to score on.\",\n \"add_to_rankings is called before search _baseline_cv_scores will be empty so we will return\",\n \"if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc\",\n \"error_callback (callable): Function called when `search()` errors and raises an Exception. Callback function\",\n \"# So that the time in this loop does not count towards the\",\n \"PipelineBase subclasses indicating the pipelines allowed in the search. The default of None\",\n \"must be a float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\\\")\",\n \"\\\"validation_score\\\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in\",\n \"pipeline name that maps to a dictionary of scores. Note that the any\",\n \"search early. Must be positive. If None, early stopping is disabled. Defaults to\",\n \"will call `log_error_callback`. additional_objectives (list): Custom set of objectives to score on. Will\",\n \"to retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided ID \\\"\\\"\\\"\",\n \"be positive. If None, early stopping is disabled. Defaults to None. tolerance (float):\",\n \"that the any pipelines that error out during training will not be included\",\n \"as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback\",\n \"[self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize\",\n \"and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) +\",\n \"to search again.\\\") return # don't show iteration plot outside of a jupyter\",\n \"pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self,\",\n \"\\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df =\",\n \"{self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\"\",\n \"True # check max_time and max_iterations elapsed = time.time() - self._start if self.max_time\",\n \"if self.objective.greater_is_better else curr_score < best_score if score_improved and significant_change: best_score = curr_score\",\n \"(0 < _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split size must be between 0\",\n \"objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0,\",\n \"f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs:\",\n \"{self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\"\",\n \"with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to the data.\",\n \"n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters,\",\n \"Defaults to None. tolerance (float): Minimum percentage difference to qualify as score improvement\",\n \"compatible with a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if\",\n \"for new pipelines after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" %\",\n \"# In the event add_to_rankings is called before search _baseline_cv_scores will be empty\",\n \"= self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return\",\n \"criterion and current state, should the search continue? Returns: bool: True if yes,\",\n \"// ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\\\"Ensembling is set to\",\n \"(bool): Whether or not to optimize the binary pipeline threshold. Defaults to True.\",\n \"and 1 exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ =\",\n \"= pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\":\",\n \"classification problems, and - R2 for regression problems. max_iterations (int): Maximum number of\",\n \"it existing_main_objective = next((obj for obj in additional_objectives if obj.name == self.objective.name), None)\",\n \"set to True, but max_iterations is too small, so ensembling will not run.\",\n \"disabled. Defaults to None. tolerance (float): Minimum percentage difference to qualify as score\",\n \"new pipeline, an untrained_pipeline containing the parameters used during training, and the AutoMLSearch\",\n \"None and max_batches < 0: raise ValueError(f\\\"Parameter max_batches must be None or non-negative.\",\n \"objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()}\",\n \"self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective\",\n \"max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None,\",\n \"to optimize for. Used to propose and rank pipelines, but not for optimizing\",\n \"duration has elapsed. If it is an integer, then the time will be\",\n \"True, search should terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop = time.time() while\",\n \"and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\\\"tolerance value must be\",\n \"data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner\",\n \"this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults\",\n \"number of unique pipelines to search over per batch is one, ensembling will\",\n \"ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline\",\n \"can be helpful for training pipelines once the search is complete. Arguments: pipelines\",\n \"run again on the same instance. Re-initialize AutoMLSearch to search again.\\\") return #\",\n \"allowed in the search. \\\"\\\"\\\" if X_train is None: raise ValueError('Must specify training\",\n \"False logger.warning(f\\\"Ensembling is set to True, but max_iterations is too small, so ensembling\",\n \"= curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if score_improved and\",\n \"if not problem_configuration or not all(p in problem_configuration for p in required_parameters): raise\",\n \"train the best pipeline before returning it. Defaults to True. pipeline_parameters (dict): A\",\n \"desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name =\",\n \"(int, str): Maximum time to search for pipelines. This will not start a\",\n \"an untrained instance of the specified pipeline initialized with the parameters used to\",\n \"pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True, return dictionary\",\n \"= self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else\",\n \"classification problems, - LogLossMulticlass for multiclass classification problems, and - R2 for regression\",\n \"exception and stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def\",\n \"pipeline evaluations made in the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the\",\n \"set to False, returns an untrained pipeline instance. Returns: PipelineBase: A trained instance\",\n \"def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data:\",\n \"(self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size <\",\n \"in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for objective, score\",\n \"found in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is\",\n \"= bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv\",\n \"fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in\",\n \"self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = []\",\n \"a total of {self.max_iterations} pipelines. \\\") elif self.max_iterations is not None: logger.info(\\\"Searching up\",\n \"= tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': []\",\n \"* (self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size < 1):\",\n \"time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if\",\n \"Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed\",\n \"When set to 'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass\",\n \"pipeline instance. \\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be run before\",\n \"default in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already been run and\",\n \"the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\") if self.max_batches\",\n \"self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline evaluations which\",\n \"handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class\",\n \"Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\"\",\n \"cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv >= {threshold}) within\",\n \"to search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator\",\n \"rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc\",\n \"[] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch()\",\n \"elapsed. If it is an integer, then the time will be in seconds.\",\n \"same instance. Re-initialize AutoMLSearch to search again.\\\") return # don't show iteration plot\",\n \"num_pipelines == 0: return True # check max_time and max_iterations elapsed = time.time()\",\n \"List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame):\",\n \"import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from\",\n \"{ensemble_nth_batch} batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1\",\n \"= input(leading_char + \\\"Do you really want to exit search (y/n)? \\\").strip().lower() if\",\n \"problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main\",\n \"returning it. Defaults to True. pipeline_parameters (dict): A dict of the parameters used\",\n \"pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list of pipelines on the given holdout\",\n \"all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results\",\n \"pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\":\",\n \"amount of the training data we'll set aside for training ensemble metalearners. Only\",\n \"to exit search (y/n)? \\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True\",\n \"max_iterations will default to max_iterations of 5. max_time (int, str): Maximum time to\",\n \"pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that\",\n \"for early stopping. Only applicable if patience is not None. Defaults to None.\",\n \"iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch +\",\n \"in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe()\",\n \"or non-negative. Received {max_batches}.\\\") if max_iterations is not None and max_iterations < 0:\",\n \"best pipeline in the rankings If self._best_pipeline already exists, check to make sure\",\n \"0: return True # check max_time and max_iterations elapsed = time.time() - self._start\",\n \"} if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"]\",\n \"[\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df\",\n \"specified as seconds, minutes, or hours. patience (int): Number of iterations without improvement\",\n \"max_iterations is too small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration}\",\n \"Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True, return dictionary of\",\n \"pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to\",\n \"rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of the\",\n \"while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm\",\n \"= ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations:\",\n \"raise ValueError(\\\"Given pipeline {} is not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self):\",\n \"ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective\",\n \"all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in\",\n \"not scores if field in objective_name_to_class: scores[field] += value return {objective: float(score) /\",\n \"of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of\",\n \"< 0): raise ValueError(\\\"patience value must be a positive integer. Received {} instead\\\".format(patience))\",\n \"if self.max_batches is not None: logger.info(f\\\"Searching up to {self.max_batches} batches for a total\",\n \"self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores\",\n \"update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives\",\n \"small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\\\")\",\n \"\\\"\\\"\\\"Score a list of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)):\",\n \"parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to\",\n \"with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML\",\n \"allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {}\",\n \"searching for new pipelines after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\"\",\n \"if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be run before selecting `best_pipeline`.\\\") return\",\n \"len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original stopping criterion and current state, should the\",\n \"> threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv >= {threshold}) within cross\",\n \"import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils\",\n \"{self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if not self.rankings.empty: rankings_str =\",\n \"the best pipeline before returning it. Defaults to True. pipeline_parameters (dict): A dict\",\n \"optional): list of feature types, either numerical or categorical. Categorical features will automatically\",\n \"will default to max_iterations of 5. max_time (int, str): Maximum time to search\",\n \"return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that allows access to a copy\",\n \"to None, which will call `log_error_callback`. additional_objectives (list): Custom set of objectives to\",\n \"of pipelines to search. Parameters max_time, and max_iterations have precedence over stopping the\",\n \"fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains\",\n \"evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score\",\n \"evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from\",\n \"MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils\",\n \"start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds\",\n \"cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline),\",\n \"to see options. Change `binary` to `multiclass` or `regression` depending on the problem\",\n \"evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective to optimize for.\",\n \"default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or\",\n \"desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives =\",\n \"and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the current AutoML batch produced a score\",\n \"gap and max_delay variables. train_best_pipeline (boolean): Whether or not to train the best\",\n \"If True, return dictionary of information about pipeline. Defaults to False. Returns: Description\",\n \"raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family\",\n \"does not work well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None,\",\n \"additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for\",\n \"= self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def\",\n \"import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import\",\n \"for %s. \\\" % self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower'))\",\n \"np.nan on the primary objective {self.objective}.\\\") self.search_duration = time.time() - self._start elapsed_time =\",\n \"pipeline then adds the results to the automl rankings with the requirement that\",\n \"max_batches (int): The maximum number of batches of pipelines to search. Parameters max_time,\",\n \"Disabled by default in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already been\",\n \"or not to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable):\",\n \"batch + each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) +\",\n \"if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train\",\n \"in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before\",\n \"IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and\",\n \"The maximum number of batches of pipelines to search. Parameters max_time, and max_iterations\",\n \"of length [n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes): type\",\n \"p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set\",\n \"`search()` errors and raises an Exception. Callback function takes three positional parameters: the\",\n \"Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations\",\n \"objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names()\",\n \"targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str,\",\n \"pipeline): \\\"\\\"\\\"Fits and evaluates a given pipeline then adds the results to the\",\n \"additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in\",\n \"to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1\",\n \"series problems, values should be passed in for the gap and max_delay variables.\",\n \"+ 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if\",\n \"search. problem_configuration (dict, None): Additional parameters needed to configure the search. For example,\",\n \"isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\\\"Parameter max_time must be a float, int,\",\n \"try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for\",\n \"Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string()\",\n \"the best pipeline and parameters found during automl search. If `train_best_pipeline` is set\",\n \"indicating the pipelines allowed in the search. The default of None indicates all\",\n \"= { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\":\",\n \"+ each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines)\",\n \"metalearners. Only used when ensembling is True. Must be between 0 and 1,\",\n \"\\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop = time.time() while True: choice = input(leading_char +\",\n \"if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if\",\n \"self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output\",\n \"will not run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\\\") else:\",\n \"that error out during scoring will not be included in the dictionary but\",\n \"and (not isinstance(patience, int) or patience < 0): raise ValueError(\\\"patience value must be\",\n \"or None): Non-negative integer describing level of parallelism used for pipelines. None and\",\n \"\\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\",\",\n \"early stopping is disabled. Defaults to None. tolerance (float): Minimum percentage difference to\",\n \"rankings with the requirement that automl search has been run. Arguments: pipeline (PipelineBase):\",\n \"def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of the best pipeline and parameters found\",\n \"False, returns an untrained pipeline instance. \\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search\",\n \"= self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling,\",\n \"and rank pipelines, but not for optimizing each pipeline during fit-time. When set\",\n \"pipeline before returning it. Defaults to True. pipeline_parameters (dict): A dict of the\",\n \"pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN -\",\n \"the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv =\",\n \"or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed\",\n \"search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator in\",\n \"automl rankings with the requirement that automl search has been run. Arguments: pipeline\",\n \"callable(function): return function.__name__ else: return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\"\",\n \"`regression` depending on the problem type. Note that if allowed_pipelines is provided, this\",\n \"(sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner class\",\n \"the ID of a pipeline training result, returns an untrained instance of the\",\n \"\\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not found in\",\n \"\\\") elif self.max_iterations is not None: logger.info(\\\"Searching up to %s pipelines. \\\" %\",\n \"X_train is None: raise ValueError('Must specify training data as a 2d array using\",\n \"path Arguments: file_path (str): location to find file to load Returns: AutoSearchBase object\",\n \"used. ensembling (boolean): If True, runs ensembling in a separate batch after every\",\n \"= 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without\",\n \"choice = input(leading_char + \\\"Do you really want to exit search (y/n)? \\\").strip().lower()\",\n \"is an integer, then the time will be in seconds. For strings, time\",\n \"dictionary but the exception and stacktrace will be displayed in the log. \\\"\\\"\\\"\",\n \"self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2):\",\n \"if no. \\\"\\\"\\\" if self._interrupted: return False # for add_to_rankings if self._searched: return\",\n \"'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score pipelines.\\\") if self.max_batches is not None:\",\n \"searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary` to\",\n \"a warning if variance is higher than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv\",\n \"p in required_parameters): raise ValueError(\\\"user_parameters must be a dict containing values for at\",\n \"logger.warning(f\\\"High coefficient of variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may\",\n \"{o.name: o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float,\",\n \"will be empty so we will return # nan for the base score.\",\n \"example, in time series problems, values should be passed in for the gap\",\n \"tolerance (float): Minimum percentage difference to qualify as score improvement for early stopping.\",\n \"cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info]\",\n \"else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1],\",\n \"if c in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean =\",\n \"is None: raise ValueError('Must specify training data target values as a 1d vector\",\n \"{self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience:\",\n \"loop_interrupted = False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except\",\n \"check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily\",\n \"self._pipelines_per_batch} iterations after that.\\\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True\",\n \"Re-initialize AutoMLSearch to search again.\\\") return # don't show iteration plot outside of\",\n \"sure it is different from the current best pipeline before training and thresholding\\\"\\\"\\\"\",\n \"evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils\",\n \"run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name']\",\n \"strings, time can be specified as seconds, minutes, or hours. patience (int): Number\",\n \"Defaults to False. Returns: Description of specified pipeline. Includes information such as type\",\n \"a list of pipelines on the training data. This can be helpful for\",\n \"from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils\",\n \"data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective =\",\n \"self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params\",\n \"specify training data as a 2d array using the X_train argument') if y_train\",\n \"not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline\",\n \"ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling\",\n \"= \\\"\\\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results:\",\n \"# nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] =\",\n \"count towards the time budget (if set) time_in_loop = time.time() - start_of_loop self._start\",\n \"return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check for\",\n \"or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective\",\n \"= -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass\",\n \"< 1): raise ValueError(f\\\"Ensembling split size must be between 0 and 1 exclusive,\",\n \"self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out\",\n \"PipelineNotFoundError(\\\"automl search must be run before selecting `best_pipeline`.\\\") return self._best_pipeline def save(self, file_path,\",\n \"for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]:\",\n \"base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline,\",\n \"def _num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline evaluations which have been made Returns:\",\n \"file to load Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb') as f: return\",\n \"ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\",\",\n \"Parameters max_time, and max_iterations have precedence over stopping the search. problem_configuration (dict, None):\",\n \"pipeline_parameters (dict): A dict of the parameters used to initalize a pipeline with.\",\n \"(not isinstance(patience, int) or patience < 0): raise ValueError(\\\"patience value must be a\",\n \"= False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\" % self.objective.name) logger.info(\\\"{}\",\n \"* (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 +\",\n \"self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def _validate_problem_configuration(self,\",\n \"\\\"\\\"\\\" if self._interrupted: return False # for add_to_rankings if self._searched: return True #\",\n \"pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs a warning if variance is\",\n \"cross-validation scores and logs a warning if variance is higher than specified threshhold.\\\"\\\"\\\"\",\n \"TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from\",\n \"file_path (str): location to find file to load Returns: AutoSearchBase object \\\"\\\"\\\" with\",\n \"n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import\",\n \"'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try:\",\n \"loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings =\",\n \"all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self,\",\n \"If it is an integer, then the time will be in seconds. For\",\n \"raise ValueError(f\\\"Ensembling split size must be between 0 and 1 exclusive, received {_ensembling_split_size}\\\")\",\n \"c in all_objective_scores: if c in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\")\",\n \"raise ValueError(\\\"patience value must be a positive integer. Received {} instead\\\".format(patience)) if tolerance\",\n \"self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p\",\n \"stopping the search. problem_configuration (dict, None): Additional parameters needed to configure the search.\",\n \"sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__\",\n \"feature types, either numerical or categorical. Categorical features will automatically be encoded show_iteration_plot\",\n \"= ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION:\",\n \"maximum number of batches of pipelines to search. Parameters max_time, and max_iterations have\",\n \"self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)]))\",\n \"terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop = time.time() while True: choice =\",\n \"to score on. Will override default objectives for problem type if not empty.\",\n \"[], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None\",\n \"import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle,\",\n \"logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds\",\n \"for the gap and max_delay variables. train_best_pipeline (boolean): Whether or not to train\",\n \"the parameters used to initalize a pipeline with. _ensembling_split_size (float): The amount of\",\n \"self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be run before selecting `best_pipeline`.\\\") return self._best_pipeline def\",\n \"when `search()` errors and raises an Exception. Callback function takes three positional parameters:\",\n \"baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class =\",\n \"from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique,\",\n \"first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations\",\n \"0: raise ValueError(f\\\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\\\") self.max_time =\",\n \"_baseline_cv_scores will be empty so we will return # nan for the base\",\n \"after every allowed pipeline class has been iterated over. If the number of\",\n \"def load(file_path): \\\"\\\"\\\"Loads AutoML object at file path Arguments: file_path (str): location to\",\n \"= allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output =\",\n \"max_batches is too small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch\",\n \"None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter\",\n \"self.tolerance is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement =\",\n \"traceback, and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able\",\n \"return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name]\",\n \"self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback =\",\n \"# Run at least one pipeline for every search num_pipelines = self._num_pipelines() if\",\n \"pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs a warning\",\n \"= tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback\",\n \"stopping is disabled. Defaults to None. tolerance (float): Minimum percentage difference to qualify\",\n \"Callback function takes three positional parameters: The pipeline class, the pipeline parameters, and\",\n \"return objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o\",\n \"if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices]\",\n \"search must be run before selecting `best_pipeline`.\\\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL):\",\n \"plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None,\",\n \"used for pipelines. None and 1 are equivalent. If set to -1, all\",\n \"\\\", \\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot =\",\n \"were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring\",\n \"f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives\",\n \"target values as a 1d vector using the y_train argument') try: self.problem_type =\",\n \"thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger,\",\n \"will be in seconds. For strings, time can be specified as seconds, minutes,\",\n \"1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search early...\\\".format(self.patience)) return\",\n \"False logger.warning(f\\\"Ensembling is set to True, but max_batches is too small, so ensembling\",\n \"logger.info(f\\\"Searching up to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \\\") elif\",\n \"self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches =\",\n \"pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a given pipeline then adds the\",\n \"def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from all pipelines searched\\\"\\\"\\\" ascending\",\n \"ValueError(f\\\"Parameter max_time must be None or non-negative. Received {max_time}.\\\") if max_batches is not\",\n \"if self.objective.greater_is_better: ascending = False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\",\",\n \"ensembling will not run.\\\") run_ensembling = False if run_ensembling and self.max_iterations is not\",\n \"self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\":\",\n \"= train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train =\",\n \"training time (including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores\",\n \"logger.info(\\\"Total training time (including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\")\",\n \"from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from\",\n \"in problem_configuration for p in required_parameters): raise ValueError(\\\"user_parameters must be a dict containing\",\n \"convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch\",\n \"LogLossMulticlass for multiclass classification problems, and - R2 for regression problems. max_iterations (int):\",\n \"it. Defaults to True. pipeline_parameters (dict): A dict of the parameters used to\",\n \"start a new pipeline search after the duration has elapsed. If it is\",\n \"self.allowed_pipelines == []: raise ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling\",\n \"(float): The amount of the training data we'll set aside for training ensemble\",\n \"max_time must be a float, int, string or None. Received {type(max_time)} with value\",\n \"search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features].\",\n \"num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search early...\\\".format(self.patience)) return False return\",\n \"defaultdict import cloudpickle import numpy as np import pandas as pd import woodwork\",\n \"since Sphinx does not work well with instance attributes. plot = PipelineSearchPlots def\",\n \"This will not start a new pipeline search after the duration has elapsed.\",\n \"self.patience = patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {},\",\n \"the duration has elapsed. If it is an integer, then the time will\",\n \"input training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target\",\n \"evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class\",\n \"self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\\\"No allowed\",\n \"( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import\",\n \"for training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List of\",\n \"current AutoML batch produced a score of np.nan on the primary objective {self.objective}.\\\")\",\n \"current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if\",\n \"file path Arguments: file_path (str): location to find file to load Returns: AutoSearchBase\",\n \"self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines)\",\n \"not empty. random_seed (int): Seed for the random number generator. Defaults to 0.\",\n \"difference to qualify as score improvement for early stopping. Only applicable if patience\",\n \"f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object at file path\",\n \"else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping\",\n \"to True, but max_iterations is too small, so ensembling will not run. Set\",\n \"y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline\",\n \"= 40 # Necessary for \\\"Plotting\\\" documentation, since Sphinx does not work well\",\n \"regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback =\",\n \"{ensemble_nth_batch + 1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\")\",\n \"not run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling\",\n \"and stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.score_batch(pipelines, X_holdout, y_holdout,\",\n \"primary objective {self.objective}.\\\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc =\",\n \"self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change =\",\n \"cross validation scores. {pipeline_name} may not perform as estimated on unseen data.\\\") return\",\n \"full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all():\",\n \"== 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if\",\n \"field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives\",\n \"estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\\\"No allowed pipelines to search\\\")\",\n \"= best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def\",\n \"Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str,\",\n \"have been made Returns: int: the number of pipeline evaluations made in the\",\n \"to the automl rankings with the requirement that automl search has been run.\",\n \"Change `binary` to `multiclass` or `regression` depending on the problem type. Note that\",\n \"not to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function\",\n \"self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not found in automl results\\\") pipeline_class\",\n \"value must be a positive integer. Received {} instead\\\".format(patience)) if tolerance and (tolerance\",\n \"time.time() - start_of_loop self._start += time_in_loop return False else: leading_char = \\\"\\\" def\",\n \"well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto',\",\n \"= False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings\",\n \"for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))):\",\n \"instance. \\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be run before selecting\",\n \"scoring will not be included in the dictionary but the exception and stacktrace\",\n \"return True # check max_time and max_iterations elapsed = time.time() - self._start if\",\n \"dict with results from each pipeline, and `search_order`: a list describing the order\",\n \"of unique pipelines to search over per batch is one, ensembling will not\",\n \"True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score']\",\n \"not None: logger.info(\\\"Will stop searching for new pipelines after %d seconds.\\\\n\\\" % self.max_time)\",\n \"None: logger.info(f\\\"Searching up to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \\\")\",\n \"iterations without improvement. Stopping search early...\\\".format(self.patience)) return False return True def _validate_problem_type(self): for\",\n \"score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores)\",\n \"the gap and max_delay variables. train_best_pipeline (boolean): Whether or not to train the\",\n \"`multiclass` or `regression` depending on the problem type. Note that if allowed_pipelines is\",\n \"AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary for \\\"Plotting\\\" documentation, since\",\n \"f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines:\",\n \"again.\\\") return # don't show iteration plot outside of a jupyter notebook if\",\n \"estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\\\"No\",\n \"self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from all\",\n \"(list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by\",\n \"one. The first batch will train a baseline pipline + one of each\",\n \"[] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try: self.plot\",\n \"evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family\",\n \"def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {}\",\n \"int) or patience < 0): raise ValueError(\\\"patience value must be a positive integer.\",\n \"\\\"\\\"\\\"Given the original stopping criterion and current state, should the search continue? Returns:\",\n \"out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except\",\n \"score of np.nan on the primary objective {self.objective}.\\\") self.search_duration = time.time() - self._start\",\n \"gap and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or {} def _handle_keyboard_interrupt(self):\",\n \"positional parameters: the Exception raised, the traceback, and the AutoMLSearch object. Must also\",\n \"f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\"\",\n \"in this loop does not count towards the time budget (if set) time_in_loop\",\n \"a float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time,\",\n \"import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import\",\n \"type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str,\",\n \"TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import\",\n \"random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not\",\n \"= (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) +\",\n \"{self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback:\",\n \"for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field\",\n \"that pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase:\",\n \"Includes information such as type of pipeline components, problem, training time, cross validation,\",\n \"self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration)\",\n \"with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds\",\n \"all objectives # but also fields like \\\"# Training\\\" and \\\"# Testing\\\", so\",\n \"self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time\",\n \"Seed for the random number generator. Defaults to 0. n_jobs (int or None):\",\n \"run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_iterations is too small,\",\n \"self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices\",\n \"will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will\",\n \"evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger,\",\n \"self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores =\",\n \"_num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline evaluations which have been made Returns: int:\",\n \"to True. start_iteration_callback (callable): Function called before each pipeline training iteration. Callback function\",\n \"all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] =\",\n \"return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters =\",\n \"input(leading_char + \\\"Do you really want to exit search (y/n)? \\\").strip().lower() if choice\",\n \"pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not found in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class')\",\n \"the results to the automl rankings with the requirement that automl search has\",\n \"split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import\",\n \"untrained pipeline instance. Returns: PipelineBase: A trained instance of the best pipeline and\",\n \"ValueError(\\\"patience value must be a positive integer. Received {} instead\\\".format(patience)) if tolerance and\",\n \"logger.info(\\\"Optimizing for %s. \\\" % self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else\",\n \"be None or non-negative. Received {max_time}.\\\") if max_batches is not None and max_batches\",\n \"pipeline. Includes information such as type of pipeline components, problem, training time, cross\",\n \"pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter:\",\n \"{}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to the data. This is\",\n \">= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search early...\\\".format(self.patience)) return False return True\",\n \") logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 #\",\n \"pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv,\",\n \"not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc\",\n \"or []: if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline {} is not compatible\",\n \"and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if\",\n \"abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False):\",\n \"set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params =\",\n \"\\\"\\\"\\\" if X_train is None: raise ValueError('Must specify training data as a 2d\",\n \"new pipelines after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\",\",\n \"be a dict containing values for at least the gap and max_delay \\\"\",\n \"ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine\",\n \"data. This can be helpful for training pipelines once the search is complete.\",\n \"pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary for binary classification\",\n \"used when ensembling is True. Must be between 0 and 1, exclusive. Defaults\",\n \"cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs a warning if variance is higher\",\n \"{ 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs =\",\n \"pipeline in the rankings If self._best_pipeline already exists, check to make sure it\",\n \"\\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline\",\n \"None. tolerance (float): Minimum percentage difference to qualify as score improvement for early\",\n \"not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\\\"Using default\",\n \"pipeline and parameters found during automl search. If `train_best_pipeline` is set to False,\",\n \"log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN\",\n \"np import pandas as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator\",\n \"to initalize a pipeline with. _ensembling_split_size (float): The amount of the training data\",\n \"log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\" % self.objective.name) logger.info(\\\"{} score is\",\n \"pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)):\",\n \"describing level of parallelism used for pipelines. None and 1 are equivalent. If\",\n \"ensembling (boolean): If True, runs ensembling in a separate batch after every allowed\",\n \"allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The\",\n \"logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search early...\\\".format(self.patience)) return False return True def _validate_problem_type(self):\",\n \"self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if objective == 'auto':\",\n \"rankings_desc = \\\"\\\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch\",\n \"except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = []\",\n \"pipeline evaluations which have been made Returns: int: the number of pipeline evaluations\",\n \"ValueError(\\\"tolerance value must be a float between 0.0 and 1.0 inclusive. Received {}\",\n \"(bool): If True, return dictionary of information about pipeline. Defaults to False. Returns:\",\n \"[])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback:\",\n \"0: raise ValueError(f\\\"Parameter max_batches must be None or non-negative. Received {max_batches}.\\\") if max_iterations\",\n \"= self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train =\",\n \"\\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from\",\n \"with scoring results from all pipelines searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better: ascending\",\n \"PipelineBase: untrained pipeline instance associated with the provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id)\",\n \"percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"]\",\n \"retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided ID \\\"\\\"\\\" pipeline_results\",\n \"= pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is None:\",\n \"batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def\",\n \"pipelines in the current AutoML batch produced a score of np.nan on the\",\n \"automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance\",\n \"the specified pipeline initialized with the parameters used to train that pipeline during\",\n \"y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for\",\n \"is the first pipeline fit during search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline\",\n \"data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout\",\n \"string or None. Received {type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time, (int, float)) and\",\n \"raise ValueError(f\\\"Parameter max_time must be None or non-negative. Received {max_time}.