0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() #","= speaker_id def collect_files(self): meta = join(self.data_root, \"train.txt\") with open(meta, \"rb\") as f:","== 0 # Set 0 for zero beginning padding # imitates initial decoder","self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None): if lengths","prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\"Target mel spectrogram\", mel_output, global_step) # Target spectrogram if linear_outputs is not","deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule import torch","None mel_outputs, attn, done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) #","mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\"Predicted mel spectrogram\",","frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1]) //","lengths or mask\") # (B, T, 1) if mask is None: mask =","else \"single\" for idx, text in enumerate(texts): signal, alignment, _, mel = synthesis.tts(","soft_mask attn_loss = (attn * soft_mask).mean() loss += attn_loss if global_step > 0","32, len(self.lengths)) if batch_group_size % batch_size != 0: batch_group_size -= batch_group_size % batch_size","is not None and attn.dim() == 4: for i, alignment in enumerate(attn): alignment","alignment alignment_dir = join(checkpoint_dir, \"alignment_ave\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \"step{:09d}_alignment.png\".format(global_step)) alignment =","data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \"\"\"Partially randmoized sampler 1. Sort","{}\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length","global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path =","Model model = build_model() if use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate,","Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset =","speaker_ids = [0, 1, 10] if ismultispeaker else [None] for speaker_id in speaker_ids:","elif train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat, _ = model.seq2seq( x,","return paths def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None):","mel_output, global_step) # Target spectrogram if linear_outputs is not None: linear_output = y[idx].cpu().data.numpy()","binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32)","a = np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths","__init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None):","nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import join, expanduser import random import librosa.display","\\\"{}\\\": {}\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input","None: batch_group_size = min(batch_size * 32, len(self.lengths)) if batch_group_size % batch_size != 0:","= nn.BCELoss() assert train_seq2seq or train_postnet global global_step, global_epoch while global_epoch < nepochs:","guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else","np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \"step{:09d}_text{}_{}_alignment.png\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag","speaker_str), signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def save_states(global_step, writer,","mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\"Predicted mel spectrogram\", mel_output, global_step) # Predicted","info=\"{}, {}, step={}\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram =","g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else soft_mask attn_loss =","* batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate:","= np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need","loss += attn_loss if global_step > 0 and global_step % checkpoint_interval == 0:","None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input) loss","args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return","= FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths","s = i * batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate","speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat,","r:, :] target_mask = target_mask[:, r:, :] else: decoder_target_mask, target_mask = None, None","self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e = 0, 0 for i in range(len(indices)","is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path,","attn.dim() == 4: for i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag =","not train_postnet: print(\"Training whole model\") train_seq2seq, train_postnet = True, True if train_seq2seq: print(\"Training","log_event_path is None: log_event_path = \"log/run-test\" + str(datetime.now()).replace(\" \", \"_\") print(\"Los event path:","= np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last",":], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) *","Feed data x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions","= model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion =","= positions # Downsample mel spectrogram if downsample_step > 1: mel = mel[:,","clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\"loss\", float(loss.data[0]),","// r // downsample_step # Feed data x, mel, y = Variable(x), Variable(mel),","https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\"Restore part of the model from: {}\".format(path)) state =","X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) #","idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length = input_lengths[idx]","(x, input_lengths, mel, y, positions, done, target_lengths, speaker_ids) \\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker","== self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast","checkpoint_dir = args[\"--checkpoint-dir\"] checkpoint_path = args[\"--checkpoint\"] checkpoint_seq2seq_path = args[\"--checkpoint-seq2seq\"] checkpoint_postnet_path = args[\"--checkpoint-postnet\"] load_embedding","lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size","flip against freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model, checkpoint_dir,","t] = 1 - np.exp(-(n / N - t / T)**2 / (2","im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep'","import sys from os.path import dirname, join from tqdm import tqdm, trange from","train_postnet global global_step, global_epoch while global_epoch < nepochs: running_loss = 0. for step,","range(T): W[n, t] = 1 - np.exp(-(n / N - t / T)**2","is not None and hparams.preset != \"\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset))","harded coded texts = [ \"Scientists at the CERN laboratory say they have","str(datetime.now()).replace(\" \", \"_\") print(\"Los event path: {}\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try:","signal {}_{}\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def","of the model. --log-event-path=<name> Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding","{}\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset","= list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64 to int","setup dataset = PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler,","\"\"\"Partially randmoized sampler 1. Sort by lengths 2. Pick a small patch and","None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\"Target linear spectrogram\", spectrogram, global_step) def","or train_postnet global global_step, global_epoch while global_epoch < nepochs: running_loss = 0. for","else None # Create mask if we use masked loss if hparams.masked_loss_weight >","text-to-speech synthesis model. usage: train.py [options] options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir>","mel_output, global_step) # Predicted spectrogram if linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy()","max_target_len % r assert max_target_len % r == 0 if max_target_len % downsample_step","if __name__ == \"__main__\": args = docopt(__doc__) print(\"Command line args:\\n\", args) checkpoint_dir =","return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\"{}, {}, step={}\".format( hparams.builder, time_string(),","= None, None # Apply model if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn,","for b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W","y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w > 0: binary_div = w *","downsample_step a = np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch = torch.LongTensor(a)","+ batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size)","+= 1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and","False # If not None, filter by speaker_id self.speaker_id = speaker_id def collect_files(self):","= torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in","s, e = 0, 0 for i in range(len(indices) // batch_group_size): s =","train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask) mel_loss","grad_norm, global_step) writer.add_scalar(\"learning rate\", current_lr, global_step) global_step += 1 running_loss += loss.data[0] averaged_loss","idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\"(Eval) Predicted audio signal {}_{}\".format(idx, speaker_str), signal, global_step,","target, lengths=None, mask=None, max_len=None): if lengths is None and mask is None: raise","l.decode(\"utf-8\").split(\"|\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker and self.speaker_id","_frontend eval_output_dir = join(checkpoint_dir, \"eval\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids = [0,","< len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self,","Reset optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use specific speaker of data","= Y # alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text,","train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda:","None data_root = args[\"--data-root\"] if data_root is None: data_root = join(dirname(__file__), \"data\", \"ljspeech\")","= torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in batch]) else: speaker_ids","e = s - 1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch),","if data_root is None: data_root = join(dirname(__file__), \"data\", \"ljspeech\") log_event_path = args[\"--log-event-path\"] reset_optimizer","\"step{:09d}_alignment.png\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \"averaged_alignment\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) *","D) mask_ = mask.expand_as(input) loss = self.criterion(input * mask_, target * mask_) return","checkpoint path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only Train only postnet model. --restore-parts=<path>","a single speaker dataset if self.speaker_id is not None: indices = np.array(speaker_ids) ==","init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda() linear_dim","speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler):","mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\"Target mel spectrogram\", mel_output, global_step) # Target spectrogram if","if max_target_len % downsample_step != 0: max_target_len += downsample_step - max_target_len % downsample_step","l = lines[0].decode(\"utf-8\").split(\"|\") assert len(l) == 4 or len(l) == 5 self.multi_speaker =","% r != 0: max_target_len += r - max_target_len % r assert max_target_len","assert mask is not None l1_loss = w * masked_l1(y_hat, y, mask=mask) +","1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \"step{:09d}_layer_{}_alignment.png\".format( global_step, i + 1)) save_alignment(path, alignment)","\"must be specified wrong args\" # Override hyper parameters hparams.parse(args[\"--hparams\"]) print(hparams_debug_string()) assert hparams.name","y, mask=mask) + (1 - w) * l1(y_hat, y) else: assert mask is","t in range(T): W[n, t] = 1 - np.exp(-(n / N - t","max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2],","= np.zeros((max_N, max_T), dtype=np.float32) for n in range(N): for t in range(T): W[n,","not None and hparams.preset != \"\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\"Override","= mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\"Target mel spectrogram\", mel_output, global_step) # Target spectrogram","def __init__(self, X, Mel, Y): self.X = X self.Mel = Mel self.Y =","max(target_lengths) if max_target_len % r != 0: max_target_len += r - max_target_len %","mel = mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker","def sequence_mask(sequence_length, max_len=None): if max_len is None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0)","s, e = s - 1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand(","# Presets if hparams.preset is not None and hparams.preset != \"\": preset =","audio signal\", signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path)","optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model =","\"Please call Stella.