\\\") if max_batches is\",\n \"self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline\",\n \"bool: True if yes, False if no. \\\"\\\"\\\" if self._interrupted: return False #\",\n \"over. If the number of unique pipelines to search over per batch is\",\n \"than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold)\",\n \"the random number generator. Defaults to 0. n_jobs (int or None): Non-negative integer\",\n \"be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number\",\n \"So that the time in this loop does not count towards the time\",\n \"def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from the highest-scoring set of\",\n \"= f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type\",\n \"and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations,\",\n \"gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod\",\n \"= self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break\",\n \"None and max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations must be None or non-negative.\",\n \"= f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\"\",\n \"time_in_loop return False else: leading_char = \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best\",\n \"self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score pipelines.\\\") if self.max_batches is\",\n \"= patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order':\",\n \"object. error_callback (callable): Function called when `search()` errors and raises an Exception. Callback\",\n \"return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of a pipeline training result,\",\n \"`log_error_callback`. additional_objectives (list): Custom set of objectives to score on. Will override default\",\n \"is not None: logger.info(\\\"Will stop searching for new pipelines after %d seconds.\\\\n\\\" %\",\n \"self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and\",\n \"propose and rank pipelines, but not for optimizing each pipeline during fit-time. When\",\n \"self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines\",\n \"int: the number of pipeline evaluations made in the search \\\"\\\"\\\" return len(self._results['pipeline_results'])\",\n \"search again.\\\") return # don't show iteration plot outside of a jupyter notebook\",\n \"per batch is one, ensembling will not run. Defaults to False. max_batches (int):\",\n \"from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from\",\n \"raise ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and\",\n \"of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change\",\n \"significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score >\",\n \"float, str, type(None))): raise TypeError(f\\\"Parameter max_time must be a float, int, string or\",\n \"is not None: logger.info(\\\"Searching up to %s pipelines. \\\" % self.max_iterations) if self.max_time\",\n \"PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def\",\n \"pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots\",\n \"`train_best_pipeline` is set to False, returns an untrained pipeline instance. Returns: PipelineBase: A\",\n \"then the time will be in seconds. For strings, time can be specified\",\n \"(int): pipeline to describe return_dict (bool): If True, return dictionary of information about\",\n \"be None or non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else\",\n \"float)) and max_time < 0: raise ValueError(f\\\"Parameter max_time must be None or non-negative.\",\n \"every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\") if self.max_batches and self.max_iterations is None:\",\n \"is None: raise PipelineNotFoundError(\\\"Pipeline not found in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters\",\n \"== 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])):\",\n \"components, problem, training time, cross validation, etc. \\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']:\",\n \"(self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family ==\",\n \"is not None and max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations must be None\",\n \"0.0 and 1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience = patience self.tolerance = tolerance\",\n \"self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters\",\n \"max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a\",\n \"maps to a dictionary of scores. Note that the any pipelines that error\",\n \"return_dict: return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a given pipeline then\",\n \"1)) if run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split\",\n \"update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40\",\n \"set to True, but max_batches is too small, so ensembling will not run.\",\n \"[]: raise ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling\",\n \"= full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in\",\n \"[n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes): type of supervised\",\n \"float between 0.0 and 1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience = patience self.tolerance\",\n \"= pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property\",\n \"if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is not compatible with a {}\",\n \"tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or\",\n \"etc. \\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline =\",\n \"set, then max_iterations will default to max_iterations of 5. max_time (int, str): Maximum\",\n \"The pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function\",\n \"raise PipelineNotFoundError(\\\"Pipeline not found in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters')\",\n \"for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all\",\n \"full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from all pipelines searched\\\"\\\"\\\" ascending =\",\n \"helpful for training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List\",\n \"called after each pipeline training iteration. Callback function takes three positional parameters: A\",\n \"self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters =\",\n \"pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}})\",\n \"Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores\",\n \"pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return\",\n \"SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter\",\n \"mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of var\\\", c] = std / mean\",\n \"first batch will train a baseline pipline + one of each pipeline family\",\n \"with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) #\",\n \"of feature types, either numerical or categorical. Categorical features will automatically be encoded\",\n \"X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed)\",\n \"obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called\",\n \"n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time']\",\n \"must be None or non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str)\",\n \"return False else: leading_char = \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline\",\n \"found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input\",\n \"additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated\",\n \"if self._searched: logger.info(\\\"AutoMLSearch.search() has already been run and will not run again on\",\n \"== 0: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_batches is\",\n \"# they are not scores if field in objective_name_to_class: scores[field] += value return\",\n \"run ensembling.\\\") else: logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines)\",\n \"def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives:\",\n \"time, cross validation, etc. \\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not\",\n \"results from all pipelines searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better: ascending = False\",\n \"search\\\") logger.info(\\\"Optimizing for %s. \\\" % self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better\",\n \"self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification pipeline thresholds for:\",\n \"% pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]]\",\n \"If None, early stopping is disabled. Defaults to None. tolerance (float): Minimum percentage\",\n \"\\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be run before selecting `best_pipeline`.\\\")\",\n \"= self.X_train y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train\",\n \"pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training\",\n \"and the AutoMLSearch object. error_callback (callable): Function called when `search()` errors and raises\",\n \"set of objectives to score on. Will override default objectives for problem type\",\n \"max_time (int, str): Maximum time to search for pipelines. This will not start\",\n \"additional_objectives (list): Custom set of objectives to score on. Will override default objectives\",\n \"None. Received {type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time, (int, float)) and max_time <\",\n \"in seconds. For strings, time can be specified as seconds, minutes, or hours.\",\n \"{} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification\",\n \"import time from collections import defaultdict import cloudpickle import numpy as np import\",\n \"= None self._interrupted = False if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to search\",\n \"# The 'all_objective_scores' field contains scores for all objectives # but also fields\",\n \"None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt:\",\n \"(1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration:\",\n \"for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross\",\n \"\\\"\\\"\\\"Class that allows access to a copy of the results from `automl_search`. Returns:\",\n \"class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize\",\n \"(callable): Function called before each pipeline training iteration. Callback function takes three positional\",\n \"\\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice == \\\"n\\\": # So that the\",\n \"logger.warning(\\\"Ensembling is set to True, but the number of unique pipelines is one,\",\n \"function takes three positional parameters: A dictionary containing the training results for the\",\n \"of information about pipeline. Defaults to False. Returns: Description of specified pipeline. Includes\",\n \"elif choice == \\\"n\\\": # So that the time in this loop does\",\n \"self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train,\",\n \"== self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train\",\n \"Jupyter notebook. Disabled by default in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has\",\n \"{} instead\\\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results = {\",\n \"percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family\",\n \"baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap\",\n \"rankings If self._best_pipeline already exists, check to make sure it is different from\",\n \"from collections import defaultdict import cloudpickle import numpy as np import pandas as\",\n \"is None or parameters is None: raise PipelineNotFoundError(\\\"Pipeline class or parameters not found\",\n \"to save file pickle_protocol (int): the pickle data stream format. Returns: None \\\"\\\"\\\"\",\n \"- R2 for regression problems. max_iterations (int): Maximum number of iterations to search.\",\n \"if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\\\"Pipeline class or parameters\",\n \"a list describing the order the pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property\",\n \"= ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False)\",\n \"a baseline pipline + one of each pipeline family allowed in the search.\",\n \"problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue,\",\n \"to load Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb') as f: return cloudpickle.load(f)\",\n \"len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores'\",\n \"(n_cpus + 1 + n_jobs) are used. ensembling (boolean): If True, runs ensembling\",\n \"to search. The default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\")\",\n \"def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name}\",\n \"_ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split size must be between 0 and 1\",\n \"{ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline\",\n \"fit during search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type\",\n \"SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary pipeline threshold. Defaults\",\n \"Returns: bool: If True, search should terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop\",\n \"None indicates all pipelines for this problem type are allowed. Setting this field\",\n \"dictionary of information about pipeline. Defaults to False. Returns: Description of specified pipeline.\",\n \"Will override default objectives for problem type if not empty. random_seed (int): Seed\",\n \"elapsed = time.time() - self._start if self.max_time and elapsed >= self.max_time: return False\",\n \"self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False\",\n \"field in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for objective,\",\n \"separate batch after every allowed pipeline class has been iterated over. If the\",\n \"= list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline in\",\n \"ensembling will not run. Defaults to False. max_batches (int): The maximum number of\",\n \"self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set to True, but\",\n \"search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary for \\\"Plotting\\\" documentation, since Sphinx does not\",\n \"objectives # but also fields like \\\"# Training\\\" and \\\"# Testing\\\", so we\",\n \"= self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices]\",\n \"(str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full\",\n \"self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores)\",\n \"an untrained_pipeline containing the parameters used during training, and the AutoMLSearch object. error_callback\",\n \"with the parameters used to train that pipeline during automl search. Arguments: pipeline_id\",\n \"families to search. The default of None searches over all model families. Run\",\n \"the same instance. Re-initialize AutoMLSearch to search again.\\\") return # don't show iteration\",\n \"def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a given pipeline then adds the results\",\n \"run. Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will\",\n \"the search. problem_configuration (dict, None): Additional parameters needed to configure the search. For\",\n \"the Exception raised, the traceback, and the AutoMLSearch object. Must also accepts kwargs,\",\n \"{threshold}) within cross validation scores. {pipeline_name} may not perform as estimated on unseen\",\n \"mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format,\",\n \"return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc =\",\n \"max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches:\",\n \"like \\\"# Training\\\" and \\\"# Testing\\\", so we want to exclude them since\",\n \"desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm\",\n \"from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import (\",\n \"self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches\",\n \"run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set to True, but the number\",\n \"self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs\",\n \"best pipeline before returning it. Defaults to True. pipeline_parameters (dict): A dict of\",\n \"automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or\",\n \"- start_of_loop self._start += time_in_loop return False else: leading_char = \\\"\\\" def search(self,\",\n \"not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True)\",\n \"(int): the pickle data stream format. Returns: None \\\"\\\"\\\" with open(file_path, 'wb') as\",\n \"Received {} instead\\\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results =\",\n \"optimizing each pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary for\",\n \"before training and thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if\",\n \"but also fields like \\\"# Training\\\" and \\\"# Testing\\\", so we want to\",\n \"= None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except\",\n \"KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids\",\n \"Received {} instead\\\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance < 0.0):\",\n \"False if no. \\\"\\\"\\\" if self._interrupted: return False # for add_to_rankings if self._searched:\",\n \"pipelines to search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for\",\n \"== ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY:\",\n \"want to stop the search. Returns: bool: If True, search should terminate early\",\n \"Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self,\",\n \"keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from all pipelines\",\n \"_ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data\",\n \"Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner class to\",\n \"configure the search. For example, in time series problems, values should be passed\",\n \"{} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines\",\n \"= pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data,\",\n \"describing the order the pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self):\",\n \"the problem type. Note that if allowed_pipelines is provided, this parameter will be\",\n \"for a total of {self.max_iterations} pipelines. \\\") elif self.max_iterations is not None: logger.info(\\\"Searching\",\n \"can be specified as seconds, minutes, or hours. patience (int): Number of iterations\",\n \"if max_batches is not None and max_batches < 0: raise ValueError(f\\\"Parameter max_batches must\",\n \"stopping criterion and current state, should the search continue? Returns: bool: True if\",\n \"is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In\",\n \"else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if\",\n \"# but also fields like \\\"# Training\\\" and \\\"# Testing\\\", so we want\",\n \"automl search has been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate.\",\n \"problems, and - R2 for regression problems. max_iterations (int): Maximum number of iterations\",\n \"None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the\",\n \"else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None: logger.info(\\\"Generating\",\n \"[] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try:\",\n \"None: raise PipelineNotFoundError(\\\"Pipeline not found in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters =\",\n \"problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The\",\n \"are used. For n_jobs below -1, (n_cpus + 1 + n_jobs) are used.\",\n \"True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective\",\n \"cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores\",\n \"pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\\\"# Training\\\",\",\n \"self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try:\",\n \"_validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise\",\n \"@staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object at file path Arguments: file_path (str): location\",\n \"= ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices\",\n \"self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc,\",\n \"'\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc = (\",\n \"the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training iteration. Callback\",\n \"True): Shows an iteration vs. score plot in Jupyter notebook. Disabled by default\",\n \"import cloudpickle import numpy as np import pandas as pd import woodwork as\",\n \"self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN\",\n \"\\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed\",\n \"= IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def\",\n \"is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score pipelines.\\\")\",\n \"== parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that allows access to\",\n \"self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance\",\n \"from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from\",\n \"or non-negative. Received {max_time}.\\\") if max_batches is not None and max_batches < 0:\",\n \"max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time\",\n \"else: return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max\",\n \"y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self):\",\n \"batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective\",\n \"X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list of pipelines on the given holdout data.\",\n \"train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring.\",\n \"training iteration. Callback function takes three positional parameters: A dictionary containing the training\",\n \"of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt:\",\n \"(int): Seed for the random number generator. Defaults to 0. n_jobs (int or\",\n \"exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to train for\",\n \"test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback,\",\n \"import numpy as np import pandas as pd import woodwork as ww from\",\n \"Returns: Description of specified pipeline. Includes information such as type of pipeline components,\",\n \"as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import\",\n \"self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start,\",\n \"pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples,\",\n \"5. max_time (int, str): Maximum time to search for pipelines. This will not\",\n \"patience is not None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase\",\n \"up to {self.max_batches} batches for a total of {self.max_iterations} pipelines. \\\") elif self.max_iterations\",\n \"= self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched\",\n \"1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\") self.max_iterations =\",\n \"best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices],\",\n \"score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if score_improved\",\n \"threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv:\",\n \"from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types\",\n \"one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback\",\n \"gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay},\",\n \"scores if field in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds\",\n \"scores for all objectives # but also fields like \\\"# Training\\\" and \\\"#\",\n \"baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type ==\",\n \"is not None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results'])\",\n \"run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_batches is too small,\",\n \"pipeline training result, returns an untrained instance of the specified pipeline initialized with\",\n \"class or parameters not found in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self,\",\n \"to the user asking if they want to stop the search. Returns: bool:\",\n \"file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file path Arguments: file_path (str): location to\",\n \"seconds, minutes, or hours. patience (int): Number of iterations without improvement to stop\",\n \"ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class\",\n \"function takes three positional parameters: The pipeline class, the pipeline parameters, and the\",\n \"None: raise ValueError('Must specify training data target values as a 1d vector using\",\n \"for optimizing each pipeline during fit-time. When set to 'auto', chooses: - LogLossBinary\",\n \"not None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating\",\n \"class has been iterated over. If the number of unique pipelines to search\",\n \"values should be passed in for the gap and max_delay variables. train_best_pipeline (boolean):\",\n \"if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] =\",\n \"PipelineBase]: Dictionary keyed by pipeline name that maps to the fitted pipeline. Note\",\n \"(int): Number of iterations without improvement to stop search early. Must be positive.\",\n \"The amount of the training data we'll set aside for training ensemble metalearners.\",\n \"* self._pipelines_per_batch} iterations after that.\\\") if self.max_batches and self.max_iterations is None: self.show_batch_output =\",\n \"if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if\",\n \"regression problems. max_iterations (int): Maximum number of iterations to search. If max_iterations and\",\n \"AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective\",\n \"rank pipelines, but not for optimizing each pipeline during fit-time. When set to\",\n \"and max_delay variables. train_best_pipeline (boolean): Whether or not to train the best pipeline\",\n \"_get_funct_name(function): if callable(function): return function.__name__ else: return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\"\",\n \"search over per batch is one, ensembling will not run. Defaults to False.\",\n \"def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list of pipelines on the\",\n \"default objectives for problem type if not empty. random_seed (int): Seed for the\",\n \"must be None or non-negative. Received {max_batches}.\\\") if max_iterations is not None and\",\n \"numerical or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows\",\n \"to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the binary pipeline threshold.\",\n \"search _baseline_cv_scores will be empty so we will return # nan for the\",\n \"evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import\",\n \"problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification pipeline\",\n \"for binary classification problems, - LogLossMulticlass for multiclass classification problems, and - R2\",\n \"pipelines. \\\") elif self.max_iterations is not None: logger.info(\\\"Searching up to %s pipelines. \\\"\",\n \"time.time() - self._start if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations\",\n \"for all objectives # but also fields like \\\"# Training\\\" and \\\"# Testing\\\",\",\n \"after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\", \\\".join([model.value for\",\n \"if our main objective is part of default set of objectives for problem_type,\",\n \"return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True)\",\n \"we'll set aside for training ensemble metalearners. Only used when ensembling is True.\",\n \"already exists, check to make sure it is different from the current best\",\n \"list of feature types, either numerical or categorical. Categorical features will automatically be\",\n \"self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train\",\n \"= False if run_ensembling and self.max_iterations is not None: # Baseline + first\",\n \"{first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\") if self.max_batches and\",\n \"Additional parameters needed to configure the search. For example, in time series problems,\",\n \"raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or\",\n \"for supervised learning tasks. problem_type (str or ProblemTypes): type of supervised learning problem.\",\n \"not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE:\",\n \"f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to\",\n \"Arguments: file_path (str): location to save file pickle_protocol (int): the pickle data stream\",\n \"is set to True, but the number of unique pipelines is one, so\",\n \"= self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if\",\n \"again on the same instance. Re-initialize AutoMLSearch to search again.\\\") return # don't\",\n \"in the search. The default of None indicates all pipelines for this problem\",\n \"will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to\",\n \"None: logger.info(\\\"Will stop searching for new pipelines after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed\",\n \"self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that allows access to a copy of\",\n \"but max_batches is too small, so ensembling will not run. Set max_batches >=\",\n \"from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed,\",\n \"default of None indicates all pipelines for this problem type are allowed. Setting\",\n \"ObjectiveBase): The objective to optimize for. Used to propose and rank pipelines, but\",\n \"batch after every allowed pipeline class has been iterated over. If the number\",\n \"{} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type:\",\n \"plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched:\",\n \"pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self)\",\n \"len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set to True, but the number of unique\",\n \"+ self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\\\"Parameter max_time must\",\n \"ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if\",\n \"not found in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe\",\n \"raise ValueError('Must specify training data as a 2d array using the X_train argument')\",\n \"loop does not count towards the time budget (if set) time_in_loop = time.time()\",\n \"show iteration plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython except\",\n \"train and score pipelines.\\\") if self.max_batches is not None: logger.info(f\\\"Searching up to {self.max_batches}\",\n \"hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline\",\n \"obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is not compatible\",\n \"on. Will override default objectives for problem type if not empty. random_seed (int):\",\n \"model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score\",\n \"[get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives\",\n \"self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores\",\n \"if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\\\"Parameter max_time must be a\",\n \"self.max_iterations is not None: logger.info(\\\"Searching up to %s pipelines. \\\" % self.max_iterations) if\",\n \"batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 -\",\n \"random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback)\",\n \"Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom set of objectives\",\n \"run_ensembling = False if run_ensembling and self.max_iterations is not None: # Baseline +\",\n \"cloudpickle import numpy as np import pandas as pd import woodwork as ww\",\n \"= len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The\",\n \"given pipeline then adds the results to the automl rankings with the requirement\",\n \"**self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs,\",\n \"AutoML object at file path Arguments: file_path (str): location to find file to\",\n \"Dictionary keyed by pipeline name that maps to a dictionary of scores. Note\",\n \"field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families\",\n \"search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max\",\n \"over per batch is one, ensembling will not run. Defaults to False. max_batches\",\n \"pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1:\",\n \"list describing the order the pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def\",\n \"log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective\",\n \"< _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split size must be between 0 and\",\n \"Stopping search early...\\\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in self.additional_objectives:\",\n \"self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after\",\n \"StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted\",\n \"pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called\",\n \"logs a warning if variance is higher than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name\",\n \"= mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event\",\n \"multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback\",\n \"self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list of pipelines on\",\n \"logger.warning(\\\"Unable to import plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families =\",\n \"time will be in seconds. For strings, time can be specified as seconds,\",\n \"time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN =\",\n \"\\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already been run and will not run again\",\n \"number of pipelines to train for every batch after the first one. The\",\n \"= start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter\",\n \"data of length [n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes):\",\n \"ValueError(f\\\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if\",\n \"and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\") if self.max_batches and self.max_iterations is\",\n \"= all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c] = std\",\n \"and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training\",\n \"%s pipelines. \\\" % self.max_iterations) if self.max_time is not None: logger.info(\\\"Will stop searching\",\n \"pipeline to describe return_dict (bool): If True, return dictionary of information about pipeline.\",\n \"return False # for add_to_rankings if self._searched: return True # Run at least\",\n \"% self.max_iterations) if self.max_time is not None: logger.info(\\\"Will stop searching for new pipelines\",\n \"'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification\",\n \"override default objectives for problem type if not empty. random_seed (int): Seed for\",\n \"None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for\",\n \"train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train)\",\n \"self.X_train y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train =\",\n \"max_iterations (int): Maximum number of iterations to search. If max_iterations and max_time is\",\n \"= False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration:\",\n \"one, ensembling will not run. Defaults to False. max_batches (int): The maximum number\",\n \"and max_iterations have precedence over stopping the search. problem_configuration (dict, None): Additional parameters\",\n \"{problem_configuration}.\\\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the user\",\n \"containing values for at least the gap and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\")\",\n \"= best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline evaluations which have been\",\n \"set to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration):\",\n \"provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not\",\n \"pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before each pipeline training\",\n \"will not start a new pipeline search after the duration has elapsed. If\",\n \"score on. Will override default objectives for problem type if not empty. random_seed\",\n \"self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators\",\n \"for. Used to propose and rank pipelines, but not for optimizing each pipeline\",\n \"is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives\",\n \"\\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape\",\n \"and max_time < 0: raise ValueError(f\\\"Parameter max_time must be None or non-negative. Received\",\n \"= (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations <\",\n \"+ (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size\",\n \"{self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc\",\n \"not problem_configuration or not all(p in problem_configuration for p in required_parameters): raise ValueError(\\\"user_parameters\",\n \"for add_to_rankings if self._searched: return True # Run at least one pipeline for\",\n \"time from collections import defaultdict import cloudpickle import numpy as np import pandas\",\n \"error out during training will not be included in the dictionary but the\",\n \"from all pipelines searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols\",\n \"coefficient of variation (cv >= {threshold}) within cross validation scores. {pipeline_name} may not\",\n \"def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the user asking if they want to\",\n \"None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches =\",\n \"self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines with IDs:\",\n \"in the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original stopping criterion\",\n \"takes three positional parameters: the Exception raised, the traceback, and the AutoMLSearch object.\",\n \"def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not\",\n \"in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective)\",\n \"max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\\\")\",\n \"stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines,\",\n \"f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\"\",\n \"num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches\",\n \"patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [],\",\n \"an untrained pipeline instance. \\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be\",\n \"self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable\",\n \"Arguments: feature_types (list, optional): list of feature types, either numerical or categorical. Categorical\",\n \"obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is not compatible with a {} problem.\\\".format(obj.name, self.problem_type.value))\",\n \"pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if\",\n \"made Returns: int: the number of pipeline evaluations made in the search \\\"\\\"\\\"\",\n \"of pipelines to train for every batch after the first one. The first\",\n \"with the requirement that automl search has been run. Arguments: pipeline (PipelineBase): pipeline\",\n \"the training data. This can be helpful for training pipelines once the search\",\n \"rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a trained\",\n \"default to max_iterations of 5. max_time (int, str): Maximum time to search for\",\n \"is not set, then max_iterations will default to max_iterations of 5. max_time (int,\",\n \"False while self._should_continue(): try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML\",\n \"batch produced a score of np.nan on the primary objective {self.objective}.\\\") self.search_duration =\",\n \"to True, but the number of unique pipelines is one, so ensembling will\",\n \"if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline\",\n \"objective {} is not compatible with a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in\",\n \"None: raise PipelineNotFoundError(\\\"Pipeline class or parameters not found in automl results\\\") return pipeline_class(parameters,\",\n \"PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines =\",\n \"( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from\",\n \"\\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters ==\",\n \"- 1)) if run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling\",\n \"self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\\\"Parameter max_time must be\",\n \"pipeline fit during search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif\",\n \"len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\":\",\n \"Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary for \\\"Plotting\\\" documentation, since Sphinx does\",\n \"len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches ==\",\n \"None self._interrupted = False if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to search over...\\\")\",\n \"False return True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise\",\n \"to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling\",\n \"= 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number =\",\n \"= self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise\",\n \"train and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']:\",\n \"and stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self,\",\n \"not None: logger.info(\\\"Searching up to %s pipelines. \\\" % self.max_iterations) if self.max_time is\",\n \"in a separate batch after every allowed pipeline class has been iterated over.\",\n \"random number generator. Defaults to 0. n_jobs (int or None): Non-negative integer describing\",\n \"See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective to optimize\",\n \"be a positive integer. Received {} instead\\\".format(patience)) if tolerance and (tolerance > 1.0\",\n \"1.0 or tolerance < 0.0): raise ValueError(\\\"tolerance value must be a float between\",\n \"not to train the best pipeline before returning it. Defaults to True. pipeline_parameters\",\n \"= full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the current AutoML\",\n \"in all_objective_scores: if c in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue\",\n \"of max_batches=1.\\\\n\\\") if patience and (not isinstance(patience, int) or patience < 0): raise\",\n \"non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already been run and will not\",\n \"show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing\",\n \"self._searched: return True # Run at least one pipeline for every search num_pipelines\",\n \"or parameters not found in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id,\",\n \"the current best pipeline before training and thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return\",\n \"search for pipelines. This will not start a new pipeline search after the\",\n \"pipeline class has been iterated over. If the number of unique pipelines to\",\n \"a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type !=\",\n \"the search continue? Returns: bool: True if yes, False if no. \\\"\\\"\\\" if\",\n \"Function called before each pipeline training iteration. Callback function takes three positional parameters:\",\n \"Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if not\",\n \"results from each pipeline, and `search_order`: a list describing the order the pipelines\",\n \"jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \\\"Beginning\",\n \"automl search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance.\",\n \"desc = f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] +\",\n \"is higher than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean())\",\n \"- LogLossMulticlass for multiclass classification problems, and - R2 for regression problems. max_iterations\",\n \"call `log_error_callback`. additional_objectives (list): Custom set of objectives to score on. Will override\",\n \"to train that pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve\",\n \"size must be between 0 and 1 exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0]))\",\n \"specified pipeline. Includes information such as type of pipeline components, problem, training time,\",\n \"problem type. Note that if allowed_pipelines is provided, this parameter will be ignored.\",\n \"= evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE\",\n \"== pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline()\",\n \"will not be included in the dictionary but the exception and stacktrace will\",\n \"= infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True,\",\n \"ensembling.\\\") else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines)\",\n \"Defaults to 0. n_jobs (int or None): Non-negative integer describing level of parallelism\",\n \"= SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p\",\n \"self._best_pipeline already exists, check to make sure it is different from the current\",\n \"get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline\",\n \"objectives for problem type if not empty. random_seed (int): Seed for the random\",\n \"first batch + each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines)\",\n \"- num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch *\",\n \"in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline\",\n \"for pipelines. None and 1 are equivalent. If set to -1, all CPUs\",\n \"evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline\",\n \"= (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False\",\n \"is set to True, but max_batches is too small, so ensembling will not\",\n \"f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'],\",\n \"problem_configuration or not all(p in problem_configuration for p in required_parameters): raise ValueError(\\\"user_parameters must\",\n \"parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that allows access to a\",\n \"pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start\",\n \"first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_iterations is too\",\n \"None: logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to\",\n \"len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN)\",\n \"= curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience:\",\n \"binary classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV): %.1f seconds\\\"\",\n \"= all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of var\\\",\",\n \"add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a given pipeline then adds the results to\",\n \"+= 1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement. Stopping search early...\\\".format(self.patience))\",\n \"= True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted =\",\n \"TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline =\",\n \"def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of pipelines on the training data. This\",\n \"_, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist()\",\n \"n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search\",\n \"raise PipelineNotFoundError(\\\"automl search must be run before selecting `best_pipeline`.\\\") return self._best_pipeline def save(self,\",\n \"= \\\"\\\\n\\\" start_of_loop = time.time() while True: choice = input(leading_char + \\\"Do you\",\n \"pipelines that error out during scoring will not be included in the dictionary\",\n \"If `train_best_pipeline` is set to False, returns an untrained pipeline instance. \\\"\\\"\\\" if\",\n \"- LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification problems, and\",\n \"model families to search. The default of None searches over all model families.\",\n \"= desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name\",\n \"pipeline {} is not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a\",\n \"best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of the best pipeline and parameters found during\",\n \"not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\")\",\n \"True, return dictionary of information about pipeline. Defaults to False. Returns: Description of\",\n \"of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase):\",\n \"evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary` to `multiclass` or `regression` depending on the\",\n \"ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all\",\n \"= {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline\",\n \"search after the duration has elapsed. If it is an integer, then the\",\n \"The objective to optimize for. Used to propose and rank pipelines, but not\",\n \"None else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None:\",\n \"-1, all CPUs are used. For n_jobs below -1, (n_cpus + 1 +\",\n \"is called before search _baseline_cv_scores will be empty so we will return #\",\n \"# don't show iteration plot outside of a jupyter notebook if show_iteration_plot: try:\",\n \"high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of a pipeline training result, returns\",\n \"default set of objectives for problem_type, remove it existing_main_objective = next((obj for obj\",\n \"Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines\",\n \"during training, and the AutoMLSearch object. error_callback (callable): Function called when `search()` errors\",\n \"( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline,\",\n \"Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features.\",\n \"ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline:\",\n \"no. \\\"\\\"\\\" if self._interrupted: return False # for add_to_rankings if self._searched: return True\",\n \"in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) >\",\n \"If the number of unique pipelines to search over per batch is one,\",\n \"problem, training time, cross validation, etc. \\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']: raise\",\n \"self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True\",\n \"None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}:\",\n \"allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method\",\n \">= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False #\",\n \"to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines allowed\",\n \"the first one. The first batch will train a baseline pipline + one\",\n \"iteration plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError:\",\n \"self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train\",\n \"three positional parameters: A dictionary containing the training results for the new pipeline,\",\n \"num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling =\",\n \"max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline):\",\n \"%.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold\",\n \"evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline\",\n \"len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling\",\n \"value return {objective: float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self,\",\n \"complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary\",\n \"pipeline training iteration. Callback function takes three positional parameters: The pipeline class, the\",\n \"None \\\"\\\"\\\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path):\",\n \"(PipelineBase): pipeline to train and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for\",\n \"num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations\",\n \"number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self)\",\n \"Set max_iterations >= {first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run at the\",\n \"in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time()\",\n \"True. start_iteration_callback (callable): Function called before each pipeline training iteration. Callback function takes\",\n \"import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline )\",\n \"high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of variation\",\n \"self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train =\",\n \"f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of pipelines on the\",\n \"iterated over. If the number of unique pipelines to search over per batch\",\n \"been iterated over. If the number of unique pipelines to search over per\",\n \"allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if\",\n \"stop searching for new pipelines after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families:\",\n \"evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective names allowed in automl.\\\") return objective() return\",\n \"of the training data we'll set aside for training ensemble metalearners. Only used\",\n \"batch after the first one. The first batch will train a baseline pipline\",\n \"objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is\",\n \"by pipeline name that maps to the fitted pipeline. Note that the any\",\n \"parameters found during automl search. If `train_best_pipeline` is set to False, returns an\",\n \"number of pipeline evaluations made in the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self):\",\n \"self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline evaluations which have\",\n \"training data we'll set aside for training ensemble metalearners. Only used when ensembling\",\n \"in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores\",\n \"{[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for\",\n \"(1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches)\",\n \"import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners\",\n \"pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc\",\n \"pipelines, but not for optimizing each pipeline during fit-time. When set to 'auto',\",\n \"Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to train for every\",\n \"= _pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches =\",\n \"return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id\",\n \"self._start if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines\",\n \"subclasses indicating the pipelines allowed in the search. The default of None indicates\",\n \"not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid data splitter\\\") if\",\n \"The target training data of length [n_samples]. Required for supervised learning tasks. problem_type\",\n \"raise ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get\",\n \"num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score\",\n \"cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for\",\n \"raise PipelineNotFoundError(\\\"Pipeline class or parameters not found in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed)\",\n \"desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1\",\n \"the automl rankings with the requirement that automl search has been run. Arguments:\",\n \"add_result_callback (callable): Function called after each pipeline training iteration. Callback function takes three\",\n \"A dict of the parameters used to initalize a pipeline with. _ensembling_split_size (float):\",\n \"_find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in the rankings If self._best_pipeline already exists, check\",\n \"for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary\",\n \"+ first batch + each pipeline iteration + 1 first_ensembling_iteration = (1 +\",\n \"(self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\\\"Ensembling\",\n \"pipeline, and `search_order`: a list describing the order the pipelines were searched. \\\"\\\"\\\"\",\n \"an integer, then the time will be in seconds. For strings, time can\",\n \"to None. tolerance (float): Minimum percentage difference to qualify as score improvement for\",\n \"+ len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False\",\n \"o in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return\",\n \"be empty so we will return # nan for the base score. percent_better\",\n \"object. add_result_callback (callable): Function called after each pipeline training iteration. Callback function takes\",\n \"False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\" % self.objective.name) logger.info(\\\"{} score\",\n \"to train and score pipelines.\\\") if self.max_batches is not None: logger.info(f\\\"Searching up to\",\n \"if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc =\",\n \"will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after\",\n \"instead\\\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance < 0.0): raise ValueError(\\\"tolerance\",\n \"to True. pipeline_parameters (dict): A dict of the parameters used to initalize a\",\n \"data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None,\",\n \"existing_main_objective = next((obj for obj in additional_objectives if obj.name == self.objective.name), None) if\",\n \"work well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None,\",\n \"all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef\",\n \"each pipeline training iteration. Callback function takes three positional parameters: A dictionary containing\",\n \"if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline,\",\n \"f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\"\",\n \"+ one of each pipeline family allowed in the search. \\\"\\\"\\\" if X_train\",\n \"ValueError('Must specify training data target values as a 1d vector using the y_train\",\n \"desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0]\",\n \"training will not be included in the dictionary but the exception and stacktrace\",\n \"return dictionary of information about pipeline. Defaults to False. Returns: Description of specified\",\n \"= get_core_objectives(self.problem_type) # if our main objective is part of default set of\",\n \"for c in all_objective_scores: if c in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] =\",\n \"the primary objective {self.objective}.\\\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc\",\n \"error out during scoring will not be included in the dictionary but the\",\n \"= True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\",\",\n \"not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to\",\n \"1 exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape,\",\n \"number of pipeline evaluations which have been made Returns: int: the number of\",\n \"least the gap and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or {}\",\n \"1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if patience and (not isinstance(patience, int) or\",\n \"= std all_objective_scores.loc[\\\"coef of var\\\", c] = std / mean if abs(mean) >\",\n \"Required for supervised learning tasks. problem_type (str or ProblemTypes): type of supervised learning\",\n \"cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search.\",\n \"get_core_objectives(self.problem_type) # if our main objective is part of default set of objectives\",\n \"self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file path Arguments: file_path\",\n \"the parameters used to train that pipeline during automl search. Arguments: pipeline_id (int):\",\n \"to find file to load Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb') as\",\n \"[]: if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline {} is not compatible with\",\n \"baseline pipeline to the data. This is the first pipeline fit during search.\",\n \"to optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called\",\n \"has been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \\\"\\\"\\\" pipeline_rows\",\n \"Note that the any pipelines that error out during scoring will not be\",\n \"allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines\",\n \"(list(class)): A list of PipelineBase subclasses indicating the pipelines allowed in the search.\",\n \"obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives =\",\n \"score pipelines.\\\") if self.max_batches is not None: logger.info(f\\\"Searching up to {self.max_batches} batches for\",\n \"training data target values as a 1d vector using the y_train argument') try:\",\n \"False, returns an untrained pipeline instance. Returns: PipelineBase: A trained instance of the\",\n \"show_iteration_plot = False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\" % self.objective.name)\",\n \"= [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline:\",\n \"return problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the user asking\",\n \"{_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc\",\n \"lines = sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function):\",\n \"if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective, return_instance=False) self.objective =\",\n \"0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines\",\n \"input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try:\",\n \"parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training\",\n \"feature_types (list, optional): list of feature types, either numerical or categorical. Categorical features\",\n \"For strings, time can be specified as seconds, minutes, or hours. patience (int):\",\n \"location to find file to load Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb')\",\n \"Used to propose and rank pipelines, but not for optimizing each pipeline during\",\n \"else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score pipelines.\\\") if self.max_batches is not\",\n \"self._searched: logger.info(\\\"AutoMLSearch.search() has already been run and will not run again on the\",\n \"None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o\",\n \"if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in\",\n \"\\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters\",\n \"in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline {} is\",\n \"accepts kwargs, so AutoMLSearch is able to pass along other appropriate parameters by\",\n \"for \\\"Plotting\\\" documentation, since Sphinx does not work well with instance attributes. plot\",\n \"def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in the rankings If self._best_pipeline already exists,\",\n \"'{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and\",\n \"small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1} to\",\n \"best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id'])\",\n \"instead\\\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results':\",\n \"None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary`\",\n \"(dict): A dict of the parameters used to initalize a pipeline with. _ensembling_split_size\",\n \"_handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the user asking if they want to stop\",\n \"def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file path Arguments: file_path (str):\",\n \"not found in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class\",\n \"< 0: raise ValueError(f\\\"Parameter max_time must be None or non-negative. Received {max_time}.\\\") if\",\n \"y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds,\",\n \"will not run.\\\") run_ensembling = False if run_ensembling and self.max_iterations is not None:\",\n \"try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id],\",\n \"that maps to a dictionary of scores. Note that the any pipelines that\",\n \"from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__)\",\n \"evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from\",\n \"pipeline training iteration. Callback function takes three positional parameters: A dictionary containing the\",\n \"all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\\\"# Training\\\", \\\"#\",\n \"try: if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of\",\n \"n_folds = len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items(): #\",\n \"Maximum number of iterations to search. If max_iterations and max_time is not set,\",\n \"takes three positional parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch\",\n \"self.objective = self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise\",\n \"additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class =\",\n \"run every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches\",\n \"mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c] =\",\n \"self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter\",\n \"= False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not\",\n \"allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False\",\n \"along other appropriate parameters by default. Defaults to None, which will call `log_error_callback`.\",\n \"self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to\",\n \"additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part of default set\",\n \"[self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\\\"Parameter max_time\",\n \"0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{} iterations without improvement.\",\n \"parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable):\",\n \"ensembling.\\\") else: logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) *\",\n \"full list. objective (str, ObjectiveBase): The objective to optimize for. Used to propose\",\n \"{best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best\",\n \"( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated\",\n \"curr_score < best_score if score_improved and significant_change: best_score = curr_score num_without_improvement = 0\",\n \"pipelines): \\\"\\\"\\\"Train a list of pipelines on the training data. This can be\",\n \"def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe\",\n \"scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary\",\n \"by pipeline name that maps to a dictionary of scores. Note that the\",\n \"\\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for the data set. Arguments:\",\n \"been made Returns: int: the number of pipeline evaluations made in the search\",\n \"too small, so ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1}\",\n \"= desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if\",\n \"rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if\",\n \"or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc =\",\n \"self._num_pipelines() if num_pipelines == 0: return True # check max_time and max_iterations elapsed\",\n \"to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params\",\n \"R2 for regression problems. max_iterations (int): Maximum number of iterations to search. If\",\n \"current state, should the search continue? Returns: bool: True if yes, False if\",\n \"The input training data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The\",\n \"data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of a pipeline training\",\n \"return function.__name__ else: return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective:\",\n \"0: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_batches is too\",\n \"optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train\",\n \"`train_best_pipeline` is set to False, returns an untrained pipeline instance. \\\"\\\"\\\" if not\",\n \"for p in required_parameters): raise ValueError(\\\"user_parameters must be a dict containing values for\",\n \"infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger =\",\n \"of default set of objectives for problem_type, remove it existing_main_objective = next((obj for\",\n \"if run_ensembling: if not (0 < _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split size\",\n \"best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >=\",\n \"type. Note that if allowed_pipelines is provided, this parameter will be ignored. data_splitter\",\n \"objective() return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o in\",\n \"ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use. Defaults to StratifiedKFold. tuner_class: The\",\n \"scoring results from all pipelines searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better: ascending =\",\n \"not all(p in problem_configuration for p in required_parameters): raise ValueError(\\\"user_parameters must be a\",\n \"num_pipelines = self._num_pipelines() if num_pipelines == 0: return True # check max_time and\",\n \"(boolean): If True, runs ensembling in a separate batch after every allowed pipeline\",\n \"logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if\",\n \"been run and will not run again on the same instance. Re-initialize AutoMLSearch\",\n \"search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS:\",\n \"highest-scoring set of parameters used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def\",\n \"check to make sure it is different from the current best pipeline before\",\n \"load Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def\",\n \"argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass,\",\n \"required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration for\",\n \"= sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return\",\n \"self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are\",\n \"f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data\",\n \"classification pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV): %.1f seconds\\\" %\",\n \"argument') if y_train is None: raise ValueError('Must specify training data target values as\",\n \"score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data']\",\n \"training, and the AutoMLSearch object. error_callback (callable): Function called when `search()` errors and\",\n \"in for the gap and max_delay variables. train_best_pipeline (boolean): Whether or not to\",\n \"A trained instance of the best pipeline and parameters found during automl search.\",\n \"is None: logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set\",\n \"0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline\",\n \"self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given\",\n \"untrained instance of the specified pipeline initialized with the parameters used to train\",\n \"leading_char = \\\"\\\\n\\\" start_of_loop = time.time() while True: choice = input(leading_char + \\\"Do\",\n \"the parameters used during training, and the AutoMLSearch object. error_callback (callable): Function called\",\n \"if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the current AutoML batch produced\",\n \"of specified pipeline. Includes information such as type of pipeline components, problem, training\",\n \"objective {} is not compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is\",\n \"n_jobs below -1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean): If\",\n \"all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of var\\\", c]\",\n \"to %s pipelines. \\\" % self.max_iterations) if self.max_time is not None: logger.info(\\\"Will stop\",\n \"\\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt\",\n \"objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field,\",\n \"the exception and stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.score_batch(pipelines,\",\n \"of var\\\", c] = std / mean if abs(mean) > 0 else np.inf\",\n \"stop the search. Returns: bool: If True, search should terminate early \\\"\\\"\\\" leading_char\",\n \"import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline,\",\n \"< 0: raise ValueError(f\\\"Parameter max_batches must be None or non-negative. Received {max_batches}.\\\") if\",\n \"X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline =\",\n \"False # check for early stopping if self.patience is None or self.tolerance is\",\n \"best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train\",\n \"from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner\",\n \"if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline =\",\n \"tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback:\",\n \"untrained_pipeline containing the parameters used during training, and the AutoMLSearch object. error_callback (callable):\",\n \"notebook. Disabled by default in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already\",\n \"This is the first pipeline fit during search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY:\",\n \"curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score) > self.tolerance score_improved\",\n \"best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return\",\n \"desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0:\",\n \"AutoMLSearchException(f\\\"All pipelines in the current AutoML batch produced a score of np.nan on\",\n \"pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results =\",\n \"Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name\",\n \"training time, cross validation, etc. \\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline\",\n \"original stopping criterion and current state, should the search continue? Returns: bool: True\",\n \"ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The default of None\",\n \"= evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {}\",\n \"search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start =\",\n \"{max_batches}.\\\") if max_iterations is not None and max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations\",\n \"a dict with results from each pipeline, and `search_order`: a list describing the\",\n \"pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\" % self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if\",\n \"(ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring.\",\n \"run_ensembling and self.max_iterations is not None: # Baseline + first batch + each\",\n \"Whether or not to train the best pipeline before returning it. Defaults to\",\n \"def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores']\",\n \"% self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to\",\n \"self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a\",\n \"y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None,\",\n \"False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched =\",\n \"1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling\",\n \"pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout\",\n \"pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class)\",\n \"between 0 and 1 exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _,\",\n \"is one, so ensembling will not run.\\\") run_ensembling = False if run_ensembling and\",\n \"pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps\",\n \"all_objective_scores.loc[\\\"coef of var\\\", c] = std / mean if abs(mean) > 0 else\",\n \"= additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives} if\",\n \"evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives\",\n \"problem_type (str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a\",\n \"bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv >=\",\n \"depending on the problem type. Note that if allowed_pipelines is provided, this parameter\",\n \"limit of max_batches=1.\\\\n\\\") if patience and (not isinstance(patience, int) or patience < 0):\",\n \"self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings\",\n \"{'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration for p in\",\n \"after each pipeline training iteration. Callback function takes three positional parameters: A dictionary\",\n \"pipeline_id): \\\"\\\"\\\"Given the ID of a pipeline training result, returns an untrained instance\",\n \"1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not\",\n \"search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap',\",\n \"then adds the results to the automl rankings with the requirement that automl\",\n \"at file path Arguments: file_path (str): location to save file pickle_protocol (int): the\",\n \"dictionary of scores. Note that the any pipelines that error out during scoring\",\n \"(int): The maximum number of batches of pipelines to search. Parameters max_time, and\",\n \"to propose and rank pipelines, but not for optimizing each pipeline during fit-time.\",\n \"instance. Re-initialize AutoMLSearch to search again.\\\") return # don't show iteration plot outside\",\n \"self.objective.greater_is_better: ascending = False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"]\",\n \"log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\"\",\n \"ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not found\",\n \"# if our main objective is part of default set of objectives for\",\n \"is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part of\",\n \"of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout\",\n \"Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\"\",\n \"return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from the\",\n \"True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False\",\n \"self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for\",\n \"is not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline\",\n \"non_core_objectives = get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is\",\n \"pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline,\",\n \"run. Defaults to False. max_batches (int): The maximum number of batches of pipelines\",\n \"default limit of max_batches=1.\\\\n\\\") if patience and (not isinstance(patience, int) or patience <\",\n \"num_pipelines >= self.max_iterations: return False # check for early stopping if self.patience is\",\n \"warning if variance is higher than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv =\",\n \"given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable,\",\n \"instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None,\",\n \"+ self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations =\",\n \"= time.time() - self._start if self.max_time and elapsed >= self.max_time: return False elif\",\n \"infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration,\",\n \"of the specified pipeline initialized with the parameters used to train that pipeline\",\n \"train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of pipelines on the training data. This can\",\n \"optimize the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before\",\n \"and 1 are equivalent. If set to -1, all CPUs are used. For\",\n \"all CPUs are used. For n_jobs below -1, (n_cpus + 1 + n_jobs)\",\n \"problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives if obj.name ==\",\n \"= X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline\",\n \"pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name],\",\n \"types, either numerical or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean,\",\n \"threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs a warning if variance is higher than\",\n \"rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a\",\n \"logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch\",\n \"MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap']\",\n \"rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of the best pipeline and\",\n \"during scoring will not be included in the dictionary but the exception and\",\n \"continue? Returns: bool: True if yes, False if no. \\\"\\\"\\\" if self._interrupted: return\",\n \"desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline\",\n \"type): if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML! \\\"\",\n \"time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = []\",\n \"they want to stop the search. Returns: bool: If True, search should terminate\",\n \"called before each pipeline training iteration. Callback function takes three positional parameters: The\",\n \"to search. Parameters max_time, and max_iterations have precedence over stopping the search. problem_configuration\",\n \"log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores =\",\n \"on the problem type. Note that if allowed_pipelines is provided, this parameter will\",\n \"== \\\"n\\\": # So that the time in this loop does not count\",\n \"else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay\",\n \"pipeline, an untrained_pipeline containing the parameters used during training, and the AutoMLSearch object.\",\n \"during automl search. If `train_best_pipeline` is set to False, returns an untrained pipeline\",\n \"\\\"to get all objective names allowed in automl.\\\") return objective() return objective def\",\n \"self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train,\",\n \"self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data\",\n \"f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if not self.rankings.empty: rankings_str\",\n \"returns an untrained pipeline instance. \\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must\",\n \"parameters: A dictionary containing the training results for the new pipeline, an untrained_pipeline\",\n \"f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\"\",\n \"split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices,\",\n \"Arguments: file_path (str): location to find file to load Returns: AutoSearchBase object \\\"\\\"\\\"\",\n \"the training data we'll set aside for training ensemble metalearners. Only used when\",\n \"self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train\",\n \"is not None and max_batches < 0: raise ValueError(f\\\"Parameter max_batches must be None\",\n \"as a 1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except\",\n \"get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of a pipeline training result, returns an untrained\",\n \"log. \\\"\\\"\\\" return self._engine.train_batch(pipelines) def score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list\",\n \"else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if\",\n \"provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to use.\",\n \"initialized with the parameters used to train that pipeline during automl search. Arguments:\",\n \"if self._searched: return True # Run at least one pipeline for every search\",\n \"__init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None,\",\n \"objectives for problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives if\",\n \"None: # Baseline + first batch + each pipeline iteration + 1 first_ensembling_iteration\",\n \"takes three positional parameters: A dictionary containing the training results for the new\",\n \"elif self.max_iterations is not None: logger.info(\\\"Searching up to %s pipelines. \\\" % self.max_iterations)\",\n \"self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is not compatible with a\",\n \"scores. Note that the any pipelines that error out during scoring will not\",\n \"pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the\",\n \"if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train\",\n \"for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def\",\n \"iterations to search. If max_iterations and max_time is not set, then max_iterations will\",\n \"run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run at\",\n \"type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data,\",\n \"pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with\",\n \"the any pipelines that error out during training will not be included in\",\n \"search has been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \\\"\\\"\\\"\",\n \"\\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline =\",\n \"number of unique pipelines is one, so ensembling will not run.\\\") run_ensembling =\",\n \"is different from the current best pipeline before training and thresholding\\\"\\\"\\\" if len(self.rankings)\",\n \"be ignored. allowed_model_families (list(str, ModelFamily)): The model families to search. The default of\",\n \"at least the gap and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or\",\n \"= None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices =\",\n \"if num_pipelines == 0: return True # check max_time and max_iterations elapsed =\",\n \"1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise\",\n \"self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter\",\n \">= {first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration} iteration\",\n \"not (0 < _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split size must be between\",\n \"f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result\",\n \"score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1\",\n \"The 'all_objective_scores' field contains scores for all objectives # but also fields like\",\n \"cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline =\",\n \"set to True, but the number of unique pipelines is one, so ensembling\",\n \">= {ensemble_nth_batch + 1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch}\",\n \"Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0)\",\n \"default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot =\",\n \"train_indices = self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline,\",\n \"so we want to exclude them since # they are not scores if\",\n \"generator. Defaults to 0. n_jobs (int or None): Non-negative integer describing level of\",\n \"the first pipeline fit during search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline =\",\n \"as estimated on unseen data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID\",\n \"of unique pipelines is one, so ensembling will not run.\\\") run_ensembling = False\",\n \"self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params )\",\n \"= random_seed self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except\",\n \"is part of default set of objectives for problem_type, remove it existing_main_objective =\",\n \"all(p in problem_configuration for p in required_parameters): raise ValueError(\\\"user_parameters must be a dict\",\n \"pipeline to train and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter\",\n \"integer describing level of parallelism used for pipelines. None and 1 are equivalent.\",\n \"import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter\",\n \"should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to\",\n \"best_score) / best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else\",\n \"problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the user asking if\",\n \"if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\\\"Using\",\n \"if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set to True, but the\",\n \"self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self,\",\n \"'max_delay'} if not problem_configuration or not all(p in problem_configuration for p in required_parameters):\",\n \"model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start =\",\n \"all pipelines searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols =\",\n \"self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids =\",\n \"if self.max_time is not None: logger.info(\\\"Will stop searching for new pipelines after %d\",\n \"of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False\",\n \"tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments:\",\n \"\\\"\\\"\\\"Find the best pipeline for the data set. Arguments: feature_types (list, optional): list\",\n \"self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize,\",\n \"= max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not\",\n \"should be passed in for the gap and max_delay variables. train_best_pipeline (boolean): Whether\",\n \"that the time in this loop does not count towards the time budget\",\n \"splitting method to use. Defaults to StratifiedKFold. tuner_class: The tuner class to use.\",\n \"logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and\",\n \"threshold. Defaults to True. start_iteration_callback (callable): Function called before each pipeline training iteration.\",\n \"pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration},\",\n \"\\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\":\",\n \"length [n_samples]. Required for supervised learning tasks. problem_type (str or ProblemTypes): type of\",\n \"mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search\",\n \"Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int)\",\n \"between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of\",\n \"is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best\",\n \"< 0: raise ValueError(f\\\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\\\") self.max_time\",\n \"in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class\",\n \"is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except\",\n \"+ 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is set to\",\n \"0. n_jobs (int or None): Non-negative integer describing level of parallelism used for\",\n \"> 0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores)\",\n \"BaseCrossValidator): raise ValueError(\\\"Not a valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective\",\n \"(str): location to find file to load Returns: AutoSearchBase object \\\"\\\"\\\" with open(file_path,\",\n \"for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps\",\n \"sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks\",\n \"n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters\",\n \"> 1.0 or tolerance < 0.0): raise ValueError(\\\"tolerance value must be a float\",\n \"Necessary for \\\"Plotting\\\" documentation, since Sphinx does not work well with instance attributes.\",\n \"so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\\\") else:\",\n \"pipelines searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\\\"id\\\",\",\n \"to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not to optimize the\",\n \"Received {problem_configuration}.\\\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the\",\n \"= False logger.warning(f\\\"Ensembling is set to True, but max_iterations is too small, so\",\n \"if self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >=\",\n \"{elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline =\",\n \"set to {[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type,\",\n \"self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update()\",\n \"+ str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and\",\n \"percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for\",\n \"self.max_iterations) if self.max_time is not None: logger.info(\\\"Will stop searching for new pipelines after\",\n \"If max_iterations and max_time is not set, then max_iterations will default to max_iterations\",\n \"if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids =\",\n \"self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline])\",\n \"self.max_iterations and num_pipelines >= self.max_iterations: return False # check for early stopping if\",\n \"problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective (str, ObjectiveBase): The objective to\",\n \"for early stopping if self.patience is None or self.tolerance is None: return True\",\n \"pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline\",\n \"\\\" % self.max_iterations) if self.max_time is not None: logger.info(\\\"Will stop searching for new\",\n \"is True. Must be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch\",\n \"Number of iterations without improvement to stop search early. Must be positive. If\",\n \"self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is set\",\n \"to pass along other appropriate parameters by default. Defaults to None, which will\",\n \"True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in\",\n \"\\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE:\",\n \"Sphinx does not work well with instance attributes. plot = PipelineSearchPlots def __init__(self,\",\n \"raises an Exception. Callback function takes three positional parameters: the Exception raised, the\",\n \"pipelines. None and 1 are equivalent. If set to -1, all CPUs are\",\n \"search (y/n)? \\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice\",\n \"keyed by pipeline name that maps to a dictionary of scores. Note that\",\n \"function.