\", \"Some have accepted this as a miracle without any physical","datetime import datetime # The deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import",":].view(-1) # Handle last elements s += batch_group_size if s < len(indices): random.shuffle(indices[s:])","__init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids = None self.multi_speaker = False #","# (B, T, D) mask_ = mask.expand_as(input) loss = self.criterion(input * mask_, target","self.multi_speaker = False return texts return texts, speaker_ids else: return texts def collect_features(self,","= prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\"Target linear spectrogram\", spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x +","dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r // downsample_step # frame","priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight # L1 loss","np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts return texts,","# Feed data x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions)","indices = np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts","1).T) * 255), global_step) # save files as well for now alignment_dir =","# Alignment path = join(eval_output_dir, \"step{:09d}_text{}_{}_alignment.png\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag =","sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader","model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\"loss\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\"done_loss\", float(done_loss.data[0]), global_step)","train_postnet: writer.add_scalar(\"linear_loss\", float(linear_loss.data[0]), global_step) writer.add_scalar(\"linear_l1_loss\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\"linear_binary_div_loss\", float( linear_binary_div.data[0]), global_step) if train_seq2seq","target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain mask","from: {}\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\"state_dict\"]) if not reset_optimizer: optimizer_state = checkpoint[\"optimizer\"] if","= self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def","for x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r","'Decoder timestep' if info is not None: xlabel += '\\n\\n' + info plt.xlabel(xlabel)","global_step, i + 1)) save_alignment(path, alignment) # Save averaged alignment alignment_dir = join(checkpoint_dir,","spectrogram\", spectrogram, global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal))","model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\"Restore part of the model","path) # Target mel spectrogram if mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy()","speaker of data in case for multi-speaker datasets. -h, --help Show this help","--checkpoint-dir=<dir> Directory where to save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default:","self.batch_group_size = batch_group_size assert batch_group_size % batch_size == 0 self.permutate = permutate def","seq2seq model\") elif train_postnet: print(\"Training postnet model\") else: assert False, \"must be specified","% batch_size == 0 self.permutate = permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size","- w) * l1(y_hat, y) else: assert mask is None l1_loss = l1(y_hat,","from torch.utils import data as data_utils from torch.utils.data.sampler import Sampler import numpy as","model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \"__main__\": args = docopt(__doc__) print(\"Command line args:\\n\", args)","global_epoch = 0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend =","= args[\"--train-seq2seq-only\"] train_postnet = args[\"--train-postnet-only\"] # train both if not specified if not","dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule import torch from torch.utils import","linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion","import pyplot as plt import sys import os from tensorboardX import SummaryWriter from","path, info=\"{}, {}, step={}\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram","attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \"averaged_alignment\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) #","+ 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention,","is not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path is not None:","def prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) -","Create mask if we use masked loss if hparams.masked_loss_weight > 0: # decoder","return np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x","(spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip","dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \"step{:09d}_predicted.wav\".format( global_step)) try: writer.add_audio(\"Predicted audio signal\",","else: assert False, \"must be specified wrong args\" # Override hyper parameters hparams.parse(args[\"--hparams\"])","None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding if load_embedding is not None:","def logit(x, eps=1e-8): return torch.log(x + eps) - torch.log(1 - x + eps)","frame_positions.cuda() if train_postnet: y = y.cuda() mel = mel.cuda() done, target_lengths = done.cuda(),","speaker_id # using multi-speaker dataset as a single speaker dataset indices = np.array(speaker_ids)","sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input) loss = self.criterion(input *","e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm =","frame_positions = frame_positions.cuda() if train_postnet: y = y.cuda() mel = mel.cuda() done, target_lengths","assert len(l) == 4 or len(l) == 5 self.multi_speaker = len(l) == 5"],"string":"[\n \"global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs,\",\n \"hparams.masked_loss_weight > 0: # decoder output domain mask decoder_target_mask = sequence_mask( target_lengths /\",\n \"torch.autograd import Variable from torch import nn from torch import optim import torch.backends.cudnn\",\n \"in enumerate(texts): signal, alignment, _, mel = synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False)\",\n \"from os.path import join, expanduser import random import librosa.display from matplotlib import pyplot\",\n \"args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" # Presets\",\n \"self.frame_lengths = list(map(int, self.frame_lengths)) return paths def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource):\",\n \"else: return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None): if\",\n \"= False return texts return texts, speaker_ids else: return texts def collect_features(self, *args):\",\n \"> 1: # spectrogram-domain mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask =\",\n \"list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker and self.speaker_id is not None: speaker_ids\",\n \"mel_loss + done_loss elif train_postnet: loss = linear_loss # attention if train_seq2seq and\",\n \"args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id)\",\n \"script for seq2seq text-to-speech synthesis model. usage: train.py [options] options: --data-root=<dir> Directory contains\",\n \"W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B = len(input_lengths) max_input_len = input_lengths.max() W\",\n \"linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r, :], y[:, r:, :], target_mask, priority_bin=n_priority_freq,\",\n \"train_postnet: n_priority_freq = int(hparams.priority_freq / (fs * 0.5) * linear_dim) linear_l1_loss, linear_binary_div =\",\n \"epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1 def save_checkpoint(model, optimizer,\",\n \"later def _pad(seq, max_len, constant_values=0): return np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values)\",\n \"laboratory say they have discovered a new particle.\\\", \\\"There's a way to measure\",\n \"speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram\",\n \"mel spectrogram\\\", mel_output, global_step) # Predicted spectrogram if linear_outputs is not None: linear_output\",\n \"batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r // downsample_step #\",\n \"r != 0: max_target_len += r - max_target_len % r assert max_target_len %\",\n \"mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) # Target spectrogram if linear_outputs\",\n \"= -y * y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w > 0: binary_div\",\n \"text_positions = Variable(text_positions) frame_positions = Variable(frame_positions) done = Variable(done) target_lengths = Variable(target_lengths) speaker_ids\",\n \"attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at step {}\\\".format(global_step)) # idx\",\n \"= len(input_lengths) max_input_len = input_lengths.max() W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b\",\n \"positions, done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is not\",\n \"# to be set later def _pad(seq, max_len, constant_values=0): return np.pad(seq, (0, max_len\",\n \"Apply model if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn, done_hat = model( x,\",\n \"pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model = build_model() if use_cuda: model = model.cuda() optimizer\",\n \"\\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size =\",\n \"np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # save files as well for now alignment_dir\",\n \"elements s += batch_group_size if s < len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self):\",\n \"0: max_target_len += r - max_target_len % r assert max_target_len % r ==\",\n \"mel spectrogram if mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output))\",\n \"and global_step % checkpoint_interval == 0: save_states( global_step, writer, mel_outputs, linear_outputs, attn, mel,\",\n \"is not None else \\\"single\\\" for idx, text in enumerate(texts): signal, alignment, _,\",\n \"torch.FloatTensor(c) # text positions text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for x\",\n \"guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W def train(model, data_loader, optimizer, writer, init_lr=0.002,\",\n \"both if not specified if not train_seq2seq and not train_postnet: print(\\\"Training whole model\\\")\",\n \"as a single speaker dataset indices = np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices])\",\n \"if b_pad > 0: # s, e = s - 1, e -\",\n \"single speaker dataset if self.speaker_id is not None: indices = np.array(speaker_ids) == self.speaker_id\",\n \"max_len - len(x) - b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment,\",\n \"= w * masked_l1( y_hat[:, :, :priority_bin], y[:, :, :priority_bin], mask=mask) \\\\ +\",\n \"load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer)\",\n \"0 for i in range(len(indices) // batch_group_size): s = i * batch_group_size e\",\n \"ismultispeaker else [None] for speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is\",\n \"> 0: if w > 0: priority_loss = w * masked_l1( y_hat[:, :,\",\n \"__name__ == \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"]\",\n \"dtype=np.float32) for b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return\",\n \"miracle without any physical explanation.