__name__ else: return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\"\",\n \"evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives,\",\n \"from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline,\",\n \"in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective names allowed in\",\n \"# check for early stopping if self.patience is None or self.tolerance is None:\",\n \"= pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted\",\n \"{_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\"\",\n \"= PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None,\",\n \"run.\\\") run_ensembling = False if run_ensembling and self.max_iterations is not None: # Baseline\",\n \"pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise\",\n \"break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and\",\n \"Dictionary keyed by pipeline name that maps to the fitted pipeline. Note that\",\n \"pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file path Arguments: file_path (str): location to save\",\n \"is too small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to\",\n \"pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch,\",\n \"None, early stopping is disabled. Defaults to None. tolerance (float): Minimum percentage difference\",\n \"add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2,\",\n \"= self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self,\",\n \"as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import\",\n \"import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family\",\n \"_validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is\",\n \"applicable if patience is not None. Defaults to None. allowed_pipelines (list(class)): A list\",\n \"about pipeline. Defaults to False. Returns: Description of specified pipeline. Includes information such\",\n \"should terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop = time.time() while True: choice\",\n \"{self._engine.__class__.__name__} to train and score pipelines.\\\") if self.max_batches is not None: logger.info(f\\\"Searching up\",\n \"error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5):\",\n \"self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling =\",\n \"the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after each\",\n \"{[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params =\",\n \"raised, the traceback, and the AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch\",\n \"non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to\",\n \"the event add_to_rankings is called before search _baseline_cv_scores will be empty so we\",\n \"SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in\",\n \"pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of\",\n \"iterations after that.\\\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True if\",\n \"exit search (y/n)? \\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif\",\n \"to max_iterations of 5. max_time (int, str): Maximum time to search for pipelines.\",\n \"if run_ensembling and self.max_iterations is not None: # Baseline + first batch +\",\n \"not run. Defaults to False. max_batches (int): The maximum number of batches of\",\n \"we will return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name,\",\n \"not for optimizing each pipeline during fit-time. When set to 'auto', chooses: -\",\n \"y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required for supervised\",\n \"Callback function takes three positional parameters: the Exception raised, the traceback, and the\",\n \"else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj\",\n \"score plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \\\"\\\"\\\" if\",\n \"in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class\",\n \"+ \\\"Do you really want to exit search (y/n)? \\\").strip().lower() if choice ==\",\n \"contains scores for all objectives # but also fields like \\\"# Training\\\" and\",\n \"% self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\", \\\".join([model.value for model in self.allowed_model_families]))\",\n \"KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids)\",\n \"= next((obj for obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective\",\n \"\\\"\\\"\\\"Checks cross-validation scores and logs a warning if variance is higher than specified\",\n \"three positional parameters: The pipeline class, the pipeline parameters, and the AutoMLSearch object.\",\n \"else: logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch}\",\n \"with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline):\",\n \"significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement\",\n \"raise ValueError(\\\"Additional objective {} is not compatible with a {} problem.\\\".format(obj.name, self.problem_type.value)) for\",\n \"containing the training results for the new pipeline, an untrained_pipeline containing the parameters\",\n \"= convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches\",\n \"keyed by pipeline name that maps to the fitted pipeline. Note that the\",\n \"input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not\",\n \"PipelineBase: A trained instance of the best pipeline and parameters found during automl\",\n \"adds the results to the automl rankings with the requirement that automl search\",\n \"woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm\",\n \"made in the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original stopping\",\n \"`search_order`: a list describing the order the pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results)\",\n \"\\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary for \\\"Plotting\\\" documentation, since Sphinx\",\n \"logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations\",\n \"\\\"# Training\\\" and \\\"# Testing\\\", so we want to exclude them since #\",\n \"results to the automl rankings with the requirement that automl search has been\",\n \"after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None: best_pipeline\",\n \"Function called after each pipeline training iteration. Callback function takes three positional parameters:\",\n \"get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed\",\n \"= False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched\",\n \"optimize_thresholds (bool): Whether or not to optimize the binary pipeline threshold. Defaults to\",\n \"1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0\",\n \"If True, search should terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop = time.time()\",\n \"pipelines on the training data. This can be helpful for training pipelines once\",\n \"parallelism used for pipelines. None and 1 are equivalent. If set to -1,\",\n \"allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling\",\n \"len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling =\",\n \"training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines\",\n \"y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline\",\n \"= n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to\",\n \"import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import\",\n \"improvement for early stopping. Only applicable if patience is not None. Defaults to\",\n \"self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3,\",\n \"seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in\",\n \"1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives() if isinstance(objective,\",\n \"not set, then max_iterations will default to max_iterations of 5. max_time (int, str):\",\n \"or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement\",\n \"to the data. This is the first pipeline fit during search. \\\"\\\"\\\" if\",\n \"max_time is not set, then max_iterations will default to max_iterations of 5. max_time\",\n \"set to False, returns an untrained pipeline instance. \\\"\\\"\\\" if not self._best_pipeline: raise\",\n \"\\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to\",\n \"in the current AutoML batch produced a score of np.nan on the primary\",\n \"self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines\",\n \"if not (0 < _ensembling_split_size < 1): raise ValueError(f\\\"Ensembling split size must be\",\n \"in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not all(p in\",\n \"{} instead\\\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance < 0.0): raise\",\n \"\\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and\",\n \"True: choice = input(leading_char + \\\"Do you really want to exit search (y/n)?\",\n \"the rankings If self._best_pipeline already exists, check to make sure it is different\",\n \"for estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator\",\n \"underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c\",\n \"been run. Arguments: pipeline (PipelineBase): pipeline to train and evaluate. \\\"\\\"\\\" pipeline_rows =\",\n \"an untrained pipeline instance. Returns: PipelineBase: A trained instance of the best pipeline\",\n \"file_path (str): location to save file pickle_protocol (int): the pickle data stream format.\",\n \"max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline\",\n \"num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1))\",\n \"in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives # but\",\n \"self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed,\",\n \"a dict containing values for at least the gap and max_delay \\\" f\\\"parameters.\",\n \"std all_objective_scores.loc[\\\"coef of var\\\", c] = std / mean if abs(mean) > 0\",\n \"pandas.DataFrame with scoring results from the highest-scoring set of parameters used with each\",\n \">= self.max_iterations: return False # check for early stopping if self.patience is None\",\n \"curr_score num_without_improvement = 0 else: num_without_improvement += 1 if num_without_improvement >= self.patience: logger.info(\\\"\\\\n\\\\n{}\",\n \"for problem type if not empty. random_seed (int): Seed for the random number\",\n \"percentage difference to qualify as score improvement for early stopping. Only applicable if\",\n \"must be a float between 0.0 and 1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience\",\n \"if self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline:\",\n \"be passed in for the gap and max_delay variables. train_best_pipeline (boolean): Whether or\",\n \"exists, check to make sure it is different from the current best pipeline\",\n \"ValueError(\\\"Not a valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is\",\n \"fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores for all objectives # but also\",\n \"self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train,\",\n \"the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of\",\n \"of pipeline evaluations which have been made Returns: int: the number of pipeline\",\n \"that error out during training will not be included in the dictionary but\",\n \"None and self._automl_algorithm.batch_number > 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1,\",\n \"all objective names allowed in automl.\\\") return objective() return objective def __str__(self): def\",\n \"type of pipeline components, problem, training time, cross validation, etc. \\\"\\\"\\\" if pipeline_id\",\n \"the binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before each\",\n \") from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import\",\n \"evaluates a given pipeline then adds the results to the automl rankings with\",\n \"number of batches of pipelines to search. Parameters max_time, and max_iterations have precedence\",\n \"pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns:\",\n \"first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]:\",\n \"try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if\",\n \"is None: raise PipelineNotFoundError(\\\"Pipeline class or parameters not found in automl results\\\") return\",\n \"empty. random_seed (int): Seed for the random number generator. Defaults to 0. n_jobs\",\n \"= False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None default_data_splitter =\",\n \"AutoMLSearch.\\\") return True elif choice == \\\"n\\\": # So that the time in\",\n \"raise ValueError(\\\"user_parameters must be a dict containing values for at least the gap\",\n \"Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for\",\n \"Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False\",\n \"= True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in the rankings If self._best_pipeline\",\n \"tolerance < 0.0): raise ValueError(\\\"tolerance value must be a float between 0.0 and\",\n \"self) desc = f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3]\",\n \"SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException,\",\n \"if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using {self._engine.__class__.__name__} to train and score pipelines.\\\") if self.max_batches\",\n \"best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline evaluations which have been made\",\n \"else curr_score < best_score if score_improved and significant_change: best_score = curr_score num_without_improvement =\",\n \"as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of pipelines on\",\n \"pipelines is one, so ensembling will not run.\\\") run_ensembling = False if run_ensembling\",\n \"1) if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is set to True,\",\n \"= [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives]\",\n \"\\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\":\",\n \"to search over per batch is one, ensembling will not run. Defaults to\",\n \"if self._interrupted: return False # for add_to_rankings if self._searched: return True # Run\",\n \"desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number\",\n \"def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for the data set. Arguments: feature_types\",\n \"mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives:\",\n \"high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if\",\n \"pipeline with. _ensembling_split_size (float): The amount of the training data we'll set aside\",\n \"def results(self): \\\"\\\"\\\"Class that allows access to a copy of the results from\",\n \"objectives): \\\"\\\"\\\"Score a list of pipelines on the given holdout data. Arguments: pipelines\",\n \"pipelines for this problem type are allowed. Setting this field will cause allowed_model_families\",\n \"if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but\",\n \"\\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot)\",\n \"return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a given pipeline then adds\",\n \"Defaults to True. pipeline_parameters (dict): A dict of the parameters used to initalize\",\n \"batches for a total of {self.max_iterations} pipelines. \\\") elif self.max_iterations is not None:\",\n \"False if run_ensembling and self.max_iterations is not None: # Baseline + first batch\",\n \"= self.data_splitter.transform_sample(X_train, y_train) X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train,\",\n \"if self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but\",\n \"as type of pipeline components, problem, training time, cross validation, etc. \\\"\\\"\\\" if\",\n \"ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame,\",\n \"self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline = None self._searched = False self.X_train = infer_feature_types(X_train)\",\n \"= self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted =\",\n \"selecting `best_pipeline`.\\\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file\",\n \"self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else:\",\n \"pipeline. Note that the any pipelines that error out during training will not\",\n \"= {o.name: o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int,\",\n \"will not run. Defaults to False. max_batches (int): The maximum number of batches\",\n \"get_objective(objective, return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__,\",\n \"all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of var\\\", c] =\",\n \".pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine\",\n \"score_pipelines(self, pipelines, X_holdout, y_holdout, objectives): \\\"\\\"\\\"Score a list of pipelines on the given\",\n \"parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self):\",\n \"evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions\",\n \"For n_jobs below -1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean):\",\n \"{len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\") if self.max_batches and self.max_iterations is None: self.show_batch_output\",\n \"the time in this loop does not count towards the time budget (if\",\n \"and the AutoMLSearch object. add_result_callback (callable): Function called after each pipeline training iteration.\",\n \"containing `pipeline_results`: a dict with results from each pipeline, and `search_order`: a list\",\n \"if self.patience is None or self.tolerance is None: return True first_id = self._results['search_order'][0]\",\n \"scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores =\",\n \"import AutoMLSearchException, PipelineNotFoundError from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives,\",\n \"for o in additional_objectives] additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives =\",\n \"False else: leading_char = \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for\",\n \"current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the current AutoML batch produced a score of\",\n \"finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is not None:\",\n \"For example, in time series problems, values should be passed in for the\",\n \"raise ValueError(f\\\"Parameter max_iterations must be None or non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time)\",\n \"1 + n_jobs) are used. ensembling (boolean): If True, runs ensembling in a\",\n \"the best pipeline in the rankings If self._best_pipeline already exists, check to make\",\n \"random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int): pipeline to\",\n \"in Jupyter notebook. Disabled by default in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search()\",\n \"self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted\",\n \"training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family ==\",\n \"too small, so ensembling will not run. Set max_iterations >= {first_ensembling_iteration} to run\",\n \"run at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\")\",\n \"len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective): non_core_objectives = get_non_core_objectives()\",\n \"return_dict (bool): If True, return dictionary of information about pipeline. Defaults to False.\",\n \"The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool): Whether or not\",\n \"None and 1 are equivalent. If set to -1, all CPUs are used.\",\n \"if field in objective_name_to_class: scores[field] += value return {objective: float(score) / n_folds for\",\n \"Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to the fitted pipeline.\",\n \"Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\"\",\n \"be between 0 and 1 exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices,\",\n \"of 5. max_time (int, str): Maximum time to search for pipelines. This will\",\n \"self.max_batches: self.max_batches = 1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if patience and (not\",\n \"names allowed in automl.\\\") return objective() return objective def __str__(self): def _print_list(obj_list): lines\",\n \"X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number of\",\n \"AutoSearchBase object \\\"\\\"\\\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines):\",\n \"compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to the\",\n \"objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data =\",\n \"exception and stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.score_batch(pipelines, X_holdout,\",\n \"not run.\\\") run_ensembling = False if run_ensembling and self.max_iterations is not None: #\",\n \"PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None,\",\n \"\\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\":\",\n \"0.2 _pipelines_per_batch (int): The number of pipelines to train for every batch after\",\n \"+ len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if not (0\",\n \"Must be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The\",\n \"of pipelines on the training data. This can be helpful for training pipelines\",\n \"ModelFamily)): The model families to search. The default of None searches over all\",\n \"{type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time, (int, float)) and max_time < 0: raise\",\n \"logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name\",\n \"ValueError(f\\\"Parameter max_batches must be None or non-negative. Received {max_batches}.\\\") if max_iterations is not\",\n \"X_train = X_train.iloc[train_indices] y_train = y_train.iloc[train_indices] best_pipeline = self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective)\",\n \"_should_continue(self): \\\"\\\"\\\"Given the original stopping criterion and current state, should the search continue?\",\n \"features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str), list(ObjectiveBase)): Objectives used\",\n \"is set to True, but max_iterations is too small, so ensembling will not\",\n \"with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with\",\n \"+ len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if self.max_iterations < first_ensembling_iteration: run_ensembling\",\n \"max_time, and max_iterations have precedence over stopping the search. problem_configuration (dict, None): Additional\",\n \"== []: raise ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if\",\n \"and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration or {} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents\",\n \"results from the highest-scoring set of parameters used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\",\",\n \"type are allowed. Setting this field will cause allowed_model_families to be ignored. allowed_model_families\",\n \"self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter =\",\n \"untrained pipeline instance associated with the provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if\",\n \"None or parameters is None: raise PipelineNotFoundError(\\\"Pipeline class or parameters not found in\",\n \"return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline =\",\n \"in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already been run and will\",\n \"Non-negative integer describing level of parallelism used for pipelines. None and 1 are\",\n \"%s\\\\n\\\" % \\\", \\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot:\",\n \"Must also accepts kwargs, so AutoMLSearch is able to pass along other appropriate\",\n \"self.max_time and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations:\",\n \"Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that maps to a\",\n \"= max_batches self._pipelines_per_batch = _pipelines_per_batch if not self.max_iterations and not self.max_time and not\",\n \"except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type') self.tuner_class =\",\n \"indicates all pipelines for this problem type are allowed. Setting this field will\",\n \"problems, values should be passed in for the gap and max_delay variables. train_best_pipeline\",\n \"return objective def __str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o in obj_list])\",\n \"== ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train =\",\n \"must be a dict containing values for at least the gap and max_delay\",\n \"self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1\",\n \"to 0. n_jobs (int or None): Non-negative integer describing level of parallelism used\",\n \"(int): pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided\",\n \"not run. Set max_iterations >= {first_ensembling_iteration} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run\",\n \"+ rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'}\",\n \"choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice == \\\"n\\\": # So\",\n \"get all objective names allowed in automl.\\\") return objective() return objective def __str__(self):\",\n \"self.max_time is not None: logger.info(\\\"Will stop searching for new pipelines after %d seconds.\\\\n\\\"\",\n \"integer, then the time will be in seconds. For strings, time can be\",\n \"= None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import plotly; skipping\",\n \"ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family\",\n \"self.search_iteration_plot = None if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines()\",\n \"pipelines allowed in the search. The default of None indicates all pipelines for\",\n \"of pipeline evaluations made in the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given\",\n \"= None default_data_splitter = make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter =\",\n \"search. Returns: bool: If True, search should terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\"\",\n \"logger.info(\\\"AutoMLSearch.search() has already been run and will not run again on the same\",\n \"obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not None:\",\n \"or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list.\",\n \"are equivalent. If set to -1, all CPUs are used. For n_jobs below\",\n \"False elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check for early\",\n \"this field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)): The model\",\n \"AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number =\",\n \"(str): location to save file pickle_protocol (int): the pickle data stream format. Returns:\",\n \"_validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration\",\n \"ID of a pipeline training result, returns an untrained instance of the specified\",\n \"and elapsed >= self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return\",\n \"parameters used to train that pipeline during automl search. Arguments: pipeline_id (int): pipeline\",\n \"this problem type are allowed. Setting this field will cause allowed_model_families to be\",\n \"Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs.\",\n \"and \\\"# Testing\\\", so we want to exclude them since # they are\",\n \"40 # Necessary for \\\"Plotting\\\" documentation, since Sphinx does not work well with\",\n \"the pipelines allowed in the search. The default of None indicates all pipelines\",\n \"raise TypeError(f\\\"Parameter max_time must be a float, int, string or None. Received {type(max_time)}\",\n \"and score pipelines.\\\") if self.max_batches is not None: logger.info(f\\\"Searching up to {self.max_batches} batches\",\n \"pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that allows access\",\n \"save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file path Arguments: file_path (str): location\",\n \"and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE:\",\n \"for problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives if obj.name\",\n \"best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\")\",\n \"the data set. Arguments: feature_types (list, optional): list of feature types, either numerical\",\n \"every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches -\",\n \"{} is not compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None:\",\n \"None): Non-negative integer describing level of parallelism used for pipelines. None and 1\",\n \"be helpful for training pipelines once the search is complete. Arguments: pipelines (list(PipelineBase)):\",\n \"logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a given\",\n \"to search for pipelines. This will not start a new pipeline search after\",\n \"for this problem type are allowed. Setting this field will cause allowed_model_families to\",\n \"in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators]\",\n \"num_ensemble_batches == 0: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_batches\",\n \"= get_non_core_objectives() if isinstance(objective, type): if objective in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not\",\n \"_post_evaluation_callback(self, pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline\",\n \"are not scores if field in objective_name_to_class: scores[field] += value return {objective: float(score)\",\n \"training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family ==\",\n \"Returns: bool: True if yes, False if no. \\\"\\\"\\\" if self._interrupted: return False\",\n \"the new pipeline, an untrained_pipeline containing the parameters used during training, and the\",\n \"without improvement to stop search early. Must be positive. If None, early stopping\",\n \"or non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations\",\n \"self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else\",\n \"additional_objectives = [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name:\",\n \"non-negative. Received {max_batches}.\\\") if max_iterations is not None and max_iterations < 0: raise\",\n \"self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All\",\n \"function takes three positional parameters: the Exception raised, the traceback, and the AutoMLSearch\",\n \"(self.max_batches - 1)) if run_ensembling: if not (0 < _ensembling_split_size < 1): raise\",\n \"\\\"\\\"\\\"Finds the best pipeline in the rankings If self._best_pipeline already exists, check to\",\n \"estimated on unseen data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of\",\n \"import defaultdict import cloudpickle import numpy as np import pandas as pd import\",\n \"/ mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format',\",\n \"for ensembler are pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for\",\n \"and num_pipelines >= self.max_iterations: return False # check for early stopping if self.patience\",\n \"ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger\",\n \"self.max_iterations is not None: # Baseline + first batch + each pipeline iteration\",\n \"if allowed_pipelines is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting\",\n \"and max_iterations elapsed = time.time() - self._start if self.max_time and elapsed >= self.max_time:\",\n \"logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator,\",\n \"= time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\"\",\n \"and not self.max_batches: self.max_batches = 1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if patience\",\n \"to train for every batch after the first one. The first batch will\",\n \"so ensembling will not run. Set max_batches >= {ensemble_nth_batch + 1} to run\",\n \"not None and max_batches < 0: raise ValueError(f\\\"Parameter max_batches must be None or\",\n \"for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is\",\n \"== ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id)\",\n \"from the highest-scoring set of parameters used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\")\",\n \"pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted = False if\",\n \"self._interrupted = False if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators\",\n \"= True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches -\",\n \"score improvement for early stopping. Only applicable if patience is not None. Defaults\",\n \"the pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame\",\n \"time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc\",\n \"elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS:\",\n \"and max_time is not set, then max_iterations will default to max_iterations of 5.\",\n \"aside for training ensemble metalearners. Only used when ensembling is True. Must be\",\n \"and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize binary classification pipeline thresholds for: {}\\\".format(self.objective))\",\n \"{\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline])\",\n \"out during training will not be included in the dictionary but the exception\",\n \"+= value return {objective: float(score) / n_folds for objective, score in scores.items()} def\",\n \"patience (int): Number of iterations without improvement to stop search early. Must be\",\n \"if self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch =\",\n \"with the provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise\",\n \"that the any pipelines that error out during scoring will not be included\",\n \"def __str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return '\\\\n'.join(lines)\",\n \"Note that the any pipelines that error out during training will not be\",\n \"{pipeline_name} may not perform as estimated on unseen data.\\\") return high_variance_cv def get_pipeline(self,\",\n \"if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return\",\n \"(int): Maximum number of iterations to search. If max_iterations and max_time is not\",\n \"notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \\\"Beginning pipeline\",\n \"# for add_to_rankings if self._searched: return True # Run at least one pipeline\",\n \"vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose\",\n \"True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in the rankings If self._best_pipeline already\",\n \"should the search continue? Returns: bool: True if yes, False if no. \\\"\\\"\\\"\",\n \"used during training, and the AutoMLSearch object. error_callback (callable): Function called when `search()`\",\n \"self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the\",\n \"show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for the data set. Arguments: feature_types (list, optional):\",\n \"pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for model_family in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids\",\n \"= { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs\",\n \"object at file path Arguments: file_path (str): location to find file to load\",\n \"- self._start elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc =\",\n \"exclude them since # they are not scores if field in objective_name_to_class: scores[field]\",\n \"{get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines\",\n \"try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression)\",\n \"c] = std all_objective_scores.loc[\\\"coef of var\\\", c] = std / mean if abs(mean)\",\n \"maps to the fitted pipeline. Note that the any pipelines that error out\",\n \"plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0\",\n \"required_parameters): raise ValueError(\\\"user_parameters must be a dict containing values for at least the\",\n \"raise ValueError(\\\"Given objective {} is not compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if\",\n \"objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search _baseline_cv_scores\",\n \"be a float between 0.0 and 1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience =\",\n \"1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1) if\",\n \"random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline\",\n \"and thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0] if not (self._best_pipeline\",\n \"[self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for\",\n \"as a 2d array using the X_train argument') if y_train is None: raise\",\n \"AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective names allowed in automl.\\\")\",\n \"full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores = full_rankings[current_batch_idx]['score'] if len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the\",\n \"pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in\",\n \"the time will be in seconds. For strings, time can be specified as\",\n \"def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if len(desc)\",\n \"in the dictionary but the exception and stacktrace will be displayed in the\",\n \"a positive integer. Received {} instead\\\".format(patience)) if tolerance and (tolerance > 1.0 or\",\n \"== 1: logger.warning(\\\"Ensembling is set to True, but the number of unique pipelines\",\n \"- 1) // ensemble_nth_batch if num_ensemble_batches == 0: run_ensembling = False logger.warning(f\\\"Ensembling is\",\n \"pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline {} is not compatible with problem_type {}.\\\".format(pipeline.name,\",\n \"are pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type))\",\n \"full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\", \\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return\",\n \"of (binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback =\",\n \"\\\"\\\"\\\"Saves AutoML object at file path Arguments: file_path (str): location to save file\",\n \"a prompt to the user asking if they want to stop the search.\",\n \"allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None,\",\n \"= PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines\",\n \"Callback function takes three positional parameters: A dictionary containing the training results for\",\n \"the data. This is the first pipeline fit during search. \\\"\\\"\\\" if self.problem_type\",\n \"AutoMLSearch to search again.\\\") return # don't show iteration plot outside of a\",\n \"0: raise ValueError(f\\\"Parameter max_time must be None or non-negative. Received {max_time}.\\\") if max_batches\",\n \"except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id\",\n \"\\\"\\\"\\\"Return the number of pipeline evaluations which have been made Returns: int: the\",\n \"pipeline, evaluation_results): training_time = evaluation_results['training_time'] cv_data = evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline =\",\n \"if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending,\",\n \"the requirement that automl search has been run. Arguments: pipeline (PipelineBase): pipeline to\",\n \"objective to optimize for. Used to propose and rank pipelines, but not for\",\n \"objective names allowed in automl.\\\") return objective() return objective def __str__(self): def _print_list(obj_list):\",\n \"new pipeline search after the duration has elapsed. If it is an integer,\",\n \"len(self.allowed_pipelines) + self._pipelines_per_batch * (self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations\",\n \"in cv_data: for field, value in fold_data['all_objective_scores'].items(): # The 'all_objective_scores' field contains scores\",\n \"Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to the fitted\",\n \"training iteration. Callback function takes three positional parameters: The pipeline class, the pipeline\",\n \"file path Arguments: file_path (str): location to save file pickle_protocol (int): the pickle\",\n \"The model families to search. The default of None searches over all model\",\n \"in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None\",\n \"str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold:\",\n \"and will not run again on the same instance. Re-initialize AutoMLSearch to search\",\n \"(pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required for supervised learning\",\n \"parameters: the Exception raised, the traceback, and the AutoMLSearch object. Must also accepts\",\n \"_add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to the data. This is the first pipeline\",\n \"0.0): raise ValueError(\\\"tolerance value must be a float between 0.0 and 1.0 inclusive.\",\n \"open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object\",\n \"attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None,\",\n \"the search. \\\"\\\"\\\" if X_train is None: raise ValueError('Must specify training data as\",\n \"self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective = get_objective(objective,\",\n \"each pipeline family allowed in the search. \\\"\\\"\\\" if X_train is None: raise\",\n \"used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed by pipeline name that\",\n \"self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if\",\n \"= self.objective_name_to_class[obj_name] # In the event add_to_rankings is called before search _baseline_cv_scores will\",\n \"the fitted pipeline. Note that the any pipelines that error out during training\",\n \"_get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for\",\n \"values as a 1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type)\",\n \"are used. ensembling (boolean): If True, runs ensembling in a separate batch after\",\n \"= len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch if num_ensemble_batches\",\n \"tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] }\",\n \"copy import time from collections import defaultdict import cloudpickle import numpy as np\",\n \"get_estimators(self.problem_type, self.allowed_model_families) logger.debug(f\\\"allowed_estimators set to {[estimator.name for estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train,\",\n \"\\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores)\",\n \"logger.warning(f\\\"Ensembling is set to True, but max_iterations is too small, so ensembling will\",\n \"before each pipeline training iteration. Callback function takes three positional parameters: The pipeline\",\n \"pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def\",\n \"scores = defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field, value\",\n \"to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for\",\n \"check max_time and max_iterations elapsed = time.time() - self._start if self.max_time and elapsed\",\n \"stopping. Only applicable if patience is not None. Defaults to None. allowed_pipelines (list(class)):\",\n \"loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores\",\n \"{} def _handle_keyboard_interrupt(self): \\\"\\\"\\\"Presents a prompt to the user asking if they want\",\n \"max_batches is not None and max_batches < 0: raise ValueError(f\\\"Parameter max_batches must be\",\n \"self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type()\",\n \"self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc +\",\n \"self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\", \\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot\",\n \"of iterations without improvement to stop search early. Must be positive. If None,\",\n \"remove it existing_main_objective = next((obj for obj in additional_objectives if obj.name == self.objective.name),\",\n \"self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output) def _validate_objective(self, objective):\",\n \"if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice == \\\"n\\\": #\",\n \"information about pipeline. Defaults to False. Returns: Description of specified pipeline. Includes information\",\n \"None or non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time\",\n \"None or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score']\",\n \"from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter ) from evalml.exceptions import AutoMLSearchException, PipelineNotFoundError\",\n \"for obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not\",\n \"that automl search has been run. Arguments: pipeline (PipelineBase): pipeline to train and\",\n \") def _pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if\",\n \"of objectives for problem_type, remove it existing_main_objective = next((obj for obj in additional_objectives\",\n \"make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters\",\n \"None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for o in additional_objectives] additional_objectives = [self._validate_objective(obj)\",\n \"best pipeline for the data set. Arguments: feature_types (list, optional): list of feature\",\n \"in non_core_objectives: raise ValueError(f\\\"{objective.name.lower()} is not allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\"\",\n \"_MAX_NAME_LEN = 40 # Necessary for \\\"Plotting\\\" documentation, since Sphinx does not work\",\n \"ensembler are pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\") logger.info(\\\"Training for {}\",\n \"to import plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families\",\n \"ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import (\",\n \"best_score if score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement\",\n \"we want to exclude them since # they are not scores if field\",\n \"(str, ObjectiveBase): The objective to optimize for. Used to propose and rank pipelines,\",\n \"is not None else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines\",\n \"id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change = abs((curr_score - best_score) / best_score)\",\n \"will return # nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan))\",\n \"= self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the\",\n \"tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False,\",\n \"except NameError: show_iteration_plot = False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s. \\\"\",\n \"time (including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores =\",\n \"(list): Custom set of objectives to score on. Will override default objectives for\",\n \"(self.max_batches - 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines)\",\n \"to False. max_batches (int): The maximum number of batches of pipelines to search.\",\n \"pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() / cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High\",\n \"BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback\",\n \"fitted pipeline. Note that the any pipelines that error out during training will\",\n \"{} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our\",\n \"of scores. Note that the any pipelines that error out during scoring will\",\n \"target training data of length [n_samples]. Required for supervised learning tasks. problem_type (str\",\n \"must be a positive integer. Received {} instead\\\".format(patience)) if tolerance and (tolerance >\",\n \"be None or non-negative. Received {max_batches}.