\\\", ] import dv3.synthesis synthesis._frontend = _frontend eval_output_dir =\",\n \"if batch_group_size is None: batch_group_size = min(batch_size * 32, len(self.lengths)) if batch_group_size %\",\n \"checkpoint_path is not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding if load_embedding\",\n \"model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path> Restore model from\",\n \"0 # Set 0 for zero beginning padding # imitates initial decoder states\",\n \"r - max_target_len % r assert max_target_len % r == 0 if max_target_len\",\n \"reset_optimizer): global global_step global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"])\",\n \"target_lengths.long().numpy() // r // downsample_step # Feed data x, mel, y = Variable(x),\",\n \"eps) def masked_mean(y, mask): # (B, T, D) mask_ = mask.expand_as(y) return (y\",\n \"need to cast numpy.int64 to int self.frame_lengths = list(map(int, self.frame_lengths)) return paths def\",\n \"r:, :] else: decoder_target_mask, target_mask = None, None # Apply model if train_seq2seq\",\n \"frame_positions), done, target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T,\",\n \"= Variable(done) target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker else None if\",\n \"+ (1 - w) * l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) else: priority_loss\",\n \"Group of 20 conference.\\\", \\\"Generative adversarial network or variational auto-encoder.\\\", \\\"Please call Stella.\\\",\",\n \"1) if mask is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D)\",\n \"Dataset and Dataloader setup dataset = PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset,\",\n \"checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(),\",\n \"hparams.downsample_step multi_speaker = len(batch[0]) == 4 # Lengths input_lengths = [len(x[0]) for x\",\n \"= seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand =\",\n \"elif train_postnet: loss = linear_loss # attention if train_seq2seq and hparams.use_guided_attention: soft_mask =\",\n \"path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use specific speaker\",\n \"attn, done_hat = None, None, None # Losses w = hparams.binary_divergence_weight # mel:\",\n \"np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c =\",\n \".expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion =\",\n \"loss if hparams.masked_loss_weight > 0: # decoder output domain mask decoder_target_mask = sequence_mask(\",\n \"Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx,\",\n \"in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line\",\n \"signal\\\", signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path) #\",\n \"import cm from warnings import warn from dv3.hparams import hparams, hparams_debug_string fs =\",\n \"= init_lr binary_criterion = nn.BCELoss() assert train_seq2seq or train_postnet global global_step, global_epoch while\",\n \"input_lengths = [len(x[0]) for x in batch] max_input_len = max(input_lengths) target_lengths = [len(x[1])\",\n \"assert len(l) == 4 or len(l) == 5 multi_speaker = len(l) == 5\",\n \"\\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # save files as\",\n \"model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\" m = model.postnet checkpoint_path = join( checkpoint_dir,\",\n \"dataset as a single speaker dataset if self.speaker_id is not None: indices =\",\n \"priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) * linear_l1_loss + w * linear_binary_div\",\n \"# Model model = build_model() if use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(),\",\n \"None # Create mask if we use masked loss if hparams.masked_loss_weight > 0:\",\n \"as data_utils from torch.utils.data.sampler import Sampler import numpy as np from numba import\",\n \"and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step)\",\n \"train_postnet: suffix = \\\"\\\" m = model elif train_seq2seq: suffix = \\\"_seq2seq\\\" m\",\n \"loss = linear_loss # attention if train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths,\",\n \"4 or len(l) == 5 multi_speaker = len(l) == 5 self.frame_lengths = list(\",\n \"\\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag\",\n \"use_cuda: if train_seq2seq: x = x.cuda() text_positions = text_positions.cuda() frame_positions = frame_positions.cuda() if\",\n \"else: binary_div = z.mean() return l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T,\",\n \"T, D) mask_ = mask.expand_as(input) loss = self.criterion(input * mask_, target * mask_)\",\n \"+= '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource):\",\n \"hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions\",\n \"tensorboardX import SummaryWriter from matplotlib import cm from warnings import warn from dv3.hparams\",\n \"mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) # Predicted spectrogram if\",\n \"from matplotlib import pyplot as plt import sys import os from tensorboardX import\",\n \"speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root,\",\n \"is not None else None data_root = args[\\\"--data-root\\\"] if data_root is None: data_root\",\n \"averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1 def save_checkpoint(model, optimizer, step,\",\n \"args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"]\",\n \"= args[\\\"--train-postnet-only\\\"] # train both if not specified if not train_seq2seq and not\",\n \"= PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory)\",\n \"mask\\\") # (B, T, 1) if mask is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1)\",\n \"data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model = build_model()\",\n \"reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is not None: print(\\\"Load optimizer state from\",\n \"torch import optim import torch.backends.cudnn as cudnn from torch.utils import data as data_utils\",\n \"> 0 and global_step % hparams.eval_interval == 0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker)\",\n \"self.Mel = Mel self.Y = Y # alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self,\",\n \"collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker = len(batch[0]) ==\",\n \"masked loss if hparams.masked_loss_weight > 0: # decoder output domain mask decoder_target_mask =\",\n \"done_loss elif train_seq2seq: loss = mel_loss + done_loss elif train_postnet: loss = linear_loss\",\n \"* 255), global_step) # save files as well for now alignment_dir = join(checkpoint_dir,\",\n \"if max_target_len % r != 0: max_target_len += r - max_target_len % r\",\n \"+ (1 - w) * l1(y_hat, y) else: assert mask is None l1_loss\",\n \"y) else: assert mask is None l1_loss = l1(y_hat, y) # Priority L1\",\n \"= args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id =\",\n \"Which model to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train\",\n \"print(\\\"Training postnet model\\\") else: assert False, \\\"must be specified wrong args\\\" # Override\",\n \":priority_bin], mask=mask) \\\\ + (1 - w) * l1(y_hat[:, :, :priority_bin], y[:, :,\",\n \"import numpy as np from numba import jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource\",\n \"frame positions s, e = 1, max_decoder_target_len + 1 # if b_pad >\",\n \"self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64 to int self.frame_lengths =\",\n \"= PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset = PyTorchDataset(X, Mel,\",\n \"max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1]) // r // downsample_step - 1),\",\n \"return model def load_checkpoint(path, model, optimizer, reset_optimizer): global global_step global global_epoch print(\\\"Load checkpoint\",\n \"batch_group_size): s = i * batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) #\",\n \"mask is not None l1_loss = w * masked_l1(y_hat, y, mask=mask) + (1\",\n \"target_mask = decoder_target_mask # shift mask decoder_target_mask = decoder_target_mask[:, r:, :] target_mask =\",\n \"torch.LongTensor([x[3] for x in batch]) else: speaker_ids = None return x_batch, input_lengths, mel_batch,\",\n \"global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state\",\n \"s, e = 1, max_decoder_target_len + 1 # if b_pad > 0: #\",\n \"+= r - max_target_len % r assert max_target_len % r == 0 if\",\n \"= \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted\",\n \"def __iter__(self): indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e = 0, 0\",\n \"done = Variable(done) target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker else None\",\n \"_frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions X = FileSourceDataset(TextDataSource(data_root,\",\n \"None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id # using\",\n \"seq2seq model. --train-postnet-only Train only postnet model. --restore-parts=<path> Restore part of the model.\",\n \"# Predicted spectrogram if linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram =\",\n \"__init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if\",\n \"max_len, constant_values=0): return np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len,\",\n \"- w) * mel_l1_loss + w * mel_binary_div # done: if train_seq2seq: done_loss\",\n \"batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm\",\n \"# Dataset and Dataloader setup dataset = PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader(\",\n \"writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at step\",\n \"assert train_seq2seq or train_postnet global global_step, global_epoch while global_epoch < nepochs: running_loss =\",\n \"alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255),\",\n \"save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path> Restore model\",\n \"x = np.pad(x, [(b_pad, max_len - len(x) - b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0)\",\n \"= [ \\\"Scientists at the CERN laboratory say they have discovered a new\",\n \"T)**2 / (2 * g * g)) return W def guided_attentions(input_lengths, target_lengths, max_target_len,\",\n \"optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model\",\n \"datetime # The deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import\",\n \"warnings import warn from dv3.hparams import hparams, hparams_debug_string fs = hparams.sample_rate global_step =\",\n \"y_batch, \\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def\",\n \"1).T) * 255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel),\",\n \"reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet\",\n \"speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) #\",\n \"T, 1) if mask is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T,\",\n \"import join, expanduser import random import librosa.display from matplotlib import pyplot as plt\",\n \"is None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path))\",\n \"data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model\",\n \"max_len=None): if lengths is None and mask is None: raise RuntimeError(\\\"Should provide either\",\n \"max(input_lengths) target_lengths = [len(x[1]) for x in batch] max_target_len = max(target_lengths) if max_target_len\",\n \"args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"]\",\n \"only seq2seq model. --train-postnet-only Train only postnet model. --restore-parts=<path> Restore part of the\",\n \"x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for\",\n \"loss = mel_loss + done_loss elif train_postnet: loss = linear_loss # attention if\",\n \"join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T)\",\n \"1)) save_alignment(path, alignment) # Save averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True)\",\n \"= spec_loss( linear_outputs[:, :-r, :], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss =\",\n \"weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path\",\n \"= \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path)\",\n \"as a miracle without any physical explanation.