\\\") if max_iterations is not None and max_iterations\",\n \"\\\"Plotting\\\" documentation, since Sphinx does not work well with instance attributes. plot =\",\n \"\\\"\\\"\\\"Given the ID of a pipeline training result, returns an untrained instance of\",\n \"True # Run at least one pipeline for every search num_pipelines = self._num_pipelines()\",\n \"Testing\\\", so we want to exclude them since # they are not scores\",\n \"binary pipeline threshold. Defaults to True. start_iteration_callback (callable): Function called before each pipeline\",\n \"f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max\",\n \"copy of the results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with\",\n \"= [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while self._should_continue():\",\n \"data of shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data\",\n \"logger.info(\\\"Searching up to %s pipelines. \\\" % self.max_iterations) if self.max_time is not None:\",\n \"ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm\",\n \"= {'gap', 'max_delay'} if not problem_configuration or not all(p in problem_configuration for p\",\n \"over stopping the search. problem_configuration (dict, None): Additional parameters needed to configure the\",\n \"or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None else {} self.search_iteration_plot\",\n \"(self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\")\",\n \"search. The default of None indicates all pipelines for this problem type are\",\n \"y_holdout, objectives): \\\"\\\"\\\"Score a list of pipelines on the given holdout data. Arguments:\",\n \"objectives (list(str), list(ObjectiveBase)): Objectives used for scoring. Returns: Dict[str, Dict[str, float]]: Dictionary keyed\",\n \"get_ipython except NameError: show_iteration_plot = False log_title(logger, \\\"Beginning pipeline search\\\") logger.info(\\\"Optimizing for %s.\",\n \"between 0.0 and 1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience = patience self.tolerance =\",\n \"in allowed_estimators] if self.allowed_pipelines == []: raise ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines)\",\n \"list of pipelines on the training data. This can be helpful for training\",\n \"import copy import time from collections import defaultdict import cloudpickle import numpy as\",\n \"error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling if\",\n \"field contains scores for all objectives # but also fields like \\\"# Training\\\"\",\n \"of the best pipeline and parameters found during automl search. If `train_best_pipeline` is\",\n \"= all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean\",\n \"is set to False, returns an untrained pipeline instance. Returns: PipelineBase: A trained\",\n \"iteration. Callback function takes three positional parameters: The pipeline class, the pipeline parameters,\",\n \"not obj.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Additional objective {} is not compatible with a {} problem.\\\".format(obj.name,\",\n \"Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize\",\n \"pickle_protocol (int): the pickle data stream format. Returns: None \\\"\\\"\\\" with open(file_path, 'wb')\",\n \"self.max_iterations = 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling:\",\n \"ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline\",\n \"Baseline + first batch + each pipeline iteration + 1 first_ensembling_iteration = (1\",\n \"budget (if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False\",\n \"the any pipelines that error out during scoring will not be included in\",\n \"If set to -1, all CPUs are used. For n_jobs below -1, (n_cpus\",\n \"self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt():\",\n \"not None and max_iterations < 0: raise ValueError(f\\\"Parameter max_iterations must be None or\",\n \"not issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise\",\n \"max_iterations of 5. max_time (int, str): Maximum time to search for pipelines. This\",\n \"n_jobs) are used. ensembling (boolean): If True, runs ensembling in a separate batch\",\n \"problem type if not empty. random_seed (int): Seed for the random number generator.\",\n \"time to search for pipelines. This will not start a new pipeline search\",\n \"every batch after the first one. The first batch will train a baseline\",\n \"able to pass along other appropriate parameters by default. Defaults to None, which\",\n \"at the {first_ensembling_iteration} iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\") if\",\n \"produced a score of np.nan on the primary objective {self.objective}.\\\") self.search_duration = time.time()\",\n \"= self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class\",\n \"the exception and stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.train_batch(pipelines)\",\n \"Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]:\",\n \"objective {self.objective}.\\\") self.search_duration = time.time() - self._start elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch\",\n \"else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines) +\",\n \"{self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds: {self.optimize_thresholds}\\\\n\\\" ) rankings_desc = \\\"\\\" if not self.rankings.empty:\",\n \"cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of pipelines on the training data.\",\n \"{self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add\",\n \"self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices =\",\n \"\\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def\",\n \"using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one\",\n \"except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx =\",\n \"`best_pipeline`.\\\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML object at file path\",\n \"collections import defaultdict import cloudpickle import numpy as np import pandas as pd\",\n \"patience < 0): raise ValueError(\\\"patience value must be a positive integer. Received {}\",\n \"search num_pipelines = self._num_pipelines() if num_pipelines == 0: return True # check max_time\",\n \"optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective =\",\n \"len(current_batch_pipeline_scores) and current_batch_pipeline_scores.isna().all(): raise AutoMLSearchException(f\\\"All pipelines in the current AutoML batch produced a\",\n \"ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap,\",\n \"-cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if\",\n \"stream format. Returns: None \\\"\\\"\\\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol)\",\n \"current_batch_pipelines = [] current_batch_pipeline_scores = [] new_pipeline_ids = [] loop_interrupted = False while\",\n \"List of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name\",\n \"pandas as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots\",\n \"to False. Returns: Description of specified pipeline. Includes information such as type of\",\n \"by default. Defaults to None, which will call `log_error_callback`. additional_objectives (list): Custom set\",\n \"self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better\",\n \"first pipeline fit during search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={})\",\n \"not perform as estimated on unseen data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given\",\n \"pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines\",\n \"__str__(self): def _print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return '\\\\n'.join(lines) def\",\n \"in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str, type(None))): raise TypeError(f\\\"Parameter\",\n \"if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr',\",\n \"if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators = get_estimators(self.problem_type, self.allowed_model_families)\",\n \"if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot = False log_title(logger, \\\"Beginning pipeline search\\\")\",\n \"not allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective names\",\n \"AutoMLSearch object. Must also accepts kwargs, so AutoMLSearch is able to pass along\",\n \"is not None. Defaults to None. allowed_pipelines (list(class)): A list of PipelineBase subclasses\",\n \"below -1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean): If True,\",\n \"= MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class = {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap =\",\n \"std / mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with\",\n \"location to save file pickle_protocol (int): the pickle data stream format. Returns: None\",\n \"allowed in AutoML! \\\" \\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective names allowed\",\n \"instance associated with the provided ID \\\"\\\"\\\" pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is\",\n \"(ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives (list(str),\",\n \"additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else:\",\n \"outside of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot =\",\n \"ensembling is True. Must be between 0 and 1, exclusive. Defaults to 0.2\",\n \"and raises an Exception. Callback function takes three positional parameters: the Exception raised,\",\n \"std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of\",\n \"for regression problems. max_iterations (int): Maximum number of iterations to search. If max_iterations\",\n \"not (self._best_pipeline and self._best_pipeline == self.get_pipeline(best_pipeline['id'])): best_pipeline = self.get_pipeline(best_pipeline['id']) if self._train_best_pipeline: if best_pipeline.model_family\",\n \"dict of the parameters used to initalize a pipeline with. _ensembling_split_size (float): The\",\n \"you really want to exit search (y/n)? \\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting\",\n \"baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\": {\\\"gap\\\": gap,\",\n \"a list of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List\",\n \"all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary` to `multiclass` or\",\n \"continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c] = mean all_objective_scores.loc[\\\"std\\\", c]\",\n \"ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\":\",\n \"batch is one, ensembling will not run. Defaults to False. max_batches (int): The\",\n \"that.\\\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling: ensemble_nth_batch\",\n \"self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices\",\n \"if self.plot: self.search_iteration_plot = self.plot.search_iteration_plot(interactive_plot=show_iteration_plot) self._start = time.time() try: self._add_baseline_pipelines() except KeyboardInterrupt: if\",\n \"non-negative. Received {max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations =\",\n \"is None: return True first_id = self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0\",\n \"with. _ensembling_split_size (float): The amount of the training data we'll set aside for\",\n \"self._engine.train_pipeline(best_pipeline, X_train, y_train, self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number\",\n \"logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted =\",\n \"return_instance=False) self.objective = self._validate_objective(objective) if self.data_splitter is not None and not issubclass(self.data_splitter.__class__, BaseCrossValidator):\",\n \"< 0.0): raise ValueError(\\\"tolerance value must be a float between 0.0 and 1.0\",\n \"our main objective is part of default set of objectives for problem_type, remove\",\n \"holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. X_holdout (ww.DataTable, pd.DataFrame):\",\n \"max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1,\",\n \"pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time,\",\n \"all_objective_scores: if c in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean\",\n \"pickle data stream format. Returns: None \\\"\\\"\\\" with open(file_path, 'wb') as f: cloudpickle.dump(self,\",\n \"self._start += time_in_loop return False else: leading_char = \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find\",\n \") rankings_desc = \\\"\\\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc =\",\n \"run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1) // ensemble_nth_batch\",\n \"= {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines,\",\n \"+= time_in_loop return False else: leading_char = \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the\",\n \"on the training data. This can be helpful for training pipelines once the\",\n \"def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to the data. This is the first\",\n \"early. Must be positive. If None, early stopping is disabled. Defaults to None.\",\n \"make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes, handle_problem_types from evalml.tuners import\",\n \"problem type are allowed. Setting this field will cause allowed_model_families to be ignored.\",\n \"LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass classification problems, and -\",\n \"scores. {pipeline_name} may not perform as estimated on unseen data.\\\") return high_variance_cv def\",\n \"positive integer. Received {} instead\\\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance\",\n \"set to -1, all CPUs are used. For n_jobs below -1, (n_cpus +\",\n \"batch will train a baseline pipline + one of each pipeline family allowed\",\n \"+ 1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\") self.max_iterations\",\n \"get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline,\",\n \"dictionary containing the training results for the new pipeline, an untrained_pipeline containing the\",\n \"/ cv_scores.mean()) > threshold) if high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv >= {threshold})\",\n \"c] = mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of var\\\", c] = std\",\n \"'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot =\",\n \"\\\"n\\\": # So that the time in this loop does not count towards\",\n \"max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None, _ensembling_split_size=0.2, _pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable):\",\n \"precedence over stopping the search. problem_configuration (dict, None): Additional parameters needed to configure\",\n \"pipeline initialized with the parameters used to train that pipeline during automl search.\",\n \"The number of pipelines to train for every batch after the first one.\",\n \"if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\\\"Parameter max_time must be\",\n \"the dictionary but the exception and stacktrace will be displayed in the log.\",\n \"\\\"\\\"\\\"Presents a prompt to the user asking if they want to stop the\",\n \"= 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline\",\n \"'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of pipelines\",\n \"hours. patience (int): Number of iterations without improvement to stop search early. Must\",\n \"\\\"\\\"\\\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a\",\n \"max_batches must be None or non-negative. Received {max_batches}.\\\") if max_iterations is not None\",\n \"= self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time\",\n \"= evaluation_results['cv_data'] cv_scores = evaluation_results['cv_scores'] is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean()\",\n \"and logs a warning if variance is higher than specified threshhold.\\\"\\\"\\\" pipeline_name =\",\n \"If True, runs ensembling in a separate batch after every allowed pipeline class\",\n \"not None else {} self.search_iteration_plot = None self._interrupted = False if self.allowed_pipelines is\",\n \"get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing import split_data from evalml.problem_types import ProblemTypes,\",\n \"= optimize_thresholds self.ensembling = ensembling if objective == 'auto': objective = get_default_primary_search_objective(self.problem_type.value) objective\",\n \"self._best_pipeline is not None: best_pipeline = self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\")\",\n \"to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters\",\n \"( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters: \\\\n{'='*20}\\\\n\\\" f\\\"Objective: {get_objective(self.objective).name}\\\\n\\\" f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\"\",\n \"if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] #\",\n \"other appropriate parameters by default. Defaults to None, which will call `log_error_callback`. additional_objectives\",\n \"search should terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop = time.time() while True:\",\n \"loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break\",\n \"a full list. objective (str, ObjectiveBase): The objective to optimize for. Used to\",\n \"\\\"\\\" if not self.rankings.empty: rankings_str = self.rankings.drop(['parameters'], axis='columns').to_string() rankings_desc = f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\"\",\n \"ww.DataColumn): The target training data of length [n_samples]. Required for supervised learning tasks.\",\n \"else: X_train = self.X_train y_train = self.y_train if hasattr(self.data_splitter, \\\"transform_sample\\\"): train_indices = self.data_splitter.transform_sample(X_train,\",\n \"equivalent. If set to -1, all CPUs are used. For n_jobs below -1,\",\n \"import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import\",\n \"0 and 1 exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _\",\n \"= percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\":\",\n \"a float between 0.0 and 1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience = patience\",\n \"self._start elapsed_time = time_elapsed(self._start) desc = f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN)\",\n \"number generator. Defaults to 0. n_jobs (int or None): Non-negative integer describing level\",\n \"self.show_batch_output = True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches\",\n \"families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary` to `multiclass` or `regression` depending\",\n \"= 1 + len(self.allowed_pipelines) + (self._pipelines_per_batch * (self.max_batches - 1)) if run_ensembling: if\",\n \"= all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict: return pipeline_results def\",\n \"high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\":\",\n \"(callable): Function called after each pipeline training iteration. Callback function takes three positional\",\n \"current best pipeline before training and thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return best_pipeline\",\n \"defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field, value in fold_data['all_objective_scores'].items():\",\n \"{self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm\",\n \"None try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import plotly; skipping pipeline\",\n \"after the duration has elapsed. If it is an integer, then the time\",\n \"= std / mean if abs(mean) > 0 else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\")\",\n \"ValueError(f\\\"Ensembling split size must be between 0 and 1 exclusive, received {_ensembling_split_size}\\\") X_shape\",\n \"import plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm\",\n \"the original stopping criterion and current state, should the search continue? Returns: bool:\",\n \"percent_better_than_baseline[obj_name] = percent_better high_variance_cv = self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = {\",\n \"but the exception and stacktrace will be displayed in the log. \\\"\\\"\\\" return\",\n \"improvement to stop search early. Must be positive. If None, early stopping is\",\n \"an Exception. Callback function takes three positional parameters: the Exception raised, the traceback,\",\n \"not None: # Baseline + first batch + each pipeline iteration + 1\",\n \"towards the time budget (if set) time_in_loop = time.time() - start_of_loop self._start +=\",\n \"= len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary,\",\n \"True. Must be between 0 and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int):\",\n \"ValueError(\\\"Additional objective {} is not compatible with a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline\",\n \"search early...\\\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in self.additional_objectives: if\",\n \"and `search_order`: a list describing the order the pipelines were searched. \\\"\\\"\\\" return\",\n \"X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets for scoring. objectives\",\n \"of pipelines on the given holdout data. Arguments: pipelines (list(PipelineBase)): List of pipelines\",\n \"tasks. problem_type (str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for\",\n \"f, protocol=pickle_protocol) @staticmethod def load(file_path): \\\"\\\"\\\"Loads AutoML object at file path Arguments: file_path\",\n \"logger.info(\\\"Will stop searching for new pipelines after %d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model\",\n \"of PipelineBase subclasses indicating the pipelines allowed in the search. The default of\",\n \"{max_iterations}.\\\") self.max_time = convert_to_seconds(max_time) if isinstance(max_time, str) else max_time self.max_iterations = max_iterations self.max_batches\",\n \"during search. \\\"\\\"\\\" if self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type ==\",\n \"for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families\",\n \"Returns: None \\\"\\\"\\\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f, protocol=pickle_protocol) @staticmethod def\",\n \"iteration. Callback function takes three positional parameters: A dictionary containing the training results\",\n \"CPUs are used. For n_jobs below -1, (n_cpus + 1 + n_jobs) are\",\n \"= allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm = None self._start = 0.0 self._baseline_cv_scores =\",\n \"load(file_path): \\\"\\\"\\\"Loads AutoML object at file path Arguments: file_path (str): location to find\",\n \"inplace=True) return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of the best\",\n \"= [self._validate_objective(obj) for obj in additional_objectives] self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o\",\n \"ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={}) else: pipeline_class\",\n \"return False return True def _validate_problem_type(self): for obj in self.additional_objectives: if not obj.is_defined_for_problem_type(self.problem_type):\",\n \"has already been run and will not run again on the same instance.\",\n \"at file path Arguments: file_path (str): location to find file to load Returns:\",\n \"the search. For example, in time series problems, values should be passed in\",\n \"exclusive, received {_ensembling_split_size}\\\") X_shape = ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train,\",\n \"`binary` to `multiclass` or `regression` depending on the problem type. Note that if\",\n \"> 0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number,\",\n \"is None: raise ValueError('Must specify training data as a 2d array using the\",\n \"ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance\",\n \"max_time < 0: raise ValueError(f\\\"Parameter max_time must be None or non-negative. Received {max_time}.\\\")\",\n \"if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if not problem_configuration or not\",\n \"if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that allows\",\n \"Returns: PipelineBase: A trained instance of the best pipeline and parameters found during\",\n \"import pandas as pd import woodwork as ww from sklearn.model_selection import BaseCrossValidator from\",\n \"`automl_search`. Returns: dict containing `pipeline_results`: a dict with results from each pipeline, and\",\n \"+ 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) * self._pipelines_per_batch + 1)\",\n \"iterations without improvement to stop search early. Must be positive. If None, early\",\n \"a valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is not\",\n \"Must be positive. If None, early stopping is disabled. Defaults to None. tolerance\",\n \"or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration\",\n \"\\\"Use evalml.objectives.utils.get_core_objective_names() \\\" \\\"to get all objective names allowed in automl.\\\") return objective()\",\n \"= desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number >\",\n \"on unseen data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of a\",\n \"_print_list(obj_list): lines = sorted(['\\\\t{}'.format(o.name) for o in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if\",\n \"Only applicable if patience is not None. Defaults to None. allowed_pipelines (list(class)): A\",\n \"True, runs ensembling in a separate batch after every allowed pipeline class has\",\n \"pipline + one of each pipeline family allowed in the search. \\\"\\\"\\\" if\",\n \"and max_batches < 0: raise ValueError(f\\\"Parameter max_batches must be None or non-negative. Received\",\n \"not run again on the same instance. Re-initialize AutoMLSearch to search again.\\\") return\",\n \"run ensembling.\\\") else: logger.info(f\\\"Ensembling will run every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1 +\",\n \"mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name] # In the event add_to_rankings\",\n \"import ( get_logger, log_subtitle, log_title, time_elapsed, update_pipeline ) logger = get_logger(__file__) class AutoMLSearch:\",\n \"plot outside of a jupyter notebook if show_iteration_plot: try: get_ipython except NameError: show_iteration_plot\",\n \"self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id]) except PipelineNotFoundError: pass if self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline,\",\n \"to 'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for multiclass\",\n \"from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import (\",\n \"best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score']\",\n \"result, returns an untrained instance of the specified pipeline initialized with the parameters\",\n \"{max_time}.\\\") if max_batches is not None and max_batches < 0: raise ValueError(f\\\"Parameter max_batches\",\n \"time.time() while True: choice = input(leading_char + \\\"Do you really want to exit\",\n \"self.rankings.iloc[0] best_pipeline_name = best_pipeline[\\\"pipeline_name\\\"] logger.info(f\\\"Best pipeline: {best_pipeline_name}\\\") logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched =\",\n \"handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import\",\n \"start_of_loop self._start += time_in_loop return False else: leading_char = \\\"\\\" def search(self, show_iteration_plot=True):\",\n \"evaluations made in the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def _should_continue(self): \\\"\\\"\\\"Given the original\",\n \"cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives = self._get_mean_cv_scores_for_all_objectives(cv_data, self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives\",\n \"< best_score if score_improved and significant_change: best_score = curr_score num_without_improvement = 0 else:\",\n \"training result, returns an untrained instance of the specified pipeline initialized with the\",\n \"Custom set of objectives to score on. Will override default objectives for problem\",\n \"to 0.2 _pipelines_per_batch (int): The number of pipelines to train for every batch\",\n \"batch_number = 1 if self._automl_algorithm is not None and self._automl_algorithm.batch_number > 0: batch_number\",\n \"to StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds (bool):\",\n \"supervised learning tasks. problem_type (str or ProblemTypes): type of supervised learning problem. See\",\n \"f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\"\",\n \"(list, optional): list of feature types, either numerical or categorical. Categorical features will\",\n \"as np import pandas as pd import woodwork as ww from sklearn.model_selection import\",\n \"def _check_for_high_variance(self, pipeline, cv_scores, threshold=0.2): \\\"\\\"\\\"Checks cross-validation scores and logs a warning if\",\n \"pipeline before training and thresholding\\\"\\\"\\\" if len(self.rankings) == 0: return best_pipeline = self.rankings.iloc[0]\",\n \"if they want to stop the search. Returns: bool: If True, search should\",\n \"not count towards the time budget (if set) time_in_loop = time.time() - start_of_loop\",\n \"\\\"score\\\": cv_score, \\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\":\",\n \"additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part\",\n \"object \\\"\\\"\\\" with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train\",\n \"fields like \\\"# Training\\\" and \\\"# Testing\\\", so we want to exclude them\",\n \"{ \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters, \\\"score\\\": cv_score,\",\n \"check for early stopping if self.patience is None or self.tolerance is None: return\",\n \"pipeline to retrieve Returns: PipelineBase: untrained pipeline instance associated with the provided ID\",\n \"get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary for \\\"Plotting\\\"\",\n \"= ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train, self.y_train, self.ensembling_indices, self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families =\",\n \"time in this loop does not count towards the time budget (if set)\",\n \"data. This is the first pipeline fit during search. \\\"\\\"\\\" if self.problem_type ==\",\n \"= False logger.warning(f\\\"Ensembling is set to True, but max_batches is too small, so\",\n \"if not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations,\",\n \"trained instance of the best pipeline and parameters found during automl search. If\",\n \"Returns: int: the number of pipeline evaluations made in the search \\\"\\\"\\\" return\",\n \"a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True, return\",\n \"allowed_model_families (list(str, ModelFamily)): The model families to search. The default of None searches\",\n \"while True: choice = input(leading_char + \\\"Do you really want to exit search\",\n \"scoring results from the highest-scoring set of parameters used with each pipeline.\\\"\\\"\\\" return\",\n \"patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None, start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True,\",\n \"max_batches=1.\\\\n\\\") if patience and (not isinstance(patience, int) or patience < 0): raise ValueError(\\\"patience\",\n \"if self.allowed_pipelines == []: raise ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling =\",\n \"True. pipeline_parameters (dict): A dict of the parameters used to initalize a pipeline\",\n \"seconds. For strings, time can be specified as seconds, minutes, or hours. patience\",\n \"an iteration vs. score plot in Jupyter notebook. Disabled by default in non-Jupyter\",\n \"encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot in Jupyter notebook.\",\n \"!= self.problem_type: raise ValueError(\\\"Given pipeline {} is not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value))\",\n \"a new pipeline search after the duration has elapsed. If it is an\",\n \"a pandas.DataFrame with scoring results from the highest-scoring set of parameters used with\",\n \"= make_data_splitter(self.X_train, self.y_train, self.problem_type, self.problem_configuration, n_splits=3, shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter\",\n \"want to exclude them since # they are not scores if field in\",\n \"logger.info(\\\"Training for {} problems.\\\".format(pipeline.problem_type)) if self.optimize_thresholds and self.objective.is_defined_for_problem_type(ProblemTypes.BINARY) and self.objective.can_optimize_threshold: logger.info(\\\"Objective to optimize\",\n \"self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is not None\",\n \"\\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values()) rankings_df = rankings_df[full_rankings_cols]\",\n \"None or non-negative. Received {max_time}.\\\") if max_batches is not None and max_batches <\",\n \"model families: %s\\\\n\\\" % \\\", \\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None\",\n \"cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids\",\n \"# Baseline + first batch + each pipeline iteration + 1 first_ensembling_iteration =\",\n \"Returns: PipelineBase: untrained pipeline instance associated with the provided ID \\\"\\\"\\\" pipeline_results =\",\n \"in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std\",\n \"True if run_ensembling: ensemble_nth_batch = len(self.allowed_pipelines) + 1 num_ensemble_batches = (self.max_batches - 1)\",\n \"Required. y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required for\",\n \"tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback self.add_result_callback = add_result_callback self.error_callback = error_callback or\",\n \"in self._automl_algorithm._best_pipeline_info] self._results['pipeline_results'][pipeline_id][\\\"input_pipeline_ids\\\"] = input_pipeline_ids self._results['search_order'].append(pipeline_id) if not is_baseline: score_to_minimize = -cv_score if\",\n \"+ \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not None\",\n \"return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list of pipelines on the training\",\n \"f\\\"\\\\nSearch Results: \\\\n{'='*20}\\\\n{rankings_str}\\\" return search_desc + rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in\",\n \"the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train. Returns:\",\n \"If self._best_pipeline already exists, check to make sure it is different from the\",\n \"rankings_desc def _validate_problem_configuration(self, problem_configuration=None): if self.problem_type in [ProblemTypes.TIME_SERIES_REGRESSION]: required_parameters = {'gap', 'max_delay'} if\",\n \"class, the pipeline parameters, and the AutoMLSearch object. add_result_callback (callable): Function called after\",\n \"self.patience is None or self.tolerance is None: return True first_id = self._results['search_order'][0] best_score\",\n \"qualify as score improvement for early stopping. Only applicable if patience is not\",\n \"main objective is part of default set of objectives for problem_type, remove it\",\n \"pipeline thresholds for: {}\\\".format(self.objective)) logger.info(\\\"Total training time (including CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"])\",\n \"results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters\",\n \"instance of the best pipeline and parameters found during automl search. If `train_best_pipeline`\",\n \"and 1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to\",\n \"\\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id (int): pipeline to describe return_dict (bool): If True,\",\n \"1, exclusive. Defaults to 0.2 _pipelines_per_batch (int): The number of pipelines to train\",\n \"data set. Arguments: feature_types (list, optional): list of feature types, either numerical or\",\n \"- best_score) / best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better\",\n \"{self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in the\",\n \"pipelines to train. X_holdout (ww.DataTable, pd.DataFrame): Holdout features. y_holdout (ww.DataTable, pd.DataFrame): Holdout targets\",\n \"during fit-time. When set to 'auto', chooses: - LogLossBinary for binary classification problems,\",\n \"self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score < best_score if\",\n \"raise ValueError('Must specify training data target values as a 1d vector using the\",\n \"issubclass(self.data_splitter.__class__, BaseCrossValidator): raise ValueError(\\\"Not a valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given\",\n \"str) else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if\",\n \"self.max_batches is not None: logger.info(f\\\"Searching up to {self.max_batches} batches for a total of\",\n \"else: leading_char = \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for the\",\n \"time series problems, values should be passed in for the gap and max_delay\",\n \"\\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] } if pipeline.model_family == ModelFamily.ENSEMBLE: input_pipeline_ids = [self._automl_algorithm._best_pipeline_info[model_family][\\\"id\\\"] for\",\n \"logger.info(f\\\"Best pipeline {self.objective.name}: {best_pipeline['score']:3f}\\\") self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline\",\n \"= self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}, \\\"Time Series Baseline Estimator\\\":\",\n \"TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from evalml.pipelines.components.utils import get_estimators from evalml.pipelines.utils import make_pipeline from evalml.preprocessing\",\n \"pipeline_results = self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not found in automl\",\n \"next((obj for obj in additional_objectives if obj.name == self.objective.name), None) if existing_main_objective is\",\n \"_pipelines_per_batch=5): \\\"\\\"\\\"Automated pipeline search Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of\",\n \"Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting:\",\n \"if high_variance_cv: logger.warning(f\\\"High coefficient of variation (cv >= {threshold}) within cross validation scores.\",\n \"- 1 - num_ensemble_batches) + num_ensemble_batches) else: self.max_iterations = 1 + len(self.allowed_pipelines) +\",\n \"self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1 logger.info(\\\"Using default limit\",\n \"is_baseline = pipeline.model_family == ModelFamily.BASELINE cv_score = cv_scores.mean() percent_better_than_baseline = {} mean_cv_all_objectives =\",\n \"if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not found in automl results\\\") pipeline_class =\",\n \"else np.inf all_objective_scores = all_objective_scores.fillna(\\\"-\\\") with pd.option_context('display.float_format', '{:.3f}'.format, 'expand_frame_repr', False): logger.info(all_objective_scores) if return_dict:\",\n \"be run before selecting `best_pipeline`.\\\") return self._best_pipeline def save(self, file_path, pickle_protocol=cloudpickle.DEFAULT_PROTOCOL): \\\"\\\"\\\"Saves AutoML\",\n \"to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run at the {first_ensembling_iteration} iteration and every\",\n \"Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0) std = all_objective_scores[c].std(axis=0) all_objective_scores.loc[\\\"mean\\\", c]\",\n \"_pre_evaluation_callback(self, pipeline): if self.start_iteration_callback: self.start_iteration_callback(pipeline.__class__, pipeline.parameters, self) desc = f\\\"{pipeline.name}\\\" if len(desc) >\",\n \"number of iterations to search. If max_iterations and max_time is not set, then\",\n \"self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\": type(pipeline), \\\"pipeline_summary\\\": pipeline.summary, \\\"parameters\\\": pipeline.parameters,\",\n \"if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\") pipeline = self.get_pipeline(pipeline_id) pipeline_results\",\n \"in required_parameters): raise ValueError(\\\"user_parameters must be a dict containing values for at least\",\n \"automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments:\",\n \"\\\"\\\"\\\"Loads AutoML object at file path Arguments: file_path (str): location to find file\",\n \"max_time must be None or non-negative. Received {max_time}.\\\") if max_batches is not None\",\n \"= None self._start = 0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration\",\n \"1): raise ValueError(f\\\"Ensembling split size must be between 0 and 1 exclusive, received\",\n \"check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set\",\n \"search. \\\"\\\"\\\" if X_train is None: raise ValueError('Must specify training data as a\",\n \"results from `automl_search`. Returns: dict containing `pipeline_results`: a dict with results from each\",\n \"self.search_iteration_plot: self.search_iteration_plot.update() if self.add_result_callback: self.add_result_callback(self._results['pipeline_results'][pipeline_id], pipeline, self) return pipeline_id def _check_for_high_variance(self, pipeline, cv_scores,\",\n \"allowed in automl.\\\") return objective() return objective def __str__(self): def _print_list(obj_list): lines =\",\n \"patience and (not isinstance(patience, int) or patience < 0): raise ValueError(\\\"patience value must\",\n \"@property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from the highest-scoring set\",\n \"\\\"score\\\", \\\"validation_score\\\", \\\"percent_better_than_baseline\\\", \\\"high_variance_cv\\\", \\\"parameters\\\"] if not self._results['pipeline_results']: return pd.DataFrame(columns=full_rankings_cols) rankings_df = pd.DataFrame(self._results['pipeline_results'].values())\",\n \"stopping if self.patience is None or self.tolerance is None: return True first_id =\",\n \"of pipelines to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that\",\n \"def __init__(self, X_train=None, y_train=None, problem_type=None, objective='auto', max_iterations=None, max_time=None, patience=None, tolerance=None, data_splitter=None, allowed_pipelines=None, allowed_model_families=None,\",\n \"c] = std / mean if abs(mean) > 0 else np.inf all_objective_scores =\",\n \"to False, returns an untrained pipeline instance. Returns: PipelineBase: A trained instance of\",\n \"PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine\",\n \"over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see options. Change `binary` to `multiclass`\",\n \"such as type of pipeline components, problem, training time, cross validation, etc. \\\"\\\"\\\"\",\n \"= defaultdict(int) n_folds = len(cv_data) for fold_data in cv_data: for field, value in\",\n \"search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for the data set. Arguments: feature_types (list,\",\n \"dict containing `pipeline_results`: a dict with results from each pipeline, and `search_order`: a\",\n \"True, but max_iterations is too small, so ensembling will not run. Set max_iterations\",\n \"is one, ensembling will not run. Defaults to False. max_batches (int): The maximum\",\n \"level of parallelism used for pipelines. None and 1 are equivalent. If set\",\n \"estimator in allowed_estimators]}\\\") self.allowed_pipelines = [make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in\",\n \"so ensembling will not run.\\\") run_ensembling = False if run_ensembling and self.max_iterations is\",\n \"{\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds\",\n \"searched\\\"\\\"\\\" ascending = True if self.objective.greater_is_better: ascending = False full_rankings_cols = [\\\"id\\\", \\\"pipeline_name\\\",\",\n \"will run every {ensemble_nth_batch} batches.\\\") self.max_iterations = (1 + len(self.allowed_pipelines) + self._pipelines_per_batch *\",\n \"{self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives:\",\n \"a 1d vector using the y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError:\",\n \"to `multiclass` or `regression` depending on the problem type. Note that if allowed_pipelines\",\n \"{} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline = train_best_pipeline self._best_pipeline =\",\n \"= self.get_pipeline(pipeline_id) pipeline_results = self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler\",\n \"is disabled. Defaults to None. tolerance (float): Minimum percentage difference to qualify as\",\n \"pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property def full_rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results\",\n \"in pipeline_results[\\\"cv_data\\\"]] all_objective_scores = pd.DataFrame(all_objective_scores) for c in all_objective_scores: if c in [\\\"#\",\n \"shape [n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of length\",\n \"and parameters found during automl search. If `train_best_pipeline` is set to False, returns\",\n \"Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner: {self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional\",\n \"plotly; skipping pipeline search plotting\\\\n\\\") self.allowed_pipelines = allowed_pipelines self.allowed_model_families = allowed_model_families self._automl_algorithm =\",\n \"_pipelines_per_batch if not self.max_iterations and not self.max_time and not self.max_batches: self.max_batches = 1\",\n \"integer. Received {} instead\\\".format(patience)) if tolerance and (tolerance > 1.0 or tolerance <\",\n \"part of default set of objectives for problem_type, remove it existing_main_objective = next((obj\",\n \"returns an untrained pipeline instance. Returns: PipelineBase: A trained instance of the best\",\n \"nan for the base score. percent_better = objective_class.calculate_percent_difference(mean_cv_all_objectives[obj_name], self._baseline_cv_scores.get(obj_name, np.nan)) percent_better_than_baseline[obj_name] = percent_better\",\n \"(int, float)) and max_time < 0: raise ValueError(f\\\"Parameter max_time must be None or\",\n \"self.optimize_thresholds, self.objective) self._best_pipeline = best_pipeline def _num_pipelines(self): \\\"\\\"\\\"Return the number of pipeline evaluations\",\n \"= mean all_objective_scores.loc[\\\"std\\\", c] = std all_objective_scores.loc[\\\"coef of var\\\", c] = std /\",\n \"ValueError(\\\"No allowed pipelines to search\\\") check_all_pipeline_names_unique(self.allowed_pipelines) run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines)\",\n \"not work well with instance attributes. plot = PipelineSearchPlots def __init__(self, X_train=None, y_train=None,\",\n \"float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results): training_time\",\n \"must be None or non-negative. Received {max_time}.