\\\", ] import dv3.synthesis synthesis._frontend = _frontend\",\n \"xlabel = 'Decoder timestep' if info is not None: xlabel += '\\\\n\\\\n' +\",\n \"prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x + eps)\",\n \"max_decoder_target_len + 1 # if b_pad > 0: # s, e = s\",\n \"model): print(\\\"Restore part of the model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict =\",\n \"MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None,\",\n \"data_root self.col = col self.frame_lengths = [] self.speaker_id = speaker_id def collect_files(self): meta\",\n \"g)) return W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B = len(input_lengths) max_input_len =\",\n \"input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x\",\n \"if batch_group_size % batch_size != 0: batch_group_size -= batch_group_size % batch_size self.batch_group_size =\",\n \"assert speaker_ids is None linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat = None, None,\",\n \"v for k, v in state.items() if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if\",\n \"max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x = np.pad(x, [(b_pad,\",\n \"args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id is not None else\",\n \"import FileSourceDataset, FileDataSource from os.path import join, expanduser import random import librosa.display from\",\n \"speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is not None # Learning\",\n \"Variable(y) text_positions = Variable(text_positions) frame_positions = Variable(frame_positions) done = Variable(done) target_lengths = Variable(target_lengths)\",\n \"self.permutate = permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e\",\n \"{}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary writer for tensorboard if log_event_path is None:\",\n \"Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions = Variable(frame_positions) done = Variable(done) target_lengths\",\n \"L1 loss if w > 0: assert mask is not None l1_loss =\",\n \"writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\",\",\n \"* mask_) return loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step\",\n \"docopt import docopt import sys from os.path import dirname, join from tqdm import\",\n \"embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary writer for tensorboard if log_event_path\",\n \"target * mask_) return loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r =\",\n \"train_postnet = args[\\\"--train-postnet-only\\\"] # train both if not specified if not train_seq2seq and\",\n \"by speaker_id # using multi-speaker dataset as a single speaker dataset if self.speaker_id\",\n \"float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention:\",\n \"def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w\",\n \"self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self,\",\n \"join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as f: lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\")\",\n \"collect_features(self, *args): if self.multi_speaker: text, speaker_id = args else: text = args[0] seq\",\n \"Y): self.X = X self.Mel = Mel self.Y = Y # alias self.multi_speaker\",\n \"done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker else None #\",\n \"positions s, e = 1, max_decoder_target_len + 1 # if b_pad > 0:\",\n \"= None return x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids\",\n \"# aha, need to cast numpy.int64 to int self.frame_lengths = list(map(int, self.frame_lengths)) return\",\n \"// r // downsample_step - 1), max_decoder_target_len, constant_values=1) for x in batch]) done\",\n \"self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort by lengths 2.\",\n \"!= 0: batch_group_size -= batch_group_size % batch_size self.batch_group_size = batch_group_size assert batch_group_size %\",\n \"new particle.\\\", \\\"There's a way to measure the acute emotional intelligence that has\",\n \"axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): # harded\",\n \"spectrogram-domain mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask # shift\",\n \"text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X)\",\n \"use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend = None # to\",\n \"r assert max_target_len % r == 0 if max_target_len % downsample_step != 0:\",\n \"from os.path import dirname, join from tqdm import tqdm, trange from datetime import\",\n \"model, checkpoint_dir, ismultispeaker): # harded coded texts = [ \\\"Scientists at the CERN\",\n \"text, p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format(\",\n \"z.mean() return l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T, g): W =\",\n \"if train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask))\",\n \"[default: checkpoints]. --hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path> Restore model from checkpoint path\",\n \"class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids = None self.multi_speaker\",\n \"list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64 to int self.frame_lengths = list(map(int, self.frame_lengths))\",\n \"try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass\",\n \"fs = hparams.sample_rate global_step = 0 global_epoch = 0 use_cuda = torch.cuda.is_available() if\",\n \"% batch_size != 0: batch_group_size -= batch_group_size % batch_size self.batch_group_size = batch_group_size assert\",\n \"= Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker else None if use_cuda: if train_seq2seq:\",\n \"\\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model =\",\n \"path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only Train only postnet model. --restore-parts=<path> Restore\",\n \"Restore model from checkpoint path if given. --checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint\",\n \"b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W def\",\n \"import dv3.synthesis synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard\",\n \"in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r // downsample_step\",\n \"frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset = PyTorchDataset(X, Mel, Y) data_loader\",\n \"= seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module):\",\n \"\\\"_postnet\\\" m = model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict()\",\n \"tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Mel writer.add_image(\\\"(Eval)\",\n \"as np from numba import jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path\",\n \"= None # to be set later def _pad(seq, max_len, constant_values=0): return np.pad(seq,\",\n \"== 0: save_states( global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint(\",\n \"assert False, \\\"must be specified wrong args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string())\",\n \"x in batch]) else: speaker_ids = None return x_batch, input_lengths, mel_batch, y_batch, \\\\\",\n \"by lengths 2. Pick a small patch and randomize it 3. Permutate mini-batchs\",\n \"speaker_id=None): self.data_root = data_root self.speaker_ids = None self.multi_speaker = False # If not\",\n \"text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r // downsample_step # frame positions\",\n \"build_model() if use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2),\",\n \"multi-speaker datasets. -h, --help Show this help message and exit \\\"\\\"\\\" from docopt\",\n \"return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class\",\n \"Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") #\",\n \"dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model = build_model() if\",\n \"if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if\",\n \"linear_outputs, attn, done_hat = model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq:\",\n \"not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding if load_embedding is not\",\n \"for idx, text in enumerate(texts): signal, alignment, _, mel = synthesis.tts( model, text,\",\n \"step, (x, input_lengths, mel, y, positions, done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train()\",\n \"except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target mel spectrogram if\",\n \"train_seq2seq: suffix = \\\"_seq2seq\\\" m = model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\" m\",\n \"= torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict = {k: v for k, v in\",\n \"not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step)\",\n \"beginning padding # imitates initial decoder states b_pad = r max_target_len += b_pad\",\n \"path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate,\",\n \"plt import sys import os from tensorboardX import SummaryWriter from matplotlib import cm\",\n \"and train_postnet: suffix = \\\"\\\" m = model elif train_seq2seq: suffix = \\\"_seq2seq\\\"\",\n \"Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use\",\n \"batch_group_size = min(batch_size * 32, len(self.lengths)) if batch_group_size % batch_size != 0: batch_group_size\",\n \"nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None): if lengths is None and\",\n \"== 5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids = list(map(lambda\",\n \"torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) y_batch =\",\n \"speaker_id # using multi-speaker dataset as a single speaker dataset if self.speaker_id is\",\n \"= torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is not\",\n \"(0, 0)], mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment, path, info=None): fig, ax =\",\n \"= model.state_dict() valid_state_dict = {k: v for k, v in state.items() if k\",\n \"== 4 # Lengths input_lengths = [len(x[0]) for x in batch] max_input_len =\",\n \"in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b\",\n \"tqdm import tqdm, trange from datetime import datetime # The deepvoice3 model from\",\n \"torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch =\",\n \"if not train_seq2seq and not train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet = True,\",\n \"* linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r, :], y[:, r:, :], target_mask,\",\n \"speaker_id self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\")\",\n \"and priority_w > 0: if w > 0: priority_loss = w * masked_l1(\",\n \"intelligence that has never gone out of style.\\\", \\\"President Trump met with other\",\n \"alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) #\",\n \"train_postnet=True): if use_cuda: model = model.cuda() linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step\",\n \"float( linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh >\",\n \"use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def load_checkpoint(path,\",\n \"in batch]) else: speaker_ids = None return x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions,\",\n \"train_seq2seq: loss = mel_loss + done_loss elif train_postnet: loss = linear_loss # attention\",\n \"use_cuda else soft_mask attn_loss = (attn * soft_mask).mean() loss += attn_loss if global_step\",\n \"= linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) # Predicted audio\",\n \"* l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) else: priority_loss = l1(y_hat[:, :, :priority_bin],\",\n \"datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step))\",\n \"= w * masked_mean(z, mask) + (1 - w) * z.mean() else: binary_div\",\n \"suffix = \\\"_seq2seq\\\" m = model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\" m =\",\n \"and train_postnet: mel_outputs, linear_outputs, attn, done_hat = model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions,\",\n \"input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r // downsample_step # Feed data x, mel,\",\n \"linear_loss = (1 - w) * linear_l1_loss + w * linear_binary_div # Combine\",\n \"The deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule import\",\n \"args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id is not None else None data_root =\",\n \"optimizer, reset_optimizer): global global_step global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path)\",\n \"text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat, _\",\n \"where to save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path>\",\n \"warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target mel spectrogram if mel_outputs is not None:\",\n \"\\\"_seq2seq\\\" m = model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\" m = model.postnet checkpoint_path\",\n \"path if given. --checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint path. --checkpoint-postnet=<path> Restore postnet\",\n \"speaker_id=None): self.data_root = data_root self.col = col self.frame_lengths = [] self.speaker_id = speaker_id\",\n \"# Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio path\",\n \"= list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64 to int self.frame_lengths = list(map(int,\",\n \"plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None):\",\n \"encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection,\",\n \"torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend = None # to be set\",\n \"exist_ok=True) # Input dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id))\",\n \"attn is not None and attn.dim() == 4: for i, alignment in enumerate(attn):\",\n \"< seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self,\",\n \"is None: raise RuntimeError(\\\"Should provide either lengths or mask\\\") # (B, T, 1)\",\n \"hard coded speaker_ids = [0, 1, 10] if ismultispeaker else [None] for speaker_id\",\n \"Target mel spectrogram if mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy() mel_output =\",\n \"for t in range(T): W[n, t] = 1 - np.exp(-(n / N -\",\n \"torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is not None:\",\n \"save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Predicted\",\n \"nn from torch import optim import torch.backends.cudnn as cudnn from torch.utils import data\",\n \"__init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1.\",\n \"enumerate(texts): signal, alignment, _, mel = synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False) signal\",\n \"decoder_lengths = target_lengths.long().numpy() // r // downsample_step # Feed data x, mel, y\",\n \"self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len is None:\",\n \"* y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w > 0: binary_div = w\",\n \"variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have accepted this as a miracle without\",\n \"trange from datetime import datetime # The deepvoice3 model from dv3.deepvoice3_pytorch import frontend,\",\n \"nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda() linear_dim = model.linear_dim r\",\n \"= _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids =\",\n \"linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at step {}\\\".format(global_step)) #\",\n \"def forward(self, input, target, lengths=None, mask=None, max_len=None): if lengths is None and mask\",\n \"= np.array([_pad(np.zeros(len(x[1]) // r // downsample_step - 1), max_decoder_target_len, constant_values=1) for x in\",\n \"* linear_binary_div # Combine losses if train_seq2seq and train_postnet: loss = mel_loss +\",\n \"single speaker dataset indices = np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths =\",\n \"randmoized sampler 1. Sort by lengths 2. Pick a small patch and randomize\",\n \"binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z = -y * y_hat_logits +\",\n \"soft_mask).mean() loss += attn_loss if global_step > 0 and global_step % checkpoint_interval ==\",\n \"dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None): self.data_root = data_root self.col =\",\n \"parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" # Presets if hparams.preset is not\",\n \"spectrogram\\\", mel_output, global_step) # Predicted spectrogram if linear_outputs is not None: linear_output =\",\n \"lines)) paths = list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker and self.speaker_id is\",\n \"as well for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path\",\n \"alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Predicted mel\",\n \"def _pad(seq, max_len, constant_values=0): return np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values) def\",\n \"= Variable(text_positions) frame_positions = Variable(frame_positions) done = Variable(done) target_lengths = Variable(target_lengths) speaker_ids =\",\n \"from tensorboardX import SummaryWriter from matplotlib import cm from warnings import warn from\",\n \":], mel[:, r:, :], decoder_target_mask) mel_loss = (1 - w) * mel_l1_loss +\",\n \"> 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step += 1\",\n \":, :priority_bin], mask=mask) \\\\ + (1 - w) * l1(y_hat[:, :, :priority_bin], y[:,\",\n \"parameters [default: ]. --checkpoint=<path> Restore model from checkpoint path if given. --checkpoint-seq2seq=<path> Restore\",\n \"max_input_len = max(input_lengths) target_lengths = [len(x[1]) for x in batch] max_target_len = max(target_lengths)\",\n \"use_cuda: cudnn.benchmark = False _frontend = None # to be set later def\",\n \"train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) print(\\\"Finished\\\") sys.exit(0)\",\n \"_frontend = None # to be set later def _pad(seq, max_len, constant_values=0): return\",\n \"message and exit \\\"\\\"\\\" from docopt import docopt import sys from os.path import\",\n \"torch.utils.data.sampler import Sampler import numpy as np from numba import jit from nnmnkwii.datasets\",\n \"> 1: mel = mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths = input_lengths.long().numpy() decoder_lengths\",\n \"use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay)\",\n \"y[:, :, :priority_bin], mask=mask) \\\\ + (1 - w) * l1(y_hat[:, :, :priority_bin],\",\n \"= list(np.array(texts)[indices]) self.multi_speaker = False return texts return texts, speaker_ids else: return texts\",\n \"np.exp(-(n / N - t / T)**2 / (2 * g * g))\",\n \"have discovered a new particle.\\\", \\\"There's a way to measure the acute emotional\",\n \"global_epoch = checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key =\",\n \"import dv3.lrschedule import torch from torch.utils import data as data_utils from torch.autograd import\",\n \"\\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is not None # Learning rate\",\n \"= checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"]\",\n \"--hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path> Restore model from checkpoint path if given.\",\n \"writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Predicted mel spectrogram if mel_outputs is\",\n \"hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram - np.min(spectrogram))\",\n \"def restore_parts(path, model): print(\\\"Restore part of the model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"]\",\n \"speaker_ids else: return texts def collect_features(self, *args): if self.multi_speaker: text, speaker_id = args\",\n \"state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part of the model from: {}\\\".format(path))\",\n \"= (1 - w) * mel_l1_loss + w * mel_binary_div # done: if\",\n \"return x def plot_alignment(alignment, path, info=None): fig, ax = plt.subplots() im = ax.imshow(\",\n \"* mask_, target * mask_) return loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\"\",\n \"+ 1), max_input_len) for x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len =\",\n \"\\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids = [0, 1, 10] if ismultispeaker\",\n \"else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None): self.data_root =\",\n \"float(attn_loss.data[0]), global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr,\",\n \"w * masked_l1( y_hat[:, :, :priority_bin], y[:, :, :priority_bin], mask=mask) \\\\ + (1\",\n \"speaker_str), prepare_spec_image(mel), global_step) # Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal,\",\n \"audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass\",\n \"= s - 1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len)\",\n \"speaker_ids.cuda() if ismultispeaker else None # Create mask if we use masked loss\",\n \"is None and mask is None: raise RuntimeError(\\\"Should provide either lengths or mask\\\")\",\n \"int(hparams.priority_freq / (fs * 0.5) * linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r,\",\n \"# Priority L1 loss if priority_bin is not None and priority_w > 0:\",\n \"idx = min(1, len(input_lengths) - 1) input_length = input_lengths[idx] # Alignment # Multi-hop\",\n \"ismultispeaker = speaker_ids is not None # Learning rate schedule if hparams.lr_schedule is\",\n \"filter by speaker_id self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with\",\n \"trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train both if not specified\",\n \"speaker_id is not None else \\\"single\\\" for idx, text in enumerate(texts): signal, alignment,\",\n \"train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat, _ = model.seq2seq( x, mel,\",\n \"== 5 self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l:\",\n \"%H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step)) def\",\n \"to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train both if\",\n \"alignment, _, mel = synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal))\",\n \"texts def collect_features(self, *args): if self.multi_speaker: text, speaker_id = args else: text =\",\n \"hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning\",\n \"model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is\",\n \"state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict = {k: v for k, v\",\n \"alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy()\",\n \"W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b in range(B): W[b] = guided_attention(input_lengths[b],\",\n \"numpy.int64 to int self.frame_lengths = list(map(int, self.frame_lengths)) return paths def collect_features(self, path): return\",\n \"# Setup summary writer for tensorboard if log_event_path is None: log_event_path = \\\"log/run-test\\\"\",\n \"if train_postnet: y = y.cuda() mel = mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda()\",\n \"data_utils from torch.autograd import Variable from torch import nn from torch import optim\",\n \"l1(y_hat, y) # Priority L1 loss if priority_bin is not None and priority_w\",\n \":, :priority_bin], y[:, :, :priority_bin]) else: priority_loss = l1(y_hat[:, :, :priority_bin], y[:, :,\",\n \"b_pad=0): x = np.pad(x, [(b_pad, max_len - len(x) - b_pad), (0, 0)], mode=\\\"constant\\\",\",\n \"w) * mel_l1_loss + w * mel_binary_div # done: if train_seq2seq: done_loss =\",\n \"join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs) except Exception as\",\n \"data_root, col, speaker_id=None): self.data_root = data_root self.col = col self.frame_lengths = [] self.speaker_id\",\n \"/ (2 * g * g)) return W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2):\",\n \"# s, e = s - 1, e - 1 frame_positions = torch.arange(s,\",\n \"# Input dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y\",\n \"loss.data[0] averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss\",\n \"return x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids def time_string():\",\n \"Directory where to save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default: ].