\\\") if max_batches is not None and\",\n \"using the X_train argument') if y_train is None: raise ValueError('Must specify training data\",\n \"will not run again on the same instance. Re-initialize AutoMLSearch to search again.\\\")\",\n \"self.max_iterations: return False # check for early stopping if self.patience is None or\",\n \"self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={}) elif self.problem_type == ProblemTypes.REGRESSION: baseline = MeanBaselineRegressionPipeline(parameters={})\",\n \"Defaults to StratifiedKFold. tuner_class: The tuner class to use. Defaults to SKOptTuner. optimize_thresholds\",\n \"= 1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if patience and (not isinstance(patience, int)\",\n \"None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is part of default\",\n \"self.problem_type == ProblemTypes.BINARY: baseline = ModeBaselineBinaryPipeline(parameters={}) elif self.problem_type == ProblemTypes.MULTICLASS: baseline = ModeBaselineMulticlassPipeline(parameters={})\",\n \"not self.max_batches: self.max_batches = 1 logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if patience and\",\n \"show_iteration_plot (boolean, True): Shows an iteration vs. score plot in Jupyter notebook. Disabled\",\n \"set to 'auto', chooses: - LogLossBinary for binary classification problems, - LogLossMulticlass for\",\n \"of a pipeline training result, returns an untrained instance of the specified pipeline\",\n \"each pipeline iteration + 1 first_ensembling_iteration = (1 + len(self.allowed_pipelines) + len(self.allowed_pipelines) *\",\n \"and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set to True, but the number of\",\n \"{_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or [])}\\\\n\\\" f\\\"Random Seed: {self.random_seed}\\\\n\\\" f\\\"n_jobs: {self.n_jobs}\\\\n\\\" f\\\"Optimize Thresholds:\",\n \"= error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds self.ensembling = ensembling\",\n \"problem_configuration (dict, None): Additional parameters needed to configure the search. For example, in\",\n \"self.max_iterations < first_ensembling_iteration: run_ensembling = False logger.warning(f\\\"Ensembling is set to True, but max_iterations\",\n \"for at least the gap and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return problem_configuration\",\n \"except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending') break try: new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines)\",\n \"pipeline.name] for parameter in pipeline_rows['parameters']: if pipeline.parameters == parameter: return self._engine.evaluate_batch([pipeline]) self._find_best_pipeline() @property\",\n \"after that.\\\") if self.max_batches and self.max_iterations is None: self.show_batch_output = True if run_ensembling:\",\n \"elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check for early stopping\",\n \"f\\\"Max Time: {self.max_time}\\\\n\\\" f\\\"Max Iterations: {self.max_iterations}\\\\n\\\" f\\\"Max Batches: {self.max_batches}\\\\n\\\" f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or\",\n \"either numerical or categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True):\",\n \"0.0 self._baseline_cv_scores = {} self.show_batch_output = False self._validate_problem_type() self.problem_configuration = self._validate_problem_configuration(problem_configuration) self._train_best_pipeline =\",\n \"not loop_interrupted: current_batch_pipelines = self._automl_algorithm.next_batch() except StopIteration: logger.info('AutoML Algorithm out of recommendations, ending')\",\n \"else max_time self.max_iterations = max_iterations self.max_batches = max_batches self._pipelines_per_batch = _pipelines_per_batch if not\",\n \"not start a new pipeline search after the duration has elapsed. If it\",\n \"start_iteration_callback=None, add_result_callback=None, error_callback=None, additional_objectives=None, random_seed=0, n_jobs=-1, tuner_class=None, optimize_thresholds=True, ensembling=False, max_batches=None, problem_configuration=None, train_best_pipeline=True, pipeline_parameters=None,\",\n \"y_train is None: raise ValueError('Must specify training data target values as a 1d\",\n \"IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import (\",\n \"Training\\\" and \\\"# Testing\\\", so we want to exclude them since # they\",\n \"\\\"high_variance_cv\\\": high_variance_cv, \\\"training_time\\\": training_time, \\\"cv_data\\\": cv_data, \\\"percent_better_than_baseline_all_objectives\\\": percent_better_than_baseline, \\\"percent_better_than_baseline\\\": percent_better_than_baseline[self.objective.name], \\\"validation_score\\\": cv_scores[0] }\",\n \"PipelineNotFoundError(\\\"Pipeline not found in automl results\\\") pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if\",\n \"but the number of unique pipelines is one, so ensembling will not run.\\\")\",\n \"return True # Run at least one pipeline for every search num_pipelines =\",\n \"is provided, this parameter will be ignored. data_splitter (sklearn.model_selection.BaseCrossValidator): Data splitting method to\",\n \"fit-time. When set to 'auto', chooses: - LogLossBinary for binary classification problems, -\",\n \"for the random number generator. Defaults to 0. n_jobs (int or None): Non-negative\",\n \"{self.tuner_class.__name__}\\\\n\\\" f\\\"Start Iteration Callback: {_get_funct_name(self.start_iteration_callback)}\\\\n\\\" f\\\"Add Result Callback: {_get_funct_name(self.add_result_callback)}\\\\n\\\" f\\\"Additional Objectives: {_print_list(self.additional_objectives or\",\n \"import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from evalml.automl.callbacks import log_error_callback from evalml.automl.engine import\",\n \"with a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type\",\n \"{} is not compatible with problem_type {}.\\\".format(pipeline.name, self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline\",\n \"logger.debug(f\\\"allowed_model_families set to {self.allowed_model_families}\\\") if len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params\",\n \"= get_logger(__file__) class AutoMLSearch: \\\"\\\"\\\"Automated Pipeline search.\\\"\\\"\\\" _MAX_NAME_LEN = 40 # Necessary for\",\n \"raise ValueError(f\\\"Parameter max_batches must be None or non-negative. Received {max_batches}.\\\") if max_iterations is\",\n \"iteration and every {len(self.allowed_pipelines) * self._pipelines_per_batch} iterations after that.\\\") if self.max_batches and self.max_iterations\",\n \"The default of None indicates all pipelines for this problem type are allowed.\",\n \"% \\\", \\\".join([model.value for model in self.allowed_model_families])) self.search_iteration_plot = None if self.plot: self.search_iteration_plot\",\n \"ValueError(\\\"user_parameters must be a dict containing values for at least the gap and\",\n \"training ensemble metalearners. Only used when ensembling is True. Must be between 0\",\n \"and 1.0 inclusive. Received {} instead\\\".format(tolerance)) self.patience = patience self.tolerance = tolerance or\",\n \"\\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from the highest-scoring set of parameters used\",\n \"set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False else: leading_char\",\n \"if patience is not None. Defaults to None. allowed_pipelines (list(class)): A list of\",\n \"time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False else: leading_char =\",\n \"pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline {}\",\n \"pipeline for the data set. Arguments: feature_types (list, optional): list of feature types,\",\n \"In the event add_to_rankings is called before search _baseline_cv_scores will be empty so\",\n \"if variance is higher than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std()\",\n \"ww.DataTable(np.arange(self.X_train.shape[0])) _, ensembling_indices, _, _ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices =\",\n \"Defaults to True. start_iteration_callback (callable): Function called before each pipeline training iteration. Callback\",\n \"values for at least the gap and max_delay \\\" f\\\"parameters. Received {problem_configuration}.\\\") return\",\n \"%s. \\\" % self.objective.name) logger.info(\\\"{} score is better.\\\\n\\\".format('Greater' if self.objective.greater_is_better else 'Lower')) logger.info(f\\\"Using\",\n \"from sklearn.model_selection import BaseCrossValidator from .pipeline_search_plots import PipelineSearchPlots from evalml.automl.automl_algorithm import IterativeAlgorithm from\",\n \"new_pipeline_ids = self._engine.evaluate_batch(current_batch_pipelines) loop_interrupted = False except KeyboardInterrupt: loop_interrupted = True if self._handle_keyboard_interrupt():\",\n \"once the search is complete. Arguments: pipelines (list(PipelineBase)): List of pipelines to train.\",\n \"pipeline (PipelineBase): pipeline to train and evaluate. \\\"\\\"\\\" pipeline_rows = self.full_rankings[self.full_rankings['pipeline_name'] == pipeline.name]\",\n \"or None. Received {type(max_time)} with value {str(max_time)}..\\\") if isinstance(max_time, (int, float)) and max_time\",\n \"path Arguments: file_path (str): location to save file pickle_protocol (int): the pickle data\",\n \"one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines == 0: return\",\n \"= f\\\"\\\\nSearch finished after {elapsed_time}\\\" desc = desc.ljust(self._MAX_NAME_LEN) logger.info(desc) self._find_best_pipeline() if self._best_pipeline is\",\n \"# check max_time and max_iterations elapsed = time.time() - self._start if self.max_time and\",\n \"by default in non-Jupyter enviroments. \\\"\\\"\\\" if self._searched: logger.info(\\\"AutoMLSearch.search() has already been run\",\n \"of {self.max_iterations} pipelines. \\\") elif self.max_iterations is not None: logger.info(\\\"Searching up to %s\",\n \"for multiclass classification problems, and - R2 for regression problems. max_iterations (int): Maximum\",\n \"learning tasks. problem_type (str or ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types\",\n \"len(self.problem_configuration): pipeline_params = {**{'pipeline': self.problem_configuration}, **self.pipeline_parameters} else: pipeline_params = self.pipeline_parameters self._automl_algorithm = IterativeAlgorithm(\",\n \"to qualify as score improvement for early stopping. Only applicable if patience is\",\n \"= True if self._handle_keyboard_interrupt(): break full_rankings = self.full_rankings current_batch_idx = full_rankings['id'].isin(new_pipeline_ids) current_batch_pipeline_scores =\",\n \"AutoMLSearch is able to pass along other appropriate parameters by default. Defaults to\",\n \"self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name, \\\"pipeline_class\\\":\",\n \"0: batch_number = self._automl_algorithm.batch_number update_pipeline(logger, desc, len(self._results['pipeline_results']) + 1, self.max_iterations, self._start, batch_number, self.show_batch_output)\",\n \"results for the new pipeline, an untrained_pipeline containing the parameters used during training,\",\n \"return {objective: float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline,\",\n \"[make_pipeline(self.X_train, self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []:\",\n \"not is_baseline: score_to_minimize = -cv_score if self.objective.greater_is_better else cv_score try: self._automl_algorithm.add_result(score_to_minimize, pipeline, self._results['pipeline_results'][pipeline_id])\",\n \"-1, (n_cpus + 1 + n_jobs) are used. ensembling (boolean): If True, runs\",\n \"baseline pipline + one of each pipeline family allowed in the search. \\\"\\\"\\\"\",\n \"the search. The default of None indicates all pipelines for this problem type\",\n \"in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\") logger.debug(f\\\"allowed_model_families set to\",\n \"= self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score = self._results['pipeline_results'][id]['score'] significant_change\",\n \"a pipeline with. _ensembling_split_size (float): The amount of the training data we'll set\",\n \"AutoML batch produced a score of np.nan on the primary objective {self.objective}.\\\") self.search_duration\",\n \"improvement. Stopping search early...\\\".format(self.patience)) return False return True def _validate_problem_type(self): for obj in\",\n \"self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise ValueError(\\\"Given pipeline {} is not\",\n \"cross validation, etc. \\\"\\\"\\\" if pipeline_id not in self._results['pipeline_results']: raise PipelineNotFoundError(\\\"Pipeline not found\\\")\",\n \"objective is part of default set of objectives for problem_type, remove it existing_main_objective\",\n \"unseen data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of a pipeline\",\n \"self.problem_type.value)) def _add_baseline_pipelines(self): \\\"\\\"\\\"Fits a baseline pipeline to the data. This is the\",\n \"ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids'])) log_subtitle(logger, \\\"Training\\\")\",\n \"= self._results['search_order'][0] best_score = self._results['pipeline_results'][first_id]['score'] num_without_improvement = 0 for id in self._results['search_order'][1:]: curr_score\",\n \"objectives to score on. Will override default objectives for problem type if not\",\n \"= add_result_callback self.error_callback = error_callback or log_error_callback self.data_splitter = data_splitter self.optimize_thresholds = optimize_thresholds\",\n \"to a copy of the results from `automl_search`. Returns: dict containing `pipeline_results`: a\",\n \"self._train_best_pipeline: if best_pipeline.model_family == ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train =\",\n \"= self.results['pipeline_results'].get(pipeline_id) if pipeline_results is None: raise PipelineNotFoundError(\\\"Pipeline not found in automl results\\\")\",\n \"stacktrace will be displayed in the log. \\\"\\\"\\\" return self._engine.score_batch(pipelines, X_holdout, y_holdout, objectives)\",\n \"None self._searched = False self.X_train = infer_feature_types(X_train) self.y_train = infer_feature_types(y_train) self.ensembling_indices = None\",\n \"data stream format. Returns: None \\\"\\\"\\\" with open(file_path, 'wb') as f: cloudpickle.dump(self, f,\",\n \"search. If `train_best_pipeline` is set to False, returns an untrained pipeline instance. \\\"\\\"\\\"\",\n \"of iterations to search. If max_iterations and max_time is not set, then max_iterations\",\n \"{objective: float(score) / n_folds for objective, score in scores.items()} def _post_evaluation_callback(self, pipeline, evaluation_results):\",\n \"unique pipelines to search over per batch is one, ensembling will not run.\",\n \"empty so we will return # nan for the base score. percent_better =\",\n \"variance is higher than specified threshhold.\\\"\\\"\\\" pipeline_name = pipeline.name high_variance_cv = bool(abs(cv_scores.std() /\",\n \"(binary, multiclass, regression) as problem_type') self.tuner_class = tuner_class or SKOptTuner self.start_iteration_callback = start_iteration_callback\",\n \"= handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary, multiclass, regression) as problem_type')\",\n \"it is an integer, then the time will be in seconds. For strings,\",\n \"to describe return_dict (bool): If True, return dictionary of information about pipeline. Defaults\",\n \"= False if self.allowed_pipelines is None: logger.info(\\\"Generating pipelines to search over...\\\") allowed_estimators =\",\n \"self.random_seed = random_seed self.n_jobs = n_jobs self.plot = None try: self.plot = PipelineSearchPlots(self)\",\n \"if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines with IDs: \\\" +\",\n \"ProblemTypes): type of supervised learning problem. See evalml.problem_types.ProblemType.all_problem_types for a full list. objective\",\n \"passed in for the gap and max_delay variables. train_best_pipeline (boolean): Whether or not\",\n \"used to initalize a pipeline with. _ensembling_split_size (float): The amount of the training\",\n \"def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the ID of a pipeline training result, returns an\",\n \"rankings_df = rankings_df[full_rankings_cols] rankings_df.sort_values(\\\"score\\\", ascending=ascending, inplace=True) rankings_df.reset_index(drop=True, inplace=True) return rankings_df @property def best_pipeline(self):\",\n \"value must be a float between 0.0 and 1.0 inclusive. Received {} instead\\\".format(tolerance))\",\n \"Minimum percentage difference to qualify as score improvement for early stopping. Only applicable\",\n \"inclusive. Received {} instead\\\".format(tolerance)) self.patience = patience self.tolerance = tolerance or 0.0 self._results\",\n \"rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results from the highest-scoring set of parameters\",\n \"ensembling in a separate batch after every allowed pipeline class has been iterated\",\n \"order the pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a\",\n \"is able to pass along other appropriate parameters by default. Defaults to None,\",\n \"unique pipelines is one, so ensembling will not run.\\\") run_ensembling = False if\",\n \"the number of pipeline evaluations made in the search \\\"\\\"\\\" return len(self._results['pipeline_results']) def\",\n \"as seconds, minutes, or hours. patience (int): Number of iterations without improvement to\",\n \"(y/n)? \\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice ==\",\n \"a score of np.nan on the primary objective {self.objective}.\\\") self.search_duration = time.time() -\",\n \"searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns a pandas.DataFrame with scoring results\",\n \"var\\\", c] = std / mean if abs(mean) > 0 else np.inf all_objective_scores\",\n \"n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples]. Required\",\n \"If `train_best_pipeline` is set to False, returns an untrained pipeline instance. Returns: PipelineBase:\",\n \"(boolean): Whether or not to train the best pipeline before returning it. Defaults\",\n \"problem_configuration for p in required_parameters): raise ValueError(\\\"user_parameters must be a dict containing values\",\n \"the number of unique pipelines to search over per batch is one, ensembling\",\n \"allowed. Setting this field will cause allowed_model_families to be ignored. allowed_model_families (list(str, ModelFamily)):\",\n \"does not count towards the time budget (if set) time_in_loop = time.time() -\",\n \"early stopping if self.patience is None or self.tolerance is None: return True first_id\",\n \"a {} problem.\\\".format(self.objective.name, self.problem_type.value)) if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if\",\n \"/ best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score\",\n \"total of {self.max_iterations} pipelines. \\\") elif self.max_iterations is not None: logger.info(\\\"Searching up to\",\n \"evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline,\",\n \"Set max_batches >= {ensemble_nth_batch + 1} to run ensembling.\\\") else: logger.info(f\\\"Ensembling will run\",\n \"stop search early. Must be positive. If None, early stopping is disabled. Defaults\",\n \"to search. If max_iterations and max_time is not set, then max_iterations will default\",\n \"ModelFamily.ENSEMBLE: X_train, y_train = self.X_train.iloc[self.ensembling_indices], self.y_train.iloc[self.ensembling_indices] else: X_train = self.X_train y_train = self.y_train\",\n \"and significant_change: best_score = curr_score num_without_improvement = 0 else: num_without_improvement += 1 if\",\n \"Arguments: X_train (pd.DataFrame, ww.DataTable): The input training data of shape [n_samples, n_features]. Required.\",\n \"IterativeAlgorithm( max_iterations=self.max_iterations, allowed_pipelines=self.allowed_pipelines, tuner_class=self.tuner_class, random_seed=self.random_seed, n_jobs=self.n_jobs, number_features=self.X_train.shape[1], pipelines_per_batch=self._pipelines_per_batch, ensembling=run_ensembling, pipeline_params=pipeline_params ) def _pre_evaluation_callback(self,\",\n \"3] + \\\"...\\\" desc = desc.ljust(AutoMLSearch._MAX_NAME_LEN) batch_number = 1 if self._automl_algorithm is not\",\n \"pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for fold in pipeline_results[\\\"cv_data\\\"]] all_objective_scores\",\n \"to True, but max_batches is too small, so ensembling will not run. Set\",\n \"self._find_best_pipeline() @property def results(self): \\\"\\\"\\\"Class that allows access to a copy of the\",\n \"{}, 'search_order': [], 'errors': [] } self.random_seed = random_seed self.n_jobs = n_jobs self.plot\",\n \"additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives} if not\",\n \"max_batches < 0: raise ValueError(f\\\"Parameter max_batches must be None or non-negative. Received {max_batches}.\\\")\",\n \"train a baseline pipline + one of each pipeline family allowed in the\",\n \"try: self.plot = PipelineSearchPlots(self) except ImportError: logger.warning(\\\"Unable to import plotly; skipping pipeline search\",\n \"self.tolerance = tolerance or 0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors':\",\n \"to train the best pipeline before returning it. Defaults to True. pipeline_parameters (dict):\",\n \"different from the current best pipeline before training and thresholding\\\"\\\"\\\" if len(self.rankings) ==\",\n \"raise ValueError(\\\"Not a valid data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {}\",\n \"value {str(max_time)}..\\\") if isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\\\"Parameter max_time\",\n \"allows access to a copy of the results from `automl_search`. Returns: dict containing\",\n \"train for every batch after the first one. The first batch will train\",\n \"self, should_continue_callback=self._should_continue, pre_evaluation_callback=self._pre_evaluation_callback, post_evaluation_callback=self._post_evaluation_callback) self.allowed_model_families = list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set\",\n \"= {ProblemTypes.TIME_SERIES_REGRESSION: TimeSeriesBaselineRegressionPipeline, ProblemTypes.TIME_SERIES_MULTICLASS: TimeSeriesBaselineMulticlassPipeline, ProblemTypes.TIME_SERIES_BINARY: TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay']\",\n \"state, should the search continue? Returns: bool: True if yes, False if no.\",\n \"best_score) > self.tolerance score_improved = curr_score > best_score if self.objective.greater_is_better else curr_score <\",\n \"run_ensembling = self.ensembling if run_ensembling and len(self.allowed_pipelines) == 1: logger.warning(\\\"Ensembling is set to\",\n \"logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice == \\\"n\\\": # So that the time\",\n \"self.objective.name), None) if existing_main_objective is not None: additional_objectives.remove(existing_main_objective) else: additional_objectives = [get_objective(o) for\",\n \"untrained pipeline instance. \\\"\\\"\\\" if not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be run\",\n \"bool: If True, search should terminate early \\\"\\\"\\\" leading_char = \\\"\\\\n\\\" start_of_loop =\",\n \"if callable(function): return function.__name__ else: return None search_desc = ( f\\\"{handle_problem_types(self.problem_type).name} Search\\\\n\\\\n\\\" f\\\"Parameters:\",\n \"of pipeline components, problem, training time, cross validation, etc. \\\"\\\"\\\" if pipeline_id not\",\n \"1 are equivalent. If set to -1, all CPUs are used. For n_jobs\",\n \"objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is not compatible with a {} problem.\\\".format(self.objective.name, self.problem_type.value))\",\n \"get_non_core_objectives, get_objective ) from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline\",\n \"search. If max_iterations and max_time is not set, then max_iterations will default to\",\n \"problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []: if pipeline.problem_type != self.problem_type: raise\",\n \"the order the pipelines were searched. \\\"\\\"\\\" return copy.deepcopy(self._results) @property def rankings(self): \\\"\\\"\\\"Returns\",\n \"self.max_time: return False elif self.max_iterations and num_pipelines >= self.max_iterations: return False # check\",\n \"= self._results['pipeline_results'][pipeline_id] pipeline.describe() if pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines with\",\n \"make sure it is different from the current best pipeline before training and\",\n \"if pipeline_parameters is not None else {} self.search_iteration_plot = None self._interrupted = False\",\n \"to train. Returns: Dict[str, PipelineBase]: Dictionary keyed by pipeline name that maps to\",\n \"be a float, int, string or None. Received {type(max_time)} with value {str(max_time)}..\\\") if\",\n \"%d seconds.\\\\n\\\" % self.max_time) logger.info(\\\"Allowed model families: %s\\\\n\\\" % \\\", \\\".join([model.value for model\",\n \"(if set) time_in_loop = time.time() - start_of_loop self._start += time_in_loop return False else:\",\n \"allowed_pipelines (list(class)): A list of PipelineBase subclasses indicating the pipelines allowed in the\",\n \"_ = split_data(X_shape, self.y_train, problem_type=self.problem_type, test_size=_ensembling_split_size, random_seed=self.random_seed) self.ensembling_indices = ensembling_indices.to_dataframe()[0].tolist() self._engine = SequentialEngine(self.X_train,\",\n \"TimeSeriesBaselineBinaryPipeline}[self.problem_type] gap = self.problem_configuration['gap'] max_delay = self.problem_configuration['max_delay'] baseline = pipeline_class(parameters={\\\"pipeline\\\": {\\\"gap\\\": gap, \\\"max_delay\\\":\",\n \"SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from evalml.utils.logger import ( get_logger, log_subtitle, log_title,\",\n \"pipeline_id (int): pipeline to describe return_dict (bool): If True, return dictionary of information\",\n \"# Necessary for \\\"Plotting\\\" documentation, since Sphinx does not work well with instance\",\n \"automatically be encoded show_iteration_plot (boolean, True): Shows an iteration vs. score plot in\",\n \"when ensembling is True. Must be between 0 and 1, exclusive. Defaults to\",\n \"data splitter\\\") if not objective.is_defined_for_problem_type(self.problem_type): raise ValueError(\\\"Given objective {} is not compatible with\",\n \"from evalml.model_family import ModelFamily from evalml.objectives import ( get_core_objectives, get_non_core_objectives, get_objective ) from\",\n \"if additional_objectives is None: additional_objectives = get_core_objectives(self.problem_type) # if our main objective is\",\n \"return rankings_df @property def best_pipeline(self): \\\"\\\"\\\"Returns a trained instance of the best pipeline\",\n \"found during automl search. If `train_best_pipeline` is set to False, returns an untrained\",\n \"positive. If None, early stopping is disabled. Defaults to None. tolerance (float): Minimum\",\n \"a 2d array using the X_train argument') if y_train is None: raise ValueError('Must\",\n \"Shows an iteration vs. score plot in Jupyter notebook. Disabled by default in\",\n \"import ProblemTypes, handle_problem_types from evalml.tuners import SKOptTuner from evalml.utils import convert_to_seconds, infer_feature_types from\",\n \"f\\\"Allowed Pipelines: \\\\n{_print_list(self.allowed_pipelines or [])}\\\\n\\\" f\\\"Patience: {self.patience}\\\\n\\\" f\\\"Tolerance: {self.tolerance}\\\\n\\\" f\\\"Data Splitting: {self.data_splitter}\\\\n\\\" f\\\"Tuner:\",\n \"not compatible with a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or []:\",\n \"False): logger.info(all_objective_scores) if return_dict: return pipeline_results def add_to_rankings(self, pipeline): \\\"\\\"\\\"Fits and evaluates a\",\n \"self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed = random_seed\",\n \"positional parameters: A dictionary containing the training results for the new pipeline, an\",\n \"self.y_train, estimator, self.problem_type, custom_hyperparameters=self.pipeline_parameters) for estimator in allowed_estimators] if self.allowed_pipelines == []: raise\",\n \"up to %s pipelines. \\\" % self.max_iterations) if self.max_time is not None: logger.info(\\\"Will\",\n \"max_iterations elapsed = time.time() - self._start if self.max_time and elapsed >= self.max_time: return\",\n \"is not None: # Baseline + first batch + each pipeline iteration +\",\n \"o for o in [self.objective] + self.additional_objectives} if not isinstance(max_time, (int, float, str,\",\n \"if y_train is None: raise ValueError('Must specify training data target values as a\",\n \"max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds = len(cv_data) for\",\n \"allowed in the search. The default of None indicates all pipelines for this\",\n \"allowed pipeline class has been iterated over. If the number of unique pipelines\",\n \"y_train argument') try: self.problem_type = handle_problem_types(problem_type) except ValueError: raise ValueError('choose one of (binary,\",\n \"is not None: logger.info(f\\\"Searching up to {self.max_batches} batches for a total of {self.max_iterations}\",\n \"before search _baseline_cv_scores will be empty so we will return # nan for\",\n \"results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False): \\\"\\\"\\\"Describe a pipeline Arguments: pipeline_id\",\n \"`pipeline_results`: a dict with results from each pipeline, and `search_order`: a list describing\",\n \"pipelines to search over per batch is one, ensembling will not run. Defaults\",\n \"= self._check_for_high_variance(pipeline, cv_scores) pipeline_id = len(self._results['pipeline_results']) self._results['pipeline_results'][pipeline_id] = { \\\"id\\\": pipeline_id, \\\"pipeline_name\\\": pipeline.name,\",\n \"with open(file_path, 'rb') as f: return cloudpickle.load(f) def train_pipelines(self, pipelines): \\\"\\\"\\\"Train a list\",\n \"logger.info(\\\"Using default limit of max_batches=1.\\\\n\\\") if patience and (not isinstance(patience, int) or patience\",\n \"needed to configure the search. For example, in time series problems, values should\",\n \"of each pipeline family allowed in the search. \\\"\\\"\\\" if X_train is None:\",\n \"self.additional_objectives = additional_objectives self.objective_name_to_class = {o.name: o for o in [self.objective] + self.additional_objectives}\",\n \"self._searched = True def _find_best_pipeline(self): \\\"\\\"\\\"Finds the best pipeline in the rankings If\",\n \"has elapsed. If it is an integer, then the time will be in\",\n \"Baseline Estimator\\\": {\\\"gap\\\": gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores =\",\n \"iteration vs. score plot in Jupyter notebook. Disabled by default in non-Jupyter enviroments.\",\n \"Run at least one pipeline for every search num_pipelines = self._num_pipelines() if num_pipelines\",\n \"isinstance(max_time, (int, float)) and max_time < 0: raise ValueError(f\\\"Parameter max_time must be None\",\n \"True, but the number of unique pipelines is one, so ensembling will not\",\n \"self.objective_name_to_class) if is_baseline: self._baseline_cv_scores = mean_cv_all_objectives for obj_name in mean_cv_all_objectives: objective_class = self.objective_name_to_class[obj_name]\",\n \"abs((curr_score - best_score) / best_score) > self.tolerance score_improved = curr_score > best_score if\",\n \"parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is None: raise PipelineNotFoundError(\\\"Pipeline\",\n \"c in [\\\"# Training\\\", \\\"# Validation\\\"]: all_objective_scores[c] = all_objective_scores[c].astype(\\\"object\\\") continue mean = all_objective_scores[c].mean(axis=0)\",\n \"with results from each pipeline, and `search_order`: a list describing the order the\",\n \"not self._best_pipeline: raise PipelineNotFoundError(\\\"automl search must be run before selecting `best_pipeline`.\\\") return self._best_pipeline\",\n \"are allowed. Setting this field will cause allowed_model_families to be ignored. allowed_model_families (list(str,\",\n \"to stop search early. Must be positive. If None, early stopping is disabled.\",\n \"pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns: PipelineBase: untrained\",\n \"The default of None searches over all model families. Run evalml.pipelines.components.utils.allowed_model_families(\\\"binary\\\") to see\",\n \") from evalml.pipelines import ( MeanBaselineRegressionPipeline, ModeBaselineBinaryPipeline, ModeBaselineMulticlassPipeline, TimeSeriesBaselineBinaryPipeline, TimeSeriesBaselineMulticlassPipeline, TimeSeriesBaselineRegressionPipeline ) from\",\n \"in obj_list]) return '\\\\n'.join(lines) def _get_funct_name(function): if callable(function): return function.__name__ else: return None\",\n \"\\\").strip().lower() if choice == \\\"y\\\": logger.info(\\\"Exiting AutoMLSearch.\\\") return True elif choice == \\\"n\\\":\",\n \"isinstance(patience, int) or patience < 0): raise ValueError(\\\"patience value must be a positive\",\n \"[n_samples, n_features]. Required. y_train (pd.Series, ww.DataColumn): The target training data of length [n_samples].\",\n \"categorical. Categorical features will automatically be encoded show_iteration_plot (boolean, True): Shows an iteration\",\n \"pipeline.model_family == ModelFamily.ENSEMBLE: logger.info(\\\"Input for ensembler are pipelines with IDs: \\\" + str(pipeline_results['input_pipeline_ids']))\",\n \"CV): %.1f seconds\\\" % pipeline_results[\\\"training_time\\\"]) log_subtitle(logger, \\\"Cross Validation\\\", underline=\\\"-\\\") all_objective_scores = [fold[\\\"all_objective_scores\\\"] for\",\n \"Received {max_batches}.\\\") if max_iterations is not None and max_iterations < 0: raise ValueError(f\\\"Parameter\",\n \"shuffle=True, random_seed=self.random_seed) self.data_splitter = self.data_splitter or default_data_splitter self.pipeline_parameters = pipeline_parameters if pipeline_parameters is\",\n \"the highest-scoring set of parameters used with each pipeline.\\\"\\\"\\\" return self.full_rankings.drop_duplicates(subset=\\\"pipeline_name\\\", keep=\\\"first\\\") @property\",\n \"training data as a 2d array using the X_train argument') if y_train is\",\n \"pipeline_class = pipeline_results.get('pipeline_class') parameters = pipeline_results.get('parameters') if pipeline_class is None or parameters is\",\n \"0.0 self._results = { 'pipeline_results': {}, 'search_order': [], 'errors': [] } self.random_seed =\",\n \"list(set([p.model_family for p in (self.allowed_pipelines)])) logger.debug(f\\\"allowed_pipelines set to {[pipeline.name for pipeline in self.allowed_pipelines]}\\\")\",\n \"three positional parameters: the Exception raised, the traceback, and the AutoMLSearch object. Must\",\n \"is not compatible with a {} problem.\\\".format(obj.name, self.problem_type.value)) for pipeline in self.allowed_pipelines or\",\n \"pipelines to search. Parameters max_time, and max_iterations have precedence over stopping the search.\",\n \"see options. Change `binary` to `multiclass` or `regression` depending on the problem type.\",\n \"f\\\"{pipeline.name}\\\" if len(desc) > AutoMLSearch._MAX_NAME_LEN: desc = desc[:AutoMLSearch._MAX_NAME_LEN - 3] + \\\"...\\\" desc\",\n \"on the same instance. Re-initialize AutoMLSearch to search again.\\\") return # don't show\",\n \"parameters not found in automl results\\\") return pipeline_class(parameters, random_seed=self.random_seed) def describe_pipeline(self, pipeline_id, return_dict=False):\",\n \"to stop the search. Returns: bool: If True, search should terminate early \\\"\\\"\\\"\",\n \"gap, \\\"max_delay\\\": max_delay}}) self._engine.evaluate_batch([baseline]) @staticmethod def _get_mean_cv_scores_for_all_objectives(cv_data, objective_name_to_class): scores = defaultdict(int) n_folds =\",\n \"This can be helpful for training pipelines once the search is complete. Arguments:\",\n \"train that pipeline during automl search. Arguments: pipeline_id (int): pipeline to retrieve Returns:\",\n \"= \\\"\\\" def search(self, show_iteration_plot=True): \\\"\\\"\\\"Find the best pipeline for the data set.\",\n \"instance of the specified pipeline initialized with the parameters used to train that\",\n \"but max_iterations is too small, so ensembling will not run. Set max_iterations >=\",\n \"log_error_callback from evalml.automl.engine import SequentialEngine from evalml.automl.utils import ( check_all_pipeline_names_unique, get_default_primary_search_objective, make_data_splitter )\",\n \"to configure the search. For example, in time series problems, values should be\",\n \"\\\"\\\"\\\"Returns a trained instance of the best pipeline and parameters found during automl\",\n \"of parallelism used for pipelines. None and 1 are equivalent. If set to\",\n \"logger.warning(f\\\"Ensembling is set to True, but max_batches is too small, so ensembling will\",\n \"perform as estimated on unseen data.\\\") return high_variance_cv def get_pipeline(self, pipeline_id): \\\"\\\"\\\"Given the\",\n \"Exception. Callback function takes three positional parameters: the Exception raised, the traceback, and\"\n]","isConversational":false}}},{"rowIdx":3597,"cells":{"ngram":{"kind":"list like","value":["graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider =","= graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root,","def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin,","SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \"\"\"Social Auth for JSON Web","provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info,","from graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators import social_auth","from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract","Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root,","graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info,","Meta: abstract = True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod","class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \"\"\"Social Auth for JSON","import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType)","class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments:","graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider = graphene.String(required=True) code =","info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth","abstract = True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie","graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs):","SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \"\"\"Social Auth for JSON Web Token",".decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract =","social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class","= graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social,","root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social","import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True","social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \"\"\"Social Auth","@setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs)","cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation):","from . import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social","import setup_jwt_cookie from . import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin,","SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider","\"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \"\"\"Social Auth for JSON Web Token (JWT)\"\"\"","True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def","class Meta: abstract = True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True)","@social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class","mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation):","return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin,","code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return","= graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider = graphene.String(required=True) code","mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class","= True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth","class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls,","graphene from graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators import","@classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social,","types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta:","social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider = graphene.String(required=True)","**kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \"\"\"Social Auth for","import graphene from graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators",". import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social =","setup_jwt_cookie from . import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation):","social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\"","**kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class","graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators import social_auth class","info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \"\"\"Social Auth Mutation\"\"\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \"\"\"Social"],"string":"[\n \"graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider =\",\n \"= graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root,\",\n \"def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin,\",\n \"SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth for JSON Web\",\n \"provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info,\",\n \"from graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators import social_auth\",\n \"from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract\",\n \"Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root,\",\n \"graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info,\",\n \"Meta: abstract = True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod\",\n \"class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth for JSON\",\n \"import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType)\",\n \"class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments:\",\n \"graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider = graphene.String(required=True) code =\",\n \"info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth\",\n \"abstract = True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie\",\n \"graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs):\",\n \"SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth for JSON Web Token\",\n \".decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract =\",\n \"social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class\",\n \"= graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social,\",\n \"root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social\",\n \"import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True\",\n \"social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth\",\n \"@setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs)\",\n \"cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation):\",\n \"from . import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social\",\n \"import setup_jwt_cookie from . import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin,\",\n \"SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider\",\n \"\\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth for JSON Web Token (JWT)\\\"\\\"\\\"\",\n \"True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def\",\n \"class Meta: abstract = True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True)\",\n \"@social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class\",\n \"mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation):\",\n \"return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin,\",\n \"code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return\",\n \"= graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider = graphene.String(required=True) code\",\n \"mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class\",\n \"= True class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth\",\n \"class Arguments: provider = graphene.String(required=True) code = graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls,\",\n \"graphene from graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators import\",\n \"@classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root, info, social,\",\n \"types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social = graphene.Field(types.SocialType) class Meta:\",\n \"social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider = graphene.String(required=True)\",\n \"**kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth for\",\n \"import graphene from graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators\",\n \". import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation): social =\",\n \"setup_jwt_cookie from . import mixins, types from .decorators import social_auth class SocialAuthMutation(mixins.SocialAuthMixin, graphene.Mutation):\",\n \"social, **kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\"\",\n \"**kwargs): return cls.resolve(root, info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class\",\n \"graphql_jwt.decorators import setup_jwt_cookie from . import mixins, types from .decorators import social_auth class\",\n \"info, social, **kwargs) class SocialAuth(mixins.ResolveMixin, SocialAuthMutation): \\\"\\\"\\\"Social Auth Mutation\\\"\\\"\\\" class SocialAuthJWT(mixins.JSONWebTokenMixin, SocialAuthMutation): \\\"\\\"\\\"Social\"\n]","isConversational":false}}},{"rowIdx":3598,"cells":{"ngram":{"kind":"list like","value":["yellowbrick.regressor.base # Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> #","Base classes for regressor Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## from ..utils import","regressor, otherwise it raises a ``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None, **kwargs): if","class is to perform a check to ensure the passed in estimator is","a check to ensure the passed in estimator is a regressor, otherwise it","functionality of this class is to perform a check to ensure the passed","from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ##","# Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 #","Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all ScoreVisualizers that evaluate","classes for regressor Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor","## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all","<<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700","information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \"\"\" Base classes","regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created:","not isregressor(model): raise YellowbrickTypeError( \"This estimator is not a regressor; try a classifier","## Packages for export __all__ = [ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization","Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME>","# ID: base.py [7d3f5e6] <EMAIL> $ \"\"\" Base classes for regressor Visualizers. \"\"\"","Base class for all ScoreVisualizers that evaluate a regression estimator. The primary functionality","########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from","classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>>","# # Copyright (C) 2016 District Data Labs # For license information, see","from ..exceptions import YellowbrickTypeError ## Packages for export __all__ = [ \"RegressionScoreVisualizer\", ]","# Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016","# Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author:","# yellowbrick.regressor.base # Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>>","a regressor, otherwise it raises a ``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None, **kwargs):","[ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\"","to ensure the passed in estimator is a regressor, otherwise it raises a","estimator is not a regressor; try a classifier or \" \"clustering score visualizer","__init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \"This estimator is not","########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all ScoreVisualizers that evaluate a regression","ScoreVisualizers that evaluate a regression estimator. The primary functionality of this class is","Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016 District Data","10:30:36 2016 -0700 # # Copyright (C) 2016 District Data Labs # For","[7d3f5e6] <EMAIL> $ \"\"\" Base classes for regressor Visualizers. \"\"\" ########################################################################## ## Imports","check to ensure the passed in estimator is a regressor, otherwise it raises","..