\",\n \"plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1]\",\n \"import tqdm, trange from datetime import datetime # The deepvoice3 model from dv3.deepvoice3_pytorch\",\n \"hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" # Presets if hparams.preset is\",\n \"for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir,\",\n \"np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root, col,\",\n \"Predicted mel spectrogram if mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output =\",\n \"in batch], dtype=np.float32) y_batch = torch.FloatTensor(c) # text positions text_positions = np.array([_pad(np.arange(1, len(x[0])\",\n \"self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size is None: batch_group_size = min(batch_size\",\n \"docopt import sys from os.path import dirname, join from tqdm import tqdm, trange\",\n \"has never gone out of style.\\\", \\\"President Trump met with other leaders at\",\n \"b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len,\",\n \"data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which\",\n \"other leaders at the Group of 20 conference.\\\", \\\"Generative adversarial network or variational\",\n \"* masked_l1( y_hat[:, :, :priority_bin], y[:, :, :priority_bin], mask=mask) \\\\ + (1 -\",\n \"z.mean() else: binary_div = z.mean() return l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N, T,\",\n \"join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\":\",\n \"attention if attn is not None and attn.dim() == 4: for i, alignment\",\n \"= lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 multi_speaker = len(l)\",\n \"None, filter by speaker_id self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\")\",\n \"speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler =\",\n \"If not None, filter by speaker_id self.speaker_id = speaker_id def collect_files(self): meta =\",\n \"(1 - w) * l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) else: priority_loss =\",\n \"m = model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if\",\n \"Presets if hparams.preset is not None and hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset]\",\n \"* masked_mean(z, mask) + (1 - w) * z.mean() else: binary_div = z.mean()\",\n \"True if train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training postnet model\\\") else: assert\",\n \"priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) * linear_l1_loss + w * linear_binary_div #\",\n \"checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id\",\n \"Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths =\",\n \"downsample_step # frame positions s, e = 1, max_decoder_target_len + 1 # if\",\n \"speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {},\",\n \"checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key\",\n \"idx): if self.multi_speaker: text, speaker_id = self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else:\",\n \"restore_parts(path, model): print(\\\"Restore part of the model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict\",\n \"mel_loss = (1 - w) * mel_l1_loss + w * mel_binary_div # done:\",\n \"= sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand)\",\n \"Priority L1 loss if priority_bin is not None and priority_w > 0: if\",\n \"lines)) # Filter by speaker_id # using multi-speaker dataset as a single speaker\",\n \"= batch_group_size assert batch_group_size % batch_size == 0 self.permutate = permutate def __iter__(self):\",\n \"< nepochs: running_loss = 0. for step, (x, input_lengths, mel, y, positions, done,\",\n \"MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight # L1 loss if w >\",\n \"data_root = args[\\\"--data-root\\\"] if data_root is None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path\",\n \"target_mask = target_mask[:, r:, :] else: decoder_target_mask, target_mask = None, None # Apply\",\n \"batch] max_input_len = max(input_lengths) target_lengths = [len(x[1]) for x in batch] max_target_len =\",\n \"= join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag,\",\n \"Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model\",\n \"global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model): state =\",\n \"writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float(\",\n \"None and attn.dim() == 4: for i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy()\",\n \"x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len,\",\n \"= (attn * soft_mask).mean() loss += attn_loss if global_step > 0 and global_step\",\n \"[ \\\"Scientists at the CERN laboratory say they have discovered a new particle.\\\",\",\n \"global_step) # save files as well for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i +\",\n \"p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step,\",\n \"mask if we use masked loss if hparams.masked_loss_weight > 0: # decoder output\",\n \"\\\"Scientists at the CERN laboratory say they have discovered a new particle.\\\", \\\"There's\",\n \"y_batch = torch.FloatTensor(c) # text positions text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len)\",\n \"# Target spectrogram if linear_outputs is not None: linear_output = y[idx].cpu().data.numpy() spectrogram =\",\n \"be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train both if not\",\n \"Dataloader setup dataset = PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers,\",\n \"# shift mask decoder_target_mask = decoder_target_mask[:, r:, :] target_mask = target_mask[:, r:, :]\",\n \"0, 0 for i in range(len(indices) // batch_group_size): s = i * batch_group_size\",\n \"model, optimizer, reset_optimizer): global global_step global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint =\",\n \"paths = list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker and self.speaker_id is not\",\n \"mask.expand_as(input) loss = self.criterion(input * mask_, target * mask_) return loss / mask_.sum()\",\n \"input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None elif train_postnet:\",\n \"import torch from torch.utils import data as data_utils from torch.autograd import Variable from\",\n \"for i in range(len(indices) // batch_group_size): s = i * batch_group_size e =\",\n \"else: decoder_target_mask, target_mask = None, None # Apply model if train_seq2seq and train_postnet:\",\n \"of the model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict =\",\n \"self.multi_speaker: text, speaker_id = self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx],\",\n \"== 0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) # Update loss.backward() if clip_thresh >\",\n \"= synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) # Alignment path\",\n \"= l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) l1_loss = (1 - priority_w) *\",\n \"print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return\",\n \"speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs)\",\n \"l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by\",\n \"l1_loss = w * masked_l1(y_hat, y, mask=mask) + (1 - w) * l1(y_hat,\",\n \"torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in batch],\",\n \"input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r // downsample_step # Feed data\",\n \"mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at step {}\\\".format(global_step))\",\n \"== 4 or len(l) == 5 self.multi_speaker = len(l) == 5 texts =\",\n \"# reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None elif train_postnet: assert\",\n \"batch_size != 0: batch_group_size -= batch_group_size % batch_size self.batch_group_size = batch_group_size assert batch_group_size\",\n \"global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1 def save_checkpoint(model, optimizer, step, checkpoint_dir,\",\n \"meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as f: lines = f.readlines() l\",\n \"if priority_bin is not None and priority_w > 0: if w > 0:\",\n \"optimizer, reset_optimizer) if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path\",\n \"# hard coded speaker_ids = [0, 1, 10] if ismultispeaker else [None] for\",\n \"= mask.expand_as(y) return (y * mask_).sum() / mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None,\",\n \"hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset,\",\n \"\\\"single\\\" for idx, text in enumerate(texts): signal, alignment, _, mel = synthesis.tts( model,\",\n \"# train both if not specified if not train_seq2seq and not train_postnet: print(\\\"Training\",\n \"specific speaker of data in case for multi-speaker datasets. -h, --help Show this\",\n \"# decoder output domain mask decoder_target_mask = sequence_mask( target_lengths / (r * downsample_step),\",\n \"and Dataloader setup dataset = PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size,\",\n \"2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings,\",\n \"writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]),\",\n \"use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return\",\n \"# Lengths input_lengths = [len(x[0]) for x in batch] max_input_len = max(input_lengths) target_lengths\",\n \"if use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps,\",\n \"= f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5\",\n \"x in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths)\",\n \"if ismultispeaker else None if use_cuda: if train_seq2seq: x = x.cuda() text_positions =\",\n \"build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2\",\n \"dataset = PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn,\",\n \"+ (1 - w) * z.mean() else: binary_div = z.mean() return l1_loss, binary_div\",\n \"+= loss.data[0] averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss:\",\n \"hparams, hparams_debug_string fs = hparams.sample_rate global_step = 0 global_epoch = 0 use_cuda =\",\n \"def __init__(self, data_root, col, speaker_id=None): self.data_root = data_root self.col = col self.frame_lengths =\",\n \"= y.cuda() mel = mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda()\",\n \"_load_embedding(load_embedding, model) # Setup summary writer for tensorboard if log_event_path is None: log_event_path\",\n \"print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state =\",\n \"sampler 1. Sort by lengths 2. Pick a small patch and randomize it\",\n \"if hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z =\",\n \"soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else soft_mask attn_loss = (attn\",\n \"/= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step,\",\n \"Variable(frame_positions) done = Variable(done) target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker else\",\n \"import datetime # The deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio\",\n \"x in batch], dtype=np.float32) y_batch = torch.FloatTensor(c) # text positions text_positions = np.array([_pad(np.arange(1,\",\n \"is None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"]\",\n \"print(\\\"Save intermediate states at step {}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx =\",\n \"= min(batch_size * 32, len(self.lengths)) if batch_group_size % batch_size != 0: batch_group_size -=\",\n \"is not None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram,\",\n \"current_lr optimizer.zero_grad() # Used for Position encoding text_positions, frame_positions = positions # Downsample\",\n \"[0, 1, 10] if ismultispeaker else [None] for speaker_id in speaker_ids: speaker_str =\",\n \"if hparams.masked_loss_weight > 0: # decoder output domain mask decoder_target_mask = sequence_mask( target_lengths\",\n \"time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder,\",\n \"texts return texts, speaker_ids else: return texts def collect_features(self, *args): if self.multi_speaker: text,\",\n \"self.speaker_id is not None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by\",\n \"np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x =\",\n \"= Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z = -y * y_hat_logits + torch.log(1\",\n \"\\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs) except Exception as e:\",\n \"join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx,\",\n \"target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker else None if use_cuda: if\",\n \"decoder_target_mask = sequence_mask( target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1:\",\n \"checkpoints if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is\",\n \"\\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T)\",\n \"train_postnet: mel_outputs, linear_outputs, attn, done_hat = model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths)\",\n \"from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path,\",\n \"T, max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32) for n in range(N): for\",\n \"== self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts return texts, speaker_ids\",\n \"# Alignment # Multi-hop attention if attn is not None and attn.dim() ==\",\n \"collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1,\",\n \"(1 - w) * linear_l1_loss + w * linear_binary_div # Combine losses if\",\n \"info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root,\",\n \"# attention if train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask\",\n \"self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X) def sequence_mask(sequence_length,\",\n \"seq2seq text-to-speech synthesis model. usage: train.py [options] options: --data-root=<dir> Directory contains preprocessed features.\",\n \"= [] self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta,\",\n \"optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\":\",\n \"= hparams.downsample_step multi_speaker = len(batch[0]) == 4 # Lengths input_lengths = [len(x[0]) for\",\n \"priority_loss = w * masked_l1( y_hat[:, :, :priority_bin], y[:, :, :priority_bin], mask=mask) \\\\\",\n \"texts, speaker_ids else: return texts def collect_features(self, *args): if self.multi_speaker: text, speaker_id =\",\n \"# frame positions s, e = 1, max_decoder_target_len + 1 # if b_pad\",\n \"/ T)**2 / (2 * g * g)) return W def guided_attentions(input_lengths, target_lengths,\",\n \"and not train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet = True, True if train_seq2seq:\",\n \"list(np.array(texts)[indices]) self.multi_speaker = False return texts return texts, speaker_ids else: return texts def\",\n \"is not None: mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output,\",\n \"255), global_step) # Predicted mel spectrogram if mel_outputs is not None: mel_output =\",\n \"data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint(\",\n \"class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort by lengths 2. Pick a small\",\n \"len(l) == 4 or len(l) == 5 self.multi_speaker = len(l) == 5 texts\",\n \"* priority_loss # Binary divergence loss if hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_())\",\n \":].contiguous() # Lengths input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r // downsample_step\",\n \"global global_step global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if\",\n \"x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids def time_string(): return\",\n \"writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir,\",\n \"spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x\",\n \"> 0: assert mask is not None l1_loss = w * masked_l1(y_hat, y,\",\n \"mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at step {}\\\".format(global_step)) # idx =\",\n \"r // downsample_step # frame positions s, e = 1, max_decoder_target_len + 1\",\n \"attn_loss = (attn * soft_mask).mean() loss += attn_loss if global_step > 0 and\",\n \"and attn.dim() == 4: for i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag\",\n \"load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if\",\n \"* l1(y_hat, y) else: assert mask is None l1_loss = l1(y_hat, y) #\",\n \"None: print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"]\",\n \"\\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path,\",\n \"hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:,\",\n \"from checkpoint. --speaker-id=<N> Use specific speaker of data in case for multi-speaker datasets.\",\n \"f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 self.multi_speaker\",\n \"- np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip against\",\n \"b_pad > 0: # s, e = s - 1, e - 1\",\n \"mel spectrogram\\\", mel_output, global_step) # Target spectrogram if linear_outputs is not None: linear_output\",\n \"def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): # harded coded texts = [ \\\"Scientists\",\n \"1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1])\",\n \"x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs\",\n \"if self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id #\",\n \"initial decoder states b_pad = r max_target_len += b_pad * downsample_step a =\",\n \"path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio\",\n \"data in case for multi-speaker datasets. -h, --help Show this help message and\",\n \"Stella.\\\", \\\"Some have accepted this as a miracle without any physical explanation.\\\", ]\",\n \"np.pad(x, [(b_pad, max_len - len(x) - b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return x\",\n \"# using multi-speaker dataset as a single speaker dataset indices = np.array(speaker_ids) ==\",\n \"path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id)\",\n \"hparams.preset is not None and hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset] import json\",\n \"return texts return texts, speaker_ids else: return texts def collect_features(self, *args): if self.multi_speaker:\",\n \"or len(l) == 5 multi_speaker = len(l) == 5 self.frame_lengths = list( map(lambda\",\n \"epoch, train_seq2seq, train_postnet): if train_seq2seq and train_postnet: suffix = \\\"\\\" m = model\",\n \"l1_loss = (1 - priority_w) * l1_loss + priority_w * priority_loss # Binary\",\n \"mel spectrogram if mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output))\",\n \"model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir\",\n \"list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths =\",\n \"data as data_utils from torch.utils.data.sampler import Sampler import numpy as np from numba\",\n \"= max(input_lengths) target_lengths = [len(x[1]) for x in batch] max_target_len = max(target_lengths) if\",\n \"FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler\",\n \"super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None): if\",\n \"* downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain mask target_mask = sequence_mask(\",\n \"synthesis model. usage: train.py [options] options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir> Directory\",\n \"not None # Learning rate schedule if hparams.lr_schedule is not None: lr_schedule_f =\",\n \"Target spectrogram if linear_outputs is not None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output))\",\n \"decoder_target_mask, target_mask = None, None # Apply model if train_seq2seq and train_postnet: mel_outputs,\",\n \"= torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model):\",\n \"Set 0 for zero beginning padding # imitates initial decoder states b_pad =\",\n \"= hparams.masked_loss_weight # L1 loss if w > 0: assert mask is not\",\n \"model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict = {k: v\",\n \"w * mel_binary_div # done: if train_seq2seq: done_loss = binary_criterion(done_hat, done) # linear:\",\n \"import sys import os from tensorboardX import SummaryWriter from matplotlib import cm from\",\n \"soft_mask = soft_mask.cuda() if use_cuda else soft_mask attn_loss = (attn * soft_mask).mean() loss\",\n \"X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text, speaker_id = self.X[idx] return text, self.Mel[idx],\",\n \"decoder_target_mask) mel_loss = (1 - w) * mel_l1_loss + w * mel_binary_div #\",\n \"if linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear\",\n \"to be set later def _pad(seq, max_len, constant_values=0): return np.pad(seq, (0, max_len -\",\n \"list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) #\",\n \"not None and attn.dim() == 4: for i, alignment in enumerate(attn): alignment =\",\n \"downsample_step > 1: # spectrogram-domain mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask\",\n \"w) * l1(y_hat, y) else: assert mask is None l1_loss = l1(y_hat, y)\",\n \"pyplot as plt import sys import os from tensorboardX import SummaryWriter from matplotlib\",\n \"MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self,\",\n \"--checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint path. --checkpoint-postnet=<path> Restore postnet model from checkpoint\",\n \"step {}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1)\",\n \"not None l1_loss = w * masked_l1(y_hat, y, mask=mask) + (1 - w)\",\n \"prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) # Predicted spectrogram if linear_outputs is not\",\n \"plt.subplots() im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder\",\n \"the CERN laboratory say they have discovered a new particle.