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages","for regressor Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from","not a regressor; try a classifier or \" \"clustering score visualizer instead!\" )","see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \"\"\" Base classes for","## Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions","ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \"This estimator is not a regressor;","for all ScoreVisualizers that evaluate a regression estimator. The primary functionality of this","isregressor(model): raise YellowbrickTypeError( \"This estimator is not a regressor; try a classifier or","try a classifier or \" \"clustering score visualizer instead!\" ) super(RegressionScoreVisualizer, self).__init__(model, ax=ax,","########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError","# # ID: base.py [7d3f5e6] <EMAIL> $ \"\"\" Base classes for regressor Visualizers.","\"\"\" Base class for all ScoreVisualizers that evaluate a regression estimator. The primary","class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all ScoreVisualizers that evaluate a regression estimator.","LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \"\"\" Base classes for regressor","########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for","Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all ScoreVisualizers","of this class is to perform a check to ensure the passed in","estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \"\"\" def __init__(self, model,","this class is to perform a check to ensure the passed in estimator","Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base import","a classifier or \" \"clustering score visualizer instead!\" ) super(RegressionScoreVisualizer, self).__init__(model, ax=ax, **kwargs)","license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \"\"\" Base","regressor Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base","\"\"\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer","Packages for export __all__ = [ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization Base","\"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base","if not isregressor(model): raise YellowbrickTypeError( \"This estimator is not a regressor; try a","raise YellowbrickTypeError( \"This estimator is not a regressor; try a classifier or \"","] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class","__all__ = [ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class","otherwise it raises a ``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None, **kwargs): if not","$ \"\"\" Base classes for regressor Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## from","# Copyright (C) 2016 District Data Labs # For license information, see LICENSE.txt","= [ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer):","primary functionality of this class is to perform a check to ensure the","all ScoreVisualizers that evaluate a regression estimator. The primary functionality of this class","<EMAIL> $ \"\"\" Base classes for regressor Visualizers. \"\"\" ########################################################################## ## Imports ##########################################################################","model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \"This estimator is not a","Data Labs # For license information, see LICENSE.txt # # ID: base.py [7d3f5e6]","class for all ScoreVisualizers that evaluate a regression estimator. The primary functionality of","The primary functionality of this class is to perform a check to ensure","Labs # For license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL>","03 10:30:36 2016 -0700 # # Copyright (C) 2016 District Data Labs #","ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__ = [ \"RegressionScoreVisualizer\",","Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all ScoreVisualizers that","in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \"\"\" def __init__(self,","for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> #","a regressor; try a classifier or \" \"clustering score visualizer instead!\" ) super(RegressionScoreVisualizer,","is a regressor, otherwise it raises a ``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None,","Copyright (C) 2016 District Data Labs # For license information, see LICENSE.txt #","Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36","ID: base.py [7d3f5e6] <EMAIL> $ \"\"\" Base classes for regressor Visualizers. \"\"\" ##########################################################################","import YellowbrickTypeError ## Packages for export __all__ = [ \"RegressionScoreVisualizer\", ] ########################################################################## ##","base.py [7d3f5e6] <EMAIL> $ \"\"\" Base classes for regressor Visualizers. \"\"\" ########################################################################## ##","import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__ = [","<NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016","regressor; try a classifier or \" \"clustering score visualizer instead!\" ) super(RegressionScoreVisualizer, self).__init__(model,","import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for","2016 -0700 # # Copyright (C) 2016 District Data Labs # For license","RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all ScoreVisualizers that evaluate a regression estimator. The","def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \"This estimator is","<<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C)","it raises a ``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model):","the passed in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \"\"\"","regression estimator. The primary functionality of this class is to perform a check","is to perform a check to ensure the passed in estimator is a","ensure the passed in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``.","is not a regressor; try a classifier or \" \"clustering score visualizer instead!\"","(C) 2016 District Data Labs # For license information, see LICENSE.txt # #","to perform a check to ensure the passed in estimator is a regressor,","``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \"This","# Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03","For license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \"\"\"","\"This estimator is not a regressor; try a classifier or \" \"clustering score","..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__ =","raises a ``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise","YellowbrickTypeError ## Packages for export __all__ = [ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression","..exceptions import YellowbrickTypeError ## Packages for export __all__ = [ \"RegressionScoreVisualizer\", ] ##########################################################################","passed in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \"\"\" def","2016 District Data Labs # For license information, see LICENSE.txt # # ID:","-0700 # # Copyright (C) 2016 District Data Labs # For license information,","export __all__ = [ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization Base Object ##########################################################################","<NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright","District Data Labs # For license information, see LICENSE.txt # # ID: base.py","Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016 District","<reponame>Juan0001/yellowbrick-docs-zh # yellowbrick.regressor.base # Base classes for regressor Visualizers. # # Author: <NAME>","\"\"\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \"This estimator","Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri","isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export","Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import","# # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun","Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016 District Data Labs","# For license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $","from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__","perform a check to ensure the passed in estimator is a regressor, otherwise","evaluate a regression estimator. The primary functionality of this class is to perform","estimator. The primary functionality of this class is to perform a check to","**kwargs): if not isregressor(model): raise YellowbrickTypeError( \"This estimator is not a regressor; try","\"\"\" Base classes for regressor Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## from ..utils","a ``YellowbrickTypeError``. \"\"\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError(","Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 # #","for export __all__ = [ \"RegressionScoreVisualizer\", ] ########################################################################## ## Regression Visualization Base Object","a regression estimator. The primary functionality of this class is to perform a","that evaluate a regression estimator. The primary functionality of this class is to","YellowbrickTypeError( \"This estimator is not a regressor; try a classifier or \" \"clustering","Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \"\"\" Base class for all ScoreVisualizers that evaluate a"],"string":"[\n \"yellowbrick.regressor.base # Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> #\",\n \"Base classes for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from ..utils import\",\n \"regressor, otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs): if\",\n \"class is to perform a check to ensure the passed in estimator is\",\n \"a check to ensure the passed in estimator is a regressor, otherwise it\",\n \"functionality of this class is to perform a check to ensure the passed\",\n \"from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ##\",\n \"# Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 #\",\n \"Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all ScoreVisualizers that evaluate\",\n \"classes for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor\",\n \"## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all\",\n \"<<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700\",\n \"information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes\",\n \"regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created:\",\n \"not isregressor(model): raise YellowbrickTypeError( \\\"This estimator is not a regressor; try a classifier\",\n \"## Packages for export __all__ = [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization\",\n \"Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME>\",\n \"# ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes for regressor Visualizers. \\\"\\\"\\\"\",\n \"Base class for all ScoreVisualizers that evaluate a regression estimator. The primary functionality\",\n \"########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from\",\n \"classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>>\",\n \"# # Copyright (C) 2016 District Data Labs # For license information, see\",\n \"from ..exceptions import YellowbrickTypeError ## Packages for export __all__ = [ \\\"RegressionScoreVisualizer\\\", ]\",\n \"# Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016\",\n \"# Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author:\",\n \"# yellowbrick.regressor.base # Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>>\",\n \"a regressor, otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs):\",\n \"[ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\"\",\n \"to ensure the passed in estimator is a regressor, otherwise it raises a\",\n \"estimator is not a regressor; try a classifier or \\\" \\\"clustering score visualizer\",\n \"__init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \\\"This estimator is not\",\n \"########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all ScoreVisualizers that evaluate a regression\",\n \"ScoreVisualizers that evaluate a regression estimator. The primary functionality of this class is\",\n \"Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016 District Data\",\n \"10:30:36 2016 -0700 # # Copyright (C) 2016 District Data Labs # For\",\n \"[7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports\",\n \"check to ensure the passed in estimator is a regressor, otherwise it raises\",\n \"..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages\",\n \"for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from\",\n \"not a regressor; try a classifier or \\\" \\\"clustering score visualizer instead!\\\" )\",\n \"see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes for\",\n \"## Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions\",\n \"ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \\\"This estimator is not a regressor;\",\n \"for all ScoreVisualizers that evaluate a regression estimator. The primary functionality of this\",\n \"isregressor(model): raise YellowbrickTypeError( \\\"This estimator is not a regressor; try a classifier or\",\n \"try a classifier or \\\" \\\"clustering score visualizer instead!\\\" ) super(RegressionScoreVisualizer, self).__init__(model, ax=ax,\",\n \"########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError\",\n \"# # ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes for regressor Visualizers.\",\n \"\\\"\\\"\\\" Base class for all ScoreVisualizers that evaluate a regression estimator. The primary\",\n \"class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all ScoreVisualizers that evaluate a regression estimator.\",\n \"LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes for regressor\",\n \"########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for\",\n \"Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all ScoreVisualizers\",\n \"of this class is to perform a check to ensure the passed in\",\n \"estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model,\",\n \"this class is to perform a check to ensure the passed in estimator\",\n \"Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base import\",\n \"a classifier or \\\" \\\"clustering score visualizer instead!\\\" ) super(RegressionScoreVisualizer, self).__init__(model, ax=ax, **kwargs)\",\n \"license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base\",\n \"regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base\",\n \"\\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer\",\n \"Packages for export __all__ = [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization Base\",\n \"\\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base\",\n \"if not isregressor(model): raise YellowbrickTypeError( \\\"This estimator is not a regressor; try a\",\n \"raise YellowbrickTypeError( \\\"This estimator is not a regressor; try a classifier or \\\"\",\n \"] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class\",\n \"__all__ = [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class\",\n \"otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs): if not\",\n \"$ \\\"\\\"\\\" Base classes for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from\",\n \"# Copyright (C) 2016 District Data Labs # For license information, see LICENSE.txt\",\n \"= [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer):\",\n \"primary functionality of this class is to perform a check to ensure the\",\n \"all ScoreVisualizers that evaluate a regression estimator. The primary functionality of this class\",\n \"<EMAIL> $ \\\"\\\"\\\" Base classes for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ##########################################################################\",\n \"model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \\\"This estimator is not a\",\n \"Data Labs # For license information, see LICENSE.txt # # ID: base.py [7d3f5e6]\",\n \"class for all ScoreVisualizers that evaluate a regression estimator. The primary functionality of\",\n \"The primary functionality of this class is to perform a check to ensure\",\n \"Labs # For license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL>\",\n \"03 10:30:36 2016 -0700 # # Copyright (C) 2016 District Data Labs #\",\n \"ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__ = [ \\\"RegressionScoreVisualizer\\\",\",\n \"Visualization Base Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all ScoreVisualizers that\",\n \"in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self,\",\n \"for regressor Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> #\",\n \"a regressor; try a classifier or \\\" \\\"clustering score visualizer instead!\\\" ) super(RegressionScoreVisualizer,\",\n \"is a regressor, otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None,\",\n \"Copyright (C) 2016 District Data Labs # For license information, see LICENSE.txt #\",\n \"Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36\",\n \"ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes for regressor Visualizers. \\\"\\\"\\\" ##########################################################################\",\n \"import YellowbrickTypeError ## Packages for export __all__ = [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ##\",\n \"base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\" Base classes for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ##\",\n \"import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__ = [\",\n \"<NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016\",\n \"regressor; try a classifier or \\\" \\\"clustering score visualizer instead!\\\" ) super(RegressionScoreVisualizer, self).__init__(model,\",\n \"import isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for\",\n \"2016 -0700 # # Copyright (C) 2016 District Data Labs # For license\",\n \"RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all ScoreVisualizers that evaluate a regression estimator. The\",\n \"def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \\\"This estimator is\",\n \"<<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C)\",\n \"it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model):\",\n \"the passed in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\"\",\n \"regression estimator. The primary functionality of this class is to perform a check\",\n \"is to perform a check to ensure the passed in estimator is a\",\n \"ensure the passed in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``.\",\n \"is not a regressor; try a classifier or \\\" \\\"clustering score visualizer instead!\\\"\",\n \"(C) 2016 District Data Labs # For license information, see LICENSE.txt # #\",\n \"to perform a check to ensure the passed in estimator is a regressor,\",\n \"``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \\\"This\",\n \"# Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun 03\",\n \"For license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $ \\\"\\\"\\\"\",\n \"\\\"This estimator is not a regressor; try a classifier or \\\" \\\"clustering score\",\n \"..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__ =\",\n \"raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise\",\n \"YellowbrickTypeError ## Packages for export __all__ = [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression\",\n \"..exceptions import YellowbrickTypeError ## Packages for export __all__ = [ \\\"RegressionScoreVisualizer\\\", ] ##########################################################################\",\n \"passed in estimator is a regressor, otherwise it raises a ``YellowbrickTypeError``. \\\"\\\"\\\" def\",\n \"2016 District Data Labs # For license information, see LICENSE.txt # # ID:\",\n \"-0700 # # Copyright (C) 2016 District Data Labs # For license information,\",\n \"export __all__ = [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization Base Object ##########################################################################\",\n \"<NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright\",\n \"District Data Labs # For license information, see LICENSE.txt # # ID: base.py\",\n \"Created: Fri Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016 District\",\n \"<reponame>Juan0001/yellowbrick-docs-zh # yellowbrick.regressor.base # Base classes for regressor Visualizers. # # Author: <NAME>\",\n \"\\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError( \\\"This estimator\",\n \"Visualizers. # # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri\",\n \"isregressor from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export\",\n \"Imports ########################################################################## from ..utils import isregressor from ..base import ScoreVisualizer from ..exceptions import\",\n \"# # Author: <NAME> <<EMAIL>> # Author: <NAME> <<EMAIL>> # Created: Fri Jun\",\n \"Jun 03 10:30:36 2016 -0700 # # Copyright (C) 2016 District Data Labs\",\n \"# For license information, see LICENSE.txt # # ID: base.py [7d3f5e6] <EMAIL> $\",\n \"from ..base import ScoreVisualizer from ..exceptions import YellowbrickTypeError ## Packages for export __all__\",\n \"perform a check to ensure the passed in estimator is a regressor, otherwise\",\n \"evaluate a regression estimator. The primary functionality of this class is to perform\",\n \"estimator. The primary functionality of this class is to perform a check to\",\n \"**kwargs): if not isregressor(model): raise YellowbrickTypeError( \\\"This estimator is not a regressor; try\",\n \"\\\"\\\"\\\" Base classes for regressor Visualizers. \\\"\\\"\\\" ########################################################################## ## Imports ########################################################################## from ..utils\",\n \"a ``YellowbrickTypeError``. \\\"\\\"\\\" def __init__(self, model, ax=None, **kwargs): if not isregressor(model): raise YellowbrickTypeError(\",\n \"Author: <NAME> <<EMAIL>> # Created: Fri Jun 03 10:30:36 2016 -0700 # #\",\n \"for export __all__ = [ \\\"RegressionScoreVisualizer\\\", ] ########################################################################## ## Regression Visualization Base Object\",\n \"a regression estimator. The primary functionality of this class is to perform a\",\n \"that evaluate a regression estimator. The primary functionality of this class is to\",\n \"YellowbrickTypeError( \\\"This estimator is not a regressor; try a classifier or \\\" \\\"clustering\",\n \"Object ########################################################################## class RegressionScoreVisualizer(ScoreVisualizer): \\\"\\\"\\\" Base class for all ScoreVisualizers that evaluate a\"\n]","isConversational":false}}},{"rowIdx":3599,"cells":{"ngram":{"kind":"list like","value":["[objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs)","image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files')","rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files,","= resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt,","isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage()","crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks","import logging import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import","'--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of","objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write')","IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines","imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName","type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int,","img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() ''' Main driver. '''","''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two","objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(),","''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master',","lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj","= cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave +","components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command","text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2","two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str,","resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt)","IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp,","parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None):","imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are","as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage()","dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks',","swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True,","createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right","type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser =","field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s',","objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth","imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp","= isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul =","imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth","dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser","= crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross =","parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str,","img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2,","import os import argparse import logging import isce import isceobj from components.stdproc.stdproc import","description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master","return parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1,","objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1,","objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def","img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks)","objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp","type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o',","parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 =","isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1,","img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ ==","parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range","run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2","= isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp =","min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName + '.int' objInt","objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul","and 2 are just text files, need to open them as image IU.copyAttributes(imageSlc1,","import ImageUtil as IU def createParser(): ''' Command Line Parser. ''' parser =","imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown","= isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True)","main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage()","objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1,","type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return","parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2,","/ rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt =","parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a',","isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length =","objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp)","isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage()","= createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage()","#!/usr/bin/env python3 import os import argparse import logging import isce import isceobj from","slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt,","= isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length","imageSlc1 and 2 are just text files, need to open them as image","Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel","dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int","objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage()","Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave',","'--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks')","as IU def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate","2 are just text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1)","from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line Parser. '''","objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in","isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files, need to open","createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between","default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser = createParser() return","now imageSlc1 and 2 are just text files, need to open them as","output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r',","isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul()","isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU","help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix',","logging import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil","imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml')","imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2","isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks)","+ '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt","return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now","objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 =","parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1","azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp,","resampAmp = resampName + '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage()","objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth()","and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int,","objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks","crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line Parser.","#right now imageSlc1 and 2 are just text files, need to open them","are just text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read')","objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1","exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() ''' Main","objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp)","objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength())","imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp)","default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1,","just text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage()","cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks,","None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1","def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 =","required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and","= None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks):","required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str,","amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks',","open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2)","objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width =","ImageUtil as IU def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser(","looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs =","def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and","Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m',","of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks')","image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul',","parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix","files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1,","def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName,","isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__","= min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName + '.int'","def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field","need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage()","+ '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write')","objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage()","objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files, need","image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth","argparse import logging import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil","import argparse import logging import isce import isceobj from components.stdproc.stdproc import crossmul from","+ '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix,","rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage()","objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp","= imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp =","from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): '''","lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName +","return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master +","os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() '''","img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix),","default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def","import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line","help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp","offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image')","= azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt,","type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output","objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross","'.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage()","argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True,","them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read')","objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width","= resampName + '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt)","'.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks,","for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None):","slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp","import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def","IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines","= int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp'","import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as","= slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2,","os import argparse import logging import isce import isceobj from components.stdproc.stdproc import crossmul","objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth =","'.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main()","'--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks',","resampName + '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth)","python3 import os import argparse import logging import isce import isceobj from components.stdproc.stdproc","+ '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__':","objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth","iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line Parser. ''' parser","run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() ''' Main driver.","objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth)","azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just text","files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 =","objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines =","'--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image')","objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return","intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName +","= lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for","objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt,","'--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser","resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just","obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master","dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir',","IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth =","help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth","parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave","cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml')","between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave',","IU def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset","objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown =","'.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt =","= isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp =","resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage()","= rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]:","help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs","to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2,","in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps =","= isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth /","imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName + '.int' objInt =","int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt","dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser = createParser()","isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser():","= isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files, need to","= argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master',","obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps","rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName","= isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if","int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks',","Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths')","help='Range looks') return parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs)","looks') return parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def","inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave"],"string":"[\n \"[objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs)\",\n \"image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files')\",\n \"rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files,\",\n \"= resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt,\",\n \"isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage()\",\n \"crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks\",\n \"import logging import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import\",\n \"'--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of\",\n \"objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write')\",\n \"IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines\",\n \"imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName\",\n \"type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int,\",\n \"img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() ''' Main driver. '''\",\n \"''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two\",\n \"objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(),\",\n \"''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master',\",\n \"lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj\",\n \"= cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave +\",\n \"components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command\",\n \"text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2\",\n \"two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str,\",\n \"resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt)\",\n \"IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp,\",\n \"parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None):\",\n \"imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are\",\n \"as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage()\",\n \"dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks',\",\n \"swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True,\",\n \"createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right\",\n \"type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser =\",\n \"field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s',\",\n \"objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth\",\n \"imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp\",\n \"= isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul =\",\n \"imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth\",\n \"dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser\",\n \"= crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross =\",\n \"parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str,\",\n \"img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2,\",\n \"import os import argparse import logging import isce import isceobj from components.stdproc.stdproc import\",\n \"description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master\",\n \"return parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1,\",\n \"objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1,\",\n \"objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def\",\n \"img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks)\",\n \"objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp\",\n \"type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o',\",\n \"parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 =\",\n \"isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1,\",\n \"img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ ==\",\n \"parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range\",\n \"run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2\",\n \"= isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp =\",\n \"min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName + '.int' objInt\",\n \"objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul\",\n \"and 2 are just text files, need to open them as image IU.copyAttributes(imageSlc1,\",\n \"import ImageUtil as IU def createParser(): ''' Command Line Parser. ''' parser =\",\n \"imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown\",\n \"= isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True)\",\n \"main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage()\",\n \"objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1,\",\n \"type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return\",\n \"parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2,\",\n \"/ rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt =\",\n \"parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a',\",\n \"isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length =\",\n \"objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp)\",\n \"isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage()\",\n \"= createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage()\",\n \"#!/usr/bin/env python3 import os import argparse import logging import isce import isceobj from\",\n \"slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt,\",\n \"= isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length\",\n \"imageSlc1 and 2 are just text files, need to open them as image\",\n \"Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel\",\n \"dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int\",\n \"objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage()\",\n \"Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave',\",\n \"'--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks')\",\n \"as IU def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate\",\n \"2 are just text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1)\",\n \"from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line Parser. '''\",\n \"objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in\",\n \"isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files, need to open\",\n \"createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between\",\n \"default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser = createParser() return\",\n \"now imageSlc1 and 2 are just text files, need to open them as\",\n \"output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r',\",\n \"isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul()\",\n \"isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU\",\n \"help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix',\",\n \"logging import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil\",\n \"imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml')\",\n \"imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2\",\n \"isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks)\",\n \"+ '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt\",\n \"return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now\",\n \"objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 =\",\n \"parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1\",\n \"azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp,\",\n \"resampAmp = resampName + '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage()\",\n \"objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth()\",\n \"and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int,\",\n \"objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks\",\n \"crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line Parser.\",\n \"#right now imageSlc1 and 2 are just text files, need to open them\",\n \"are just text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read')\",\n \"objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1\",\n \"exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() ''' Main\",\n \"objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp)\",\n \"objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength())\",\n \"imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp)\",\n \"default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1,\",\n \"just text files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage()\",\n \"cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks,\",\n \"None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1\",\n \"def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 =\",\n \"required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and\",\n \"= None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks):\",\n \"required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str,\",\n \"amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks',\",\n \"open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2)\",\n \"objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width =\",\n \"ImageUtil as IU def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser(\",\n \"looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs =\",\n \"def run(imageSlc1, imageSlc2, resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and\",\n \"Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m',\",\n \"of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks')\",\n \"image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul',\",\n \"parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix\",\n \"files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1,\",\n \"def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def run(imageSlc1, imageSlc2, resampName,\",\n \"isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__\",\n \"= min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName + '.int'\",\n \"def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field\",\n \"need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage()\",\n \"+ '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write')\",\n \"objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage()\",\n \"objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files, need\",\n \"image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth\",\n \"argparse import logging import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil\",\n \"import argparse import logging import isce import isceobj from components.stdproc.stdproc import crossmul from\",\n \"+ '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix,\",\n \"rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage()\",\n \"objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp\",\n \"= imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp =\",\n \"from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): '''\",\n \"lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName +\",\n \"return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master +\",\n \"os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() '''\",\n \"img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix),\",\n \"default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def\",\n \"import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line\",\n \"help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp\",\n \"offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image')\",\n \"= azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt,\",\n \"type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output\",\n \"objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross\",\n \"'.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage()\",\n \"argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True,\",\n \"them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read')\",\n \"objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp = isceobj.createAmpImage() IU.copyAttributes(objAmp, imageAmp) objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width\",\n \"= resampName + '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt)\",\n \"'.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks,\",\n \"for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None):\",\n \"slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp\",\n \"import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def\",\n \"IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines\",\n \"= int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp'\",\n \"import isce import isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as\",\n \"= slcWidth objCrossmul.length = lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2,\",\n \"os import argparse import logging import isce import isceobj from components.stdproc.stdproc import crossmul\",\n \"objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth =\",\n \"'.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main()\",\n \"'--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files') parser.add_argument('-a', '--alks',\",\n \"resampName + '.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth)\",\n \"python3 import os import argparse import logging import isce import isceobj from components.stdproc.stdproc\",\n \"+ '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__':\",\n \"objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth\",\n \"iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser(): ''' Command Line Parser. ''' parser\",\n \"run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if __name__ == '__main__': main() ''' Main driver.\",\n \"objInt.setWidth(intWidth) imageInt = isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth)\",\n \"azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just text\",\n \"files, need to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 =\",\n \"objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth / rgLooks) lines =\",\n \"'--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image')\",\n \"objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return\",\n \"intWidth = int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName +\",\n \"= lines objCrossmul.LooksDown = azLooks objCrossmul.LooksAcross = rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for\",\n \"objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt,\",\n \"'--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser\",\n \"resampName, azLooks, rgLooks): objSlc1 = isceobj.createSlcImage() #right now imageSlc1 and 2 are just\",\n \"obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master\",\n \"dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave image') parser.add_argument('-o', '--outdir',\",\n \"IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth =\",\n \"help='Prefix of output int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth\",\n \"parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave', type=str, dest='slave', required=True, help='Slave\",\n \"cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave + '.xml')\",\n \"between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master', required=True, help='Master image') parser.add_argument('-s', '--slave',\",\n \"IU def createParser(): ''' Command Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset\",\n \"objAmp.setAccessMode('write') objAmp.createImage() objCrossmul = crossmul.createcrossmul() objCrossmul.width = slcWidth objCrossmul.length = lines objCrossmul.LooksDown =\",\n \"'.amp' resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt =\",\n \"= isceobj.createIntImage() IU.copyAttributes(objInt, imageInt) objInt.setAccessMode('write') objInt.createImage() objAmp = isceobj.createAmpImage() objAmp.setFilename(resampAmp) objAmp.setWidth(intWidth) imageAmp =\",\n \"resampInt = resampName + '.int' objInt = isceobj.createIntImage() objInt.setFilename(resampInt) objInt.setWidth(intWidth) imageInt = isceobj.createIntImage()\",\n \"= rgLooks objCrossmul.crossmul(objSlc1, objSlc2, objInt, objAmp) for obj in [objInt, objAmp, objSlc1, objSlc2]:\",\n \"help='Azimuth looks') parser.add_argument('-r', '--rlks', type=int, dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs\",\n \"to open them as image IU.copyAttributes(imageSlc1, objSlc1) objSlc1.setAccessMode('read') objSlc1.createImage() objSlc2 = isceobj.createSlcImage() IU.copyAttributes(imageSlc2,\",\n \"in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps =\",\n \"= isceobj.createSlcImage() IU.copyAttributes(imageSlc2, objSlc2) objSlc2.setAccessMode('read') objSlc2.createImage() slcWidth = imageSlc1.getWidth() intWidth = int(slcWidth /\",\n \"imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName + '.int' objInt =\",\n \"int(slcWidth / rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt\",\n \"dest='rglooks', default=1, help='Range looks') return parser def cmdLineParse(iargs = None): parser = createParser()\",\n \"isceobj from components.stdproc.stdproc import crossmul from iscesys.ImageUtil.ImageUtil import ImageUtil as IU def createParser():\",\n \"= isceobj.createSlcImage() #right now imageSlc1 and 2 are just text files, need to\",\n \"= argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths') parser.add_argument('-m', '--master', type=str, dest='master',\",\n \"obj in [objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps\",\n \"rgLooks) lines = min(imageSlc1.getLength(), imageSlc2.getLength()) resampAmp = resampName + '.amp' resampInt = resampName\",\n \"= isceobj.createImage() img2.load(inps.slave + '.xml') os.makedirs(os.path.dirname(inps.prefix), exist_ok=True) run(img1, img2, inps.prefix, inps.azlooks, inps.rglooks) if\",\n \"int and amp files') parser.add_argument('-a', '--alks', type=int, dest='azlooks', default=1, help='Azimuth looks') parser.add_argument('-r', '--rlks',\",\n \"Line Parser. ''' parser = argparse.ArgumentParser( description='Generate offset field between two Sentinel swaths')\",\n \"help='Range looks') return parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs)\",\n \"looks') return parser def cmdLineParse(iargs = None): parser = createParser() return parser.parse_args(args=iargs) def\",\n \"inps = cmdLineParse(iargs) img1 = isceobj.createImage() img1.load(inps.master + '.xml') img2 = isceobj.createImage() img2.load(inps.slave\"\n]","isConversational":false}}}],"truncated":false,"partial":false},"paginationData":{"pageIndex":35,"numItemsPerPage":100,"numTotalItems":12866649,"offset":3500,"length":100}},"jwt":"eyJhbGciOiJFZERTQSIsImtpZCI6Im1sR1pFYktIVHRVSGRhX1RRdEczQ0N1S3k5ME9Qd25DNHY5elRjM3FVYzgifQ.eyJyZWFkIjp0cnVlLCJwZXJtaXNzaW9ucyI6eyJyZXBvLmNvbnRlbnQucmVhZCI6dHJ1ZX0sImlhdCI6MTc4MzUzMTcwNCwianRpIjoiMzI4OTVkMzYtNTBhMy00ODE5LTkxMjQtYWIzNGUwNGZlYjBlIiwic3ViIjoiL2RhdGFzZXRzL3RlcnJ5eXovc3RhcmNvZGVyZGF0YV9uZ3JhbV8xMyIsImV4cCI6MTc4MzUzNTMwNCwiaXNzIjoiaHR0cHM6Ly9odWdnaW5nZmFjZS5jbyJ9.MlqzOEsoKU7Wnnv_HlI0rGJUgt24G5BYIt123oSCAYPTuxABu_7fH8p_JYqWIl1O1CCiJF4wbFjrV8egq9SGCg","displayUrls":true,"splitSizeSummaries":[{"config":"default","split":"train","numRows":12866649,"numBytesParquet":149232926747}]},"discussionsStats":{"closed":0,"open":0,"total":0},"fullWidth":true,"hasGatedAccess":true,"hasFullAccess":true,"isEmbedded":false,"savedQueries":{"community":[],"user":[]}}">
ngram listlengths 0 67.8k |
|---|
[
"KIND, either express or implied. # See the License for the specific language",
"Unless required by applicable law or agreed to in writing, software # distributed",
"Return the rounded values of a tensor. \"\"\" error = tf.stop_gradient(tf.math.round(x) - x)",
"tensor. \"\"\" error = tf.stop_gradient(tf.math.... |
[
"\"\"\"Mark everything from ``start`` to the end of the line as Error.\"\"\" end",
"text[start] in self.LOGIC self.cur.append((start, Operator, text[start])) start = self.formula(start+1, text) assert text[start] ==",
"None groups = sorted(match.regs[1:]) # exclude whole match for group in groups: if",
"numbe... |
[
"django.urls import path from contacts.views import ( ContactsListView, CreateContactView, ContactDetailView, UpdateContactView, RemoveContactView, GetContactsView,",
"AddCommentView.as_view(), name=\"add_comment\"), path('comment/edit/', UpdateCommentView.as_view(), name=\"edit_comment\"), path('comment/remove/'... |
[
"@property def remote_addr(self): \"\"\"remote address IP :type: :class:`str`\"\"\" if not self.established: return None",
"\"\"\"Local address IP :type: :class:`str`\"\"\" return self._str_ipv6_addr(self.ucLocalAddr) @property def remote_addr(self): \"\"\"remote address IP",
"gdef.BSTR() self.get_Name(name) re... |
[
"if minArgs is not None and minArgs <= len(newArgs) + len(newKwArgs): return func(*newArgs,",
"100): print(a, b, c, d, e, f, g) # NOTE: '<====' means \"this",
"6 -1\" print(\"\\n------\\n\") # Example 4. def printTree(func, level = None): if level",
"# again, () forces the evaluation # it prints \"1 2 3 4",
... |
[
"data is temporarily stored in plaintext while the server handles message broadcasting Now",
"of server software written by <NAME> to allow people to talk to eachother",
"a raw TCP socket and data is temporarily stored in plaintext while the",
"not follow it. All data is sent to and from this server over",
... |
[
"\"wj\": \"저\", \"wh\": \"조\", \"wn\": \"주\", \"ak\": \"마\", \"aj\": \"머\", \"ah\": \"모\", \"an\":",
"cv2.waitKey(0) == 27: # break #cv2.destroyAllWindows() end = time.time() total_time = (end -",
": (1, 32, 42) out_best = list(np.argmax(out[0, 2:], axis=1)) # get max index",
"[k for k, g in itertools.groupby... |
[
"dtype, scope=None, device=\"\"): \"\"\" Convert 1dim numpy array to a torcharrow column (zero",
"pa.types.is_null(t): return dt.Void() if pa.types.is_string(t): return dt.String(nullable) if pa.types.is_map(t): return dt.Map(t.item_type, t.key_type, nullable)",
"pa.types.is_int8(t): return dt.Int8(nullable) if... |
[
"written to.') flags.DEFINE_string('eval_dir', '/tmp/compression/', 'Directory where the results are saved to.') flags.DEFINE_integer('max_number_of_evaluations', None,",
"KIND, either express or implied. # See the License for the specific language",
"Unless required by applicable law or agreed to in writing, s... |
[
"= \"\" for i in range(len(row_values) - 1): if row_values[i] is not None:",
"= <PASSWORD>, database = \"metrics\" #\"datacamp\" ) cursor = db_connection.cursor() query = \"use",
"os.environ['QUERY_ON'] def dump_query_results(): \"\"\" This is a simple SQL table dump of a",
"the query/results. Otherwise it is... |
[
"desktop_local_tests.windows.windows_dns_force_public_dns_servers_disrupter import WindowsDNSForcePublicDNSServersDisrupter class TestWindowsPacketCaptureDisruptForcePublicDNSServers(LocalPacketCaptureTestCaseWithDisrupter): # TODO: Make the packet capture here DNS",
"class TestWindowsPacketCaptureDisruptForcePub... |
[
"default loading image for using a such loader:: from kivy import * image",
"and pass to the client''' # want to start it ? if self._start_wanted:",
"ImageLoader.load(filename=image) else: self._error_image = image error_image = property(_get_error_image, _set_error_image) '''Image used for error.",
"OSError:... |
[
"So, the Output needs to be a bit clearer. friends = ['Pythobit','boy','Pythoman'] friend",
"friends are {friend}') # Output - My friends are Pythobit, boy, Pythoman #",
"are {friend}') # Output - My friends are Pythobit, boy, Pythoman # Here",
"# 13. Join # it allows to print list a bit better friends",
"[... |
[
"if set, mobile builds are built with 'ant'. requires the Android SDK and",
"SDK and appropriate mobile/local.properties file configured # preBuild - optional - an array",
"name of the default build to use if none is specified on the",
"URL base for external resources (all resources embedded in standard IITC)... |
[
"except: pass return self.upload(path, dest) print(\"As Dir\") try: self.sftp.mkdir(rpath) except IOError: pass for",
"path to upload to \"\"\" self.sftp.put(path, rpath) return True def get(self, url, path):",
"scheme: @param host - may be None, host to connect to implementor must",
"- may be None, password ... |
[
"// RENDER_RATIO, 120 // RENDER_RATIO)) self.rect = self.surf.get_rect() self.surf2 = pygame.Surface((30 // RENDER_RATIO,",
"RENDER_RATIO)) self.rect2 = self.surf2.get_rect() self.surf3 = pygame.Surface((30 // RENDER_RATIO, 40 // RENDER_RATIO)) self.rect3",
"0: b_rect.top = self.rect4.bottom b_speed[1] = -b_spe... |
[
"states): self.all_states = states self.train_set = None self.validation_set = None self.test_set = None",
"self.test_set = [s for s in all_states if s.state_id in test_ids] def shuffle_train_set(self):",
"for s in all_states])) random.shuffle(unique_state_ids) train_ids = set(unique_state_ids[0:int(0.6 * len(u... |
[
"colorspace to use for l2, ssim model types\", ) parser.add_argument( \"--batch_size\", type=int, default=50,",
"to use GPU\") parser.add_argument( \"--model_path\", type=str, default=None, help=\"location of model, will default to",
"for linearly calibrated net, [net] for off-the-shelf network, [l2] for euclid... |
[
"= np.std(data_first, axis=0) / 2. y_second_err = np.std(data_second, axis=0) / 2. plt.fill_between(x, y_first",
"y_second, color='g', label='Task B (transfer learning)') plt.legend(bbox_to_anchor=(0.8, 0.3), loc=2, ncol=1, fontsize=15) axes =",
"numpy as np import matplotlib.pyplot as plt plt.style.use('ggplot... |
[
"key, value in model.named_parameters(): if not value.requires_grad: continue param_lr = base_lr param_weight_decay =",
"not value.requires_grad: continue param_lr = base_lr param_weight_decay = weight_decay if \"bias\" in key:",
"= base_lr param_weight_decay = weight_decay if \"bias\" in key: param_lr = base_l... |
[
"= Swap(1.0).mutate(solution) self.assertEqual(result.variables[0][0], 1) self.assertEqual(result.variables[0][1], 0) def test_swap2b(self): problem = Problem(1, 0) problem.types[0]",
"Python module for designing and using # evolutionary algorithms (EAs) and multiobjective evolutionary",
"distributed in the hop... |
[
"h + 2 * e, h + 2 * e, ..., h +",
"[0., 1., 2.], [0., 1., 2.], [0., 1., 2.], [0., 1., 2.]], dtype=float32)",
"# handle both texture border types _edge_size = 0 # 如果包围方式是border, 那么边界长度是1, 在h和w维两侧加一排0",
"tf.variable_scope(name): ''' [num_batch, height, width, num_channels] ''' _num_batch = tf.shape(input_images... |
[
"spectral data in JSON format from OSN z (float) - redshift of SN",
"import matplotlib.pyplot as plt from astropy.io import ascii import json from IPython.display import",
"lightcurve_object): if lightcurve_object in self.lightcurves: print('already exists') else: self.lightcurves.append(lightcurve_object) #def... |
[
"from userbot import ALIVE_NAME from userbot.utils import admin_cmd DEFAULTUSER = str(ALIVE_NAME) if ALIVE_NAME",
"userbot.utils import admin_cmd DEFAULTUSER = str(ALIVE_NAME) if ALIVE_NAME else \"No name set yet",
"\"➖➖➖➖➖➖➖➖➖\\n\" \"`.fuk` فاكيو\\n\" \"➖➖➖➖➖➖➖➖➖\\n\" \"`.get_id` ايدي اي شخص دزه بمحادثته\\n\" ... |
[
"PNG bytes\"\"\" s = self.device_session_factory().exec_stream('screencap -p') data = recvall(s, 4194304) return data def",
"= self.device_session_factory().exec_stream('(echo -n %s; screencap) | nc %s %d' % (future.cookie.decode(), self.loopback, self.rch.port))",
"def __adb_device_name_detector(self): devices... |
[
"is None: self.new_dir = Path(inspect.getabsfile(inspect.stack()[1][0])).parent # type: ignore else: self.new_dir = directory def",
"of the file from which this function was called. \"\"\" def __init__(self, directory:",
"change_directory: \"\"\" A class for changing the working directory using a \"with\" state... |
[
"gui import MgallManager def main(): app = QApplication(sys.argv) ex = MgallManager() app.aboutToQuit.connect(ex.ExitHandler) sys.exit(app.exec_())",
"sys from PyQt5.QtWidgets import QApplication from gui import MgallManager def main(): app =",
"from gui import MgallManager def main(): app = QApplication(sys.ar... |
[
"return hash((self.x,self.y,self.z)) def __repr__(self): return f'({self.x},{self.y},{self.z})' def __add__(self,value): return Point3D(self.x + value.x, self.y",
"= z ''' Returns the distance between two 3D points ''' def distance(self,",
"x self.y = y self.z = z ''' Returns the distance between two",
"= y s... |
[
"can go in a set(). return self.id def __str__(self): return self.str() def str(self,",
"864, 3700), Function.ability(421, \"Research_TerranVehicleAndShipPlatingLevel2_quick\", cmd_quick, 865, 3700), Function.ability(422, \"Research_TerranVehicleAndShipPlatingLevel3_quick\", cmd_quick, 866, 3700), Function.abilit... |
[
"does not match' assert np.all(np.add(np.equal(target, 1), np.equal(target, 0))), \\ 'targets should be binary",
"/= ((self.targets - 1.0).sum() * -1.0) # calculating area under curve using trapezoidal",
"be higher when the model is more convinced that the example should be",
"and a randomly selected negative... |
[
"@property def current_x(self) -> int | None: \"\"\"Return cached value of the current_x",
"suppress(KeyError): return self.cluster[\"color_capabilities\"] if self.cluster.get(\"color_temperature\") is not None: return self.CAPABILITIES_COLOR_XY | self.CAPABILITIES_COLOR_TEMP return",
"\"\"\"Return cached value... |
[
"if self.qlimit: s += \" qlimit {.qlimit}\".format(self) return s class PFQueueStats(PFObject): \"\"\" \"\"\"",
"q.parent_qid = self.parent_qid q.realtime = self.realtime._to_struct() q.linkshare = self.linkshare._to_struct() q.upperlimit = self.upperlimit._to_struct() q.flowqueue",
"\"\"\" _struct_type = pf._s... |
[
"# coding=utf-8 # *** WARNING: this file was generated by the Pulumi SDK",
".environment_setting import * from .gallery_image import * from .get_environment import * from .get_environment_setting",
"typ == \"azure-native:labservices/v20181015:GalleryImage\": return GalleryImage(name, pulumi.ResourceOptions(urn=... |
[
"patch.object(schema_reader, 'validate_object') # @UndefinedVariable self.patcher_schema = patch.object(schema_reader, # @UndefinedVariable 'get_schema_fields', mock_schema_instance) self.patcher_schema.start() self.patcher_validate.start()",
"UserCollectionSerializer(data={'username': 'user', 'password': '<PASSW... |
[
"user and self.model.check_password(user, data['password']): expires = timedelta(days=365) user = UserSchema(only=('id', 'email', 'is_admin')).dump(user).data return",
"data = request.json if not data: return make_response(jsonify({'error': 'No data'}), 400) user =",
"class UserLoginResource(Resource): model = ... |
[
"\"rank\" FROM querybuilder_tests_order' self.assertEqual(query_str, expected_query, get_comparison_str(query_str, expected_query)) def test_rank_over_order(self): query = Query().from_table( table=Order,",
"(ORDER BY margin DESC) AS \"dense_rank\" ' 'FROM querybuilder_tests_order ' 'ORDER BY dense_rank",
"test... |
[
"import numpy h = .25 s = 1 bitmap = numpy.array([ [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0], [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],",
"<reponame>wbprice/ojimoji import numpy h = .25 s = 1 bitmap = numpy.array([ [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],",
"= 1 bitmap = numpy.array([ [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0], [0,0,0,0,0,0... |
[
"parser=argparse.ArgumentParser() parser.add_argument(\"--end\",type=float,default=100000.,help='end ratio') parser.add_argument(\"--save\",type=str,default=\"test_\",help='save name') parser.add_argument(\"--network\",type=str,default=\"rnn\",help='network name on symbols/') parser.add_argument(\"--right\",type=st... |
[
"position # action constant HOLD = 0 BUY = 1 SELL = 2",
"0 self.reward = 0 self.portfolio = 0 self.shares_held = 0 self.cost_basis = 0",
"+ temp_reward) elif (self.position == SHORT): temp_reward = ((self.entry_price - self.close_price) / self.close_price",
"round(new_portfolio,2) def step(self, action): # Ex... |
[
"JSON heatmaps...\") # compute_json_heatmaps(state_ids) print(\"Aggregating JSON files into one...\") aggegrate_json_files(state_ids) end = time.time()",
"geojson = json.load(f) with open(\"public/data/pop.json\", \"a\") as f: f.write(json.dumps(geojson[\"features\"])[1:-1] + (\",\" if state_id",
"+ 2 : file.in... |
[
"import os, glob import numpy as np from sklearn import model_selection classes =",
"y_train, y_test = model_selection.train_test_split(X, Y) xy = (X_train, X_test, y_train, y_test) np.save(\"./vehicle.npy\", xy)",
"sklearn import model_selection classes = [\"car\", \"bycycle\", \"motorcycle\", \"pedestrian\"] ... |
[
"waypoint_scenarios, quest_scenarios from services.build_campaign import Campaign from log_setup import log if __name__ ==",
"= quest_scenarios.get_number_of_quest_scenarios() log.info(f\"We have {number_quests_available} quests available\") random_waypoint_scenario = waypoint_scenarios.get_random_scenario(10) ra... |
[
"\"has_prefix\") print(info_dir) assert os.path.isfile(has_prefix_file) with open(has_prefix_file) as f: assert \"test2\" not in f.read()",
"= os.path.join(os.environ[\"ROOT\"], \"pkgs\") info_dir = os.path.join(pkgs, \"conda-build-test-ignore-some-prefix-files-1.0-0\", \"info\") has_prefix_file = os.path.join(in... |
[
"loss = alpha * loss_f + (1 - alpha) * loss_kv #print(loss_f.item(), loss_kv.item())",
"ConfigBinaryClassificationDistill from config import ConfigTripleClassification if __name__ == \"__main__\": args = get_args() print_args(args)",
"vocab from args import get_args, print_args from config import ConfigBinaryCl... |
[
"# OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. # IN",
"iso-8859-1 -*- # Copyright (c) 2016, <NAME> <<EMAIL>> # All rights reserved. #",
"this list of conditions and the following disclaimer. # 2. Redistributions in binary",
"ANY # THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LI... |
[
"downsample(infile: Path, outfile: Path, size: int, random): targets: List[FASTAItem] = list(read_fasta(infile)) if size",
"than the number of targets.\") targets = random.choice(targets, size, replace=False).tolist() with FASTAWriter(outfile) as",
"List[FASTAItem] = list(read_fasta(infile)) if size > len(targe... |
[
"= \"<KEY>\" # yahoo finance api key TICKER = \"TSLA\" INTERVAL = \"1m\"",
"LOOK_BACK = 30 # hard limit to not reach rate limit of 100",
"# API keys # YF_API_KEY = \"<KEY>\" # yahoo finance api key TICKER",
"API keys # YF_API_KEY = \"<KEY>\" # yahoo finance api key TICKER =",
"key TICKER = \"TSLA\" INTERVAL... |
[
"invoker_cmd_log, invoker_cmd_merge, invoker_cmd_pull_image, invoker_cmd_repo_check, invoker_cmd_repo_clear, invoker_cmd_sampling, invoker_cmd_terminate, invoker_cmd_user_create, invoker_task_factory, ) from proto import",
"invoker_cmd_pull_image, invoker_cmd_repo_check, invoker_cmd_repo_clear, invoker_cmd_sampli... |
[
"timedelta due_date_format = '%Y-%m-%d' datepicker_date_format = '%m%d%Y' def current_date(): return datetime.utcnow().strftime(due_date_format) def datepicker_current_date():",
"return datetime.utcnow().strftime(datepicker_date_format) def _date_from_today(days_to_add): return datetime.utcnow() + timedelta(days=... |
[
"import guvectorize import unittest def ufunc_add_core(a, b, c): for i in range(c.size): c[i]",
"in range(inp.shape[0]): np.testing.assert_equal(inp[i].sum(), out[i]) def test_gufunc_scalar_input_saxpy(self): def axpy(a, x, y, out): for i",
"= saxpy(A, X, Y) for j in range(10): exp = A * X[j]",
"range(c.size)... |
[
"def database_entry(data): try: load_dotenv() mongo_string = os.getenv('MONGODB_AUTH_URI') client = MongoClient(mongo_string) database = client[os.getenv('MONGODB_DB')]",
"import os from pymongo import MongoClient from dotenv import load_dotenv def database_entry(data): try:",
"os from pymongo import MongoClien... |
[
"try: with open(FILE, encoding='utf-8') as file: content = file.readlines() except FileNotFoundError: print('File does",
"except FileNotFoundError: print('File does not exist') except PermissionError: print('Permission denied') for line in",
"record['ip'] == ip: record['hostnames'].update(hosts) break else: hos... |
[
"dictionary , The headers of the request. Returns ------- n_sub : int, default=0",
"for top challengers. headers : dictionary , The headers of the request. Returns",
"we can make per day for the selected challenge. Parameters ---------- url :",
"): \"\"\"Get the number of submissions we can make per day for t... |
[
"-*- coding: utf-8 -*- import unittest from src.graph import Graph from src.maximum_cut import",
"('a', 'c'), ('d', 'e'), ('f', 'e'), ('f', 'g')], directed=False) (left, right) = maximum_cut_for_bipartite_graph(g)",
"graph. (u)-3-(v) | \\ / | | 5\\/1 4 2 /\\ | |",
"4') self.assertIn(len(right), [3,4], 'eighte... |
[
"print('rb magic: %s' % ((intptr_ptr + self.rb_magic_offset).cast(self.int_type.pointer()).dereference())) print('tree magic: %s' % ((intptr_ptr +",
"KIND, either express or implied. # See the License for the specific language",
"expr) return if ptr == 0: print('(null)') else: intptr_ptr = ptr.cast(self.intptr_... |
[
"= \"function (obj) { \" + \\ \"obj.total = obj.vowels + obj.cons; \"",
"\"with totals:\") displayGroup(results) if __name__==\"__main__\": mongo = MongoClient('mongodb://localhost:27017/') db = mongo['words'] collection =",
"if __name__==\"__main__\": mongo = MongoClient('mongodb://localhost:27017/') db = mong... |
[
"if key in combine: for sub_key, sub_val in output[key].items(): assert sub_key not in",
"5: assert batch['rgb'].shape[0] == 1, 'Only batch size 1 is supported for multi-cameras'",
"'logits_semantic', 'ord_probability', 'inv_depths', 'inv_depths_context', 'inv_depths1', 'inv_depths2', 'pred_depth', 'pred_depth_... |
[
"t5 = dt ** 5 c1 = (6.*dist + (a1 - a0) /",
"** 2 v1 = 0 v0 = xd * tau t1 = dt",
"* t3 + c4 * t2 + c5 * t1 + c6 xd",
"c1 * t5 + c2 * t4 + c3 * t3 + c4",
"end of traj reached!\" dist = goal - x a1 = 0 a0",
"2 v1 = 0 v0 = xd * tau t1 = dt t2",
"+ c4 * t2 + c5 * t1 + c6 xd = 5.*c1",
"c4 * t1 + c5 xdd = 2... |
[] |
[
"if exon_id == exon: test = translate(exon, human_fasta) break break true = 'MAEDADMRNELEEMQRRADQLADE'",
"# assert test == true # def test_make_sequence_array(finalsequencedf): # from orthoexon.util import make_sequence_array",
"test = separate(exon_id) true = \"ENSE00001229068\" assert test == true @pytest.fix... |
[
"= 0 feature = predict(path) for i, person_f in enumerate(person_feature): # 计算相识度 dist",
"audios = os.listdir(audio_db_path) for audio in audios: path = os.path.join(audio_db_path, audio) name =",
"from utils.record import RecordAudio parser = argparse.ArgumentParser() parser.add_argument('--audio_db', default... |
[
"loop products :) return Response( \"\\n\".join(t.name for t in _model.STORE.all_dataset_types()), content_type=\"text/plain\", ) @bp.route(\"/metadata-types.txt\")",
"f\"Metadata Type {type_.name}\", include_source_url=url_for( \".raw_metadata_type_doc\", name=type_.name, _external=True ), ) for type_ in _model.... |
[
"Copyright (c) 2021 <NAME> <<EMAIL>> # SPDX-License-Identifier: BSD-2-Clause # Board diagram/pinout: # https://user-images.githubusercontent.com/1450143/133655492-532d5e9a-0635-4889-85c9-68683d06cae0.png",
"Pins(\"35\"), IOStandard(\"LVCMOS33\")), # Leds (\"user_led\", 0, Pins(\"16\"), IOStandard(\"LVCMOS33\")), ... |
[
"search step_size = 1.0 for _ in range(ls_max_steps): vector_to_parameters(old_params - step_size * step,",
"+ 1): # lower learning rate after first update (for MAML, as described",
"<NAME>, <NAME>, \"Trust Region Policy Optimization\", 2015 (https://arxiv.org/abs/1502.05477) \"\"\" def __init__(self, sampler, ... |
[
"plural duration_plural = '{0}s'.format(duration) else: duration_plural = duration if duration_plural not in DURATION_OPTIONS:",
"{2} requests. Are you sure you want to do this? Type 'yes' to",
"return if options.get('interactive'): confirm = input(''' You have requested a database reset. This",
"not in DURAT... |
[
"assert message_to_sign in signed_blob decoded_blob = windows.crypto.verify_signature(cert, signed_blob) assert decoded_blob == message_to_sign def",
"do something else (bytes in generated ctypes ?) # This is fucking dumb,",
"= \"<PASSWORD>\" TEST_PFX = b\"\"\" <KEY> \"\"\" @pytest.fixture() def rawcert(): retu... |
[
"= self.graph_min_slider callback.args[\"high_slider\"] = self.graph_max_slider #ax.set_xlabel('RA') #ax.set_ylabel('DEC') #ax.grid(True, color='white') #fig.colorbar(im, ax=ax) #plugins.connect(fig, plugins.MousePosition(fontsize=14))",
"[] for px, py, err in zip(x, y, yerr): y_err_x.append((px, px)) y_err_y.app... |
[
"dir1 / 'file1.yml' file1.touch() file2 = dir1 / 'file2.yml' file2.touch() file3 = dir2",
"/ 'file2.yml' file2.touch() file3 = dir2 / 'file3.yml' file3.touch() my_pathlist = [Path(file3), Path(file1),",
"None def test_testaid_unit_pathlist_roles_whitelist(testvars_roles_whitelist): assert testvars_roles_whiteli... |
[
"= repr_str.replace('\\n', '\\\\n') # We check just the begin of the string: assert",
"Unless required by applicable law or agreed to in writing, software # distributed",
"PORT11_OID, 'parent': self.faked_fcp1.uri, 'class': 'storage-port', 'index': 1, 'name': 'fake-port11-name', 'description': 'FCP #1 Port",
... |
[
"tests on Windows for now. # Override the parent directory's \"unsupported\" decision until",
"only run a small set of tests on Windows for now. # Override",
"directory's \"unsupported\" decision until we can handle # all of its tests. if",
"a small set of tests on Windows for now. # Override the parent",
"... |
[
"sys.stderr = sys.__stderr__ orig_unraisablehook(unraisable) finally: sys.stderr = old_stderr def setup_unraisable_hook(): global orig_unraisablehook orig_unraisablehook",
"doctest.master = None try: ctypes = sys.modules['ctypes'] except KeyError: pass else: ctypes._reset_cache() try:",
"enumerate(names): basen... |
[
"valid') def base_tree(self, path_on_disk): connection = sqlite3.connect(path_on_disk) cursor = connection.cursor() base_tree = []",
"(1)', 'key': u'sqlite_master', 'folder': False, 'lazy': False }) else: for child in self.get_tables(key,",
"AND JSONIFY, SEE: http://stackoverflow.com/questions/12435297/how-do-i... |
[
"requests, json url = 'http://educacao.dadosabertosbr.com/api/cidades/ce' cidades = requests.get(url).content cidades = cidades.decode('utf-8') cidades =",
"import requests, json url = 'http://educacao.dadosabertosbr.com/api/cidades/ce' cidades = requests.get(url).content cidades = cidades.decode('utf-8') cidades... |
[] |
[
"transform = lookahead - state.pos transform = transform.getRotated(-state.angle) # Use the transformed vector",
"index def updateTargetVelocities(self, state): \"\"\"Update the target velocities of the left and right",
"= velocity # Limit the acceleration of the velocities for i in reversed(range(0,",
"state... |
[
"self # pylint: disable=arguments-differ def extend(self, *schemas, **kwargs): extra = kwargs.pop('extra', None) if",
"required_keys = all_required_keys.copy() # Build a map of all provided key-value pairs. #",
"schema, invalid_msg=None): invalid_msg = invalid_msg or 'mapping value' # Check some things that",
... |
[
"1 else: harmfulness += 1 if usefulness >= harmfulness: es.append(row_index) es_pred.append(pred) return es,",
"original data, otherwise filter the complete data :param complete: the complete dataframe :param",
"is auto select. :return: dc that local density threshold \"\"\" if self.dc ==",
"sort_rho_idx = np... |
[
"= Queue() que.put(root) ans, tmp, k = [], [], 1 while que.qsize() !=",
"Queue que = Queue() que.put(root) ans, tmp, k = [], [], 1 while",
"\"\"\" # Definition for a Node. \"\"\" class TreeNode(object): def __init__(self, val, children):",
"0: node = que.get() tmp.append(node.val) k -= 1 for child in node.chi... |
[
"You can customize using 'options'. If you set \"header\" list object, you can",
"\"\"\" return self.frame_buffer.recv_frame() def send_close(self, status=STATUS_NORMAL, reason=six.b(\"\")): \"\"\" send close data to the",
"list or dict. \"cookie\" -> cookie value. \"origin\" -> custom origin url. \"host\"",
... |
[
"return self.name class City(models.Model): name = models.CharField(max_length=50, verbose_name='Nome') state = models.ForeignKey(State, on_delete=models.CASCADE, verbose_name=State._meta.verbose_name)",
"= models.CharField(max_length=150, verbose_name='Bairro') cep = models.CharField(max_length=8, verbose_name='... |
[
"node = self.driver.list_nodes()[0] self.driver.reboot_node(node) def test_destroy_node(self): # Raises exception on failure node =",
"0.28] , \"host_uptime_s\" : 3378276 , \"host_mem_mb_free\" : 51825 , \"host_mem_mb_total\" : 73719",
", \"vps_parameters\" : { \"memory_mb\" : 160 , \"disk_space_mb\" : 4096 , \... |
[
"in polygon.holes for vertex in hole.vertices) @given(strategies.polygons_with_points) def test_convex_hull(polygon_with_point: Tuple[Polygon, Point]) -> None:",
"in polygon, point in polygon.convex_hull) @given(strategies.polygons_with_points) def test_indexing(polygon_with_point: Tuple[Polygon, Point]) -> None:... |
[
"multiple points l_pos_points = Points([[7.960064093299587, 1.3019528769064523, -2.6697181763370965], [7.960064093299587, 1.3019528769064523, -2.6697181763370965]]) p_dis_outs = chunk.project_local_points_on_raw(l_pos_points,",
"pytest.raises(LookupError): attempt3 = ReconsProject(\"not_supported_sfm\") def test_... |
[
"523, 532, 3017, 3018, 3019, ) STATUS_UNAUTHORIZED = (214, 277, 2553, 2554, 2555)",
"301, 302, 303, 304, 321, 323, 324, 325, 326, 327, 328, 329, 330,",
"276, 283, 284, 285, 286, 287, 288, 290, 293, 294, 295, 297, 300,",
"274, 278, 279, 280, 281, 282, 289, 291, 292, 296, 298, 305, 306,",
"402, 516, 517, 518,... |
[
"super(BezierTest, self).__init__(*args, **kwargs) self.d = 10 self.points = points self.loop = loop self.current_point",
"+ 1 return True return super(BezierTest, self).on_touch_down(touch) def on_touch_up(self, touch): if self.collide_point(touch.pos[0], touch.pos[1]):",
"self.collide_point(touch.pos[0], touc... |
[
"''), ('[\\t ]+', ' '), ('(-)?Effective user is.*', '')] init_script_1 = \"\"\"\"\"\" db_1",
"# min_versions: ['3.0.8'] # versions: 3.0.8 # qmid: None import pytest from firebird.qa",
"list on; select r.rdb$description as role_descr_blob_id from rdb$roles r where r.rdb$role_name = upper('rdb$admin');",
"OK. #... |
[
"dst_bbox[:, 1::4] = ctr_y - 0.5 * h dst_bbox[:, 2::4] = ctr_x +",
"= dy * src_height + src_ctr_y w = torch.exp(dw) * src_width h =",
"ctr_x = dx * src_width + src_ctr_x ctr_y = dy * src_height +",
"#-----------------------------------------------------------------# # 把y轴放前面是因为方便预测框和图像的宽高进行相乘 #---------------... |
[] |
[
"pyclipper.PFT_NONZERO) def test_clear(self): self.pc.Clear() with self.assertRaises(pyclipper.ClipperException): self.pc.Execute(*self.default_args) def test_exact_results(self): \"\"\" Test whether coordinates",
"pyclipper.PT_CLIP) self.pc.AddPaths([PATH_CLIP_1, INVALID_PATH], pyclipper.PT_CLIP) except pyclippe... |
[
"cells_dict[type_.find('Type').text] = cells return cells_dict #Load Xml Files xml_locs = ['/path/to/xml/files'] #same as",
"range(len(vals['counter'][cell_type])): print 'Iteration ' + str(i+1) + '/' + str(len(vals['counter'][cell_type])) #Convert Px Downsampled",
"ValueError('Found 0 or >1 matching file') #Ge... |
[
"max_iter = 10000 env.set_agent_at([2, 2], 0) env.set_pig_at([4, 4], 0) for i in range(max_iter):",
"for i in range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0, 4) print('action",
"i in range(max_iter): print(\"iter= \", i) env.render() action = random.randint(0, 4) print('action is',... |
[
"== {str} assert isinstance(d, pd.Series), d.columns assert all(map(_is_valid_dimension_name, d.index.names)) d.index.set_levels( [level.str.strip() for level",
"data = _read_tsv_gz(tsv_gz_path) data.to_hdf( hdf_path, _HDF_TABLE_PATH, complevel=conf[\"hdf_complevel\"], complib=conf[\"hdf_complib\"], ) # Replace e... |
[
"curr_data[k] = v cnn_feats = get_cnn_feats(cnn, curr_data, num_steps=num_frames_per_step * num_steps, training=False) emb_feats =",
"' does not have training argument.') training = kwargs['training'] if training: # Set",
"stride, stride) return steps def get_indices(curr_idx, num_steps, seq_len): steps = range... |
[
"= reader.read_float() if version >= (5, 4, 2) or ( version >= (5,",
"(4, 5): # 4.5 and up writer.write_vector4(self.m_Border) writer.write_float(self.m_PixelsToUnits) if version >= (5, 4,",
"self.packingRotation = SpritePackingRotation((self.m_settingsRaw >> 2) & 0xF) # 4 self.meshType = SpriteMeshType((self.m... |
[
"self.Alambda = f99(self.wave) self.Aflux = 10**(-0.4*self.Alambda) def extinction_correction(self, EBV, Rv=3.1, mag=True, source_lam=None, source_flux=None):",
"= 1. if self.throughput is not None: self.norm = np.trapz(self.throughput/self.wave, self.wave) def __repr__(self):",
"ctw95 = np.interp([0.025, 0.975... |
[
"None r = postorder.pop() root = TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:],",
"class Solution: def buildTree(self, inorder: List[int], postorder: List[int]) -> TreeNode: if not inorder:",
"TreeNode(r) index = inorder.index(r) root.right = self.buildTree(inorder[index+1:]... |
[
"[] new_pipeline_ids = [] loop_interrupted = False while self._should_continue(): try: if not loop_interrupted:",
"try: self._add_baseline_pipelines() except KeyboardInterrupt: if self._handle_keyboard_interrupt(): self._interrupted = True current_batch_pipelines = [] current_batch_pipeline_scores",
"percent_be... |
[
"graphene.Mutation): social = graphene.Field(types.SocialType) class Meta: abstract = True class Arguments: provider =",
"= graphene.String(required=True) @classmethod @setup_jwt_cookie @social_auth def mutate(cls, root, info, social, **kwargs): return cls.resolve(root,",
"def mutate(cls, root, info, social, **... |
[
"yellowbrick.regressor.base # Base classes for regressor Visualizers. # # Author: <NAME> <<EMAIL>> #",
"Base classes for regressor Visualizers. \"\"\" ########################################################################## ## Imports ########################################################################## fr... |
[
"[objInt, objAmp, objSlc1, objSlc2]: obj.finalizeImage() return imageInt, imageAmp def main(iargs=None): inps = cmdLineParse(iargs)",
"image') parser.add_argument('-o', '--outdir', type=str, dest='prefix', default='crossmul', help='Prefix of output int and amp files')",
"rgLooks): objSlc1 = isceobj.createSlcIma... |