\\\", \\\"There's a way\",\n \"(text_positions, frame_positions), done, target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn):\",\n \"Restore postnet model from checkpoint path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only Train\",\n \"not None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id #\",\n \"= list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id # using multi-speaker dataset\",\n \"batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size is None:\",\n \"if info is not None: xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep')\",\n \"path, info=None): fig, ax = plt.subplots() im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none')\",\n \"self.data_root = data_root self.col = col self.frame_lengths = [] self.speaker_id = speaker_id def\",\n \"r // downsample_step - 1), max_decoder_target_len, constant_values=1) for x in batch]) done =\",\n \"// downsample_step # Feed data x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions\",\n \"global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\",\",\n \"model def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key]\",\n \"= plt.subplots() im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel =\",\n \"global_step) # Target spectrogram if linear_outputs is not None: linear_output = y[idx].cpu().data.numpy() spectrogram\",\n \"decoder_target_mask = decoder_target_mask[:, r:, :] target_mask = target_mask[:, r:, :] else: decoder_target_mask, target_mask\",\n \"if mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel\",\n \"seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input,\",\n \"max_target_len, max_input_len), dtype=np.float32) for b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len,\",\n \"max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) y_batch = torch.FloatTensor(c) # text positions\",\n \"/ N - t / T)**2 / (2 * g * g)) return\",\n \"if not specified if not train_seq2seq and not train_postnet: print(\\\"Training whole model\\\") train_seq2seq,\",\n \"force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def\",\n \"random import librosa.display from matplotlib import pyplot as plt import sys import os\",\n \"np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource):\",\n \"train_postnet: suffix = \\\"_postnet\\\" m = model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix))\",\n \"optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use specific speaker of data in\",\n \"Use specific speaker of data in case for multi-speaker datasets. -h, --help Show\",\n \"train_postnet) if global_step > 0 and global_step % hparams.eval_interval == 0: eval_model(global_step, writer,\",\n \"else None data_root = args[\\\"--data-root\\\"] if data_root is None: data_root = join(dirname(__file__), \\\"data\\\",\",\n \"= l1(y_hat, y) # Priority L1 loss if priority_bin is not None and\",\n \"= join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids = [0, 1, 10]\",\n \"else: speaker_ids = None return x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions), done,\",\n \"else: return texts def collect_features(self, *args): if self.multi_speaker: text, speaker_id = args else:\",\n \"= self.batch_group_size s, e = 0, 0 for i in range(len(indices) // batch_group_size):\",\n \"y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions = Variable(frame_positions) done =\",\n \"None and mask is None: raise RuntimeError(\\\"Should provide either lengths or mask\\\") #\",\n \"downsample_step - max_target_len % downsample_step assert max_target_len % downsample_step == 0 # Set\",\n \"mel, y, positions, done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids\",\n \"ismultispeaker else None # Create mask if we use masked loss if hparams.masked_loss_weight\",\n \"float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step)\",\n \"speaker_id = args else: text = args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker:\",\n \"= i * batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches\",\n \"= text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet: y = y.cuda() mel = mel.cuda()\",\n \"train_seq2seq: done_loss = binary_criterion(done_hat, done) # linear: if train_postnet: n_priority_freq = int(hparams.priority_freq /\",\n \"train_seq2seq: x = x.cuda() text_positions = text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet: y\",\n \"Restore part of the model. --log-event-path=<name> Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path>\",\n \"from torch import nn from torch import optim import torch.backends.cudnn as cudnn from\",\n \"xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class\",\n \"= attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step)\",\n \"Handle last elements s += batch_group_size if s < len(indices): random.shuffle(indices[s:]) return iter(indices)\",\n \"/ (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch +=\",\n \"print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch,\",\n \"* 0.5) * linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r, :], y[:, r:,\",\n \"= checkpoint[\\\"optimizer\\\"] if optimizer_state is not None: print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"])\",\n \"> 0: # s, e = s - 1, e - 1 frame_positions\",\n \"if load_embedding is not None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup\",\n \"i * batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if\",\n \"+ 1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path,\",\n \"= hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker = len(batch[0]) == 4 # Lengths input_lengths\",\n \"- w) * z.mean() else: binary_div = z.mean() return l1_loss, binary_div @jit(nopython=True) def\",\n \"int self.frame_lengths = list(map(int, self.frame_lengths)) return paths def collect_features(self, path): return np.load(path) class\",\n \"- torch.log(1 - x + eps) def masked_mean(y, mask): # (B, T, D)\",\n \"# Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" # Presets if\",\n \"join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model to\",\n \"not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path,\",\n \"e = 1, max_decoder_target_len + 1 # if b_pad > 0: # s,\",\n \"= model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state\",\n \"model\\\") else: assert False, \\\"must be specified wrong args\\\" # Override hyper parameters\",\n \"downsample_step # Feed data x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions =\",\n \"- priority_w) * l1_loss + priority_w * priority_loss # Binary divergence loss if\",\n \"loss if hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z\",\n \"if s < len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object):\",\n \"Mel self.Y = Y # alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if\",\n \"spectrogram\\\", mel_output, global_step) # Target spectrogram if linear_outputs is not None: linear_output =\",\n \"+= b_pad * downsample_step a = np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int)\",\n \"b_pad=b_pad) for x in batch], dtype=np.float32) y_batch = torch.FloatTensor(c) # text positions text_positions\",\n \"join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i +\",\n \"= z.mean() return l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T, g): W\",\n \"= args[\\\"--reset-optimizer\\\"] # Which model to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet =\",\n \"l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) l1_loss = (1 - priority_w) * l1_loss\",\n \"np from numba import jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import\",\n \"= torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad)\",\n \"= torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) y_batch\",\n \"\\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion\",\n \"global_step > 0 and global_step % checkpoint_interval == 0: save_states( global_step, writer, mel_outputs,\",\n \"join from tqdm import tqdm, trange from datetime import datetime # The deepvoice3\",\n \"to save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path> Restore\",\n \"writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x + eps) -\",\n \"4 or len(l) == 5 self.multi_speaker = len(l) == 5 texts = list(map(lambda\",\n \"(1 - w) * mel_l1_loss + w * mel_binary_div # done: if train_seq2seq:\",\n \"texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts return texts, speaker_ids else: return\",\n \"from checkpoint path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only Train only postnet model.\",\n \"model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None\",\n \"for x in batch] max_target_len = max(target_lengths) if max_target_len % r != 0:\",\n \"1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and train_postnet:\",\n \"= speaker_ids.cuda() if ismultispeaker else None # Create mask if we use masked\",\n \"loss.backward() if clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs\",\n \"lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() #\",\n \"os from tensorboardX import SummaryWriter from matplotlib import cm from warnings import warn\",\n \"max_len, b_pad=0): x = np.pad(x, [(b_pad, max_len - len(x) - b_pad), (0, 0)],\",\n \"torch import nn from torch import optim import torch.backends.cudnn as cudnn from torch.utils\",\n \"# Load embedding if load_embedding is not None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding,\",\n \"indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last elements s += batch_group_size if s <\",\n \"= np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for x in batch], dtype=np.int) text_positions =\",\n \"y = y.cuda() mel = mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids =\",\n \"float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]),\",\n \"rate\\\", current_lr, global_step) global_step += 1 running_loss += loss.data[0] averaged_loss = running_loss /\",\n \"4: for i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i +\",\n \"origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep' if info is not None:\",\n \"dataset as a single speaker dataset indices = np.array(speaker_ids) == self.speaker_id paths =\",\n \"] import dv3.synthesis synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) #\",\n \"conference.\\\", \\\"Generative adversarial network or variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have accepted\",\n \"__init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None):\",\n \"= list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root, f), paths)) if\",\n \"* downsample_step a = np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch =\",\n \"Lengths input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r // downsample_step # Feed\",\n \"= mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None elif train_postnet: assert speaker_ids is None\",\n \"linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r, :], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight)\",\n \"= getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1,\",\n \"np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None): self.data_root = data_root self.col\",\n \"elif train_postnet: print(\\\"Training postnet model\\\") else: assert False, \\\"must be specified wrong args\\\"\",\n \"self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self):\",\n \"batch_size = sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand =\",\n \"dtype=np.float32) for n in range(N): for t in range(T): W[n, t] = 1\",\n \"writer, model, checkpoint_dir, ismultispeaker) # Update loss.backward() if clip_thresh > 0: grad_norm =\",\n \"binary_div = w * masked_mean(z, mask) + (1 - w) * z.mean() else:\",\n \"as e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths,\",\n \"else: priority_loss = l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) l1_loss = (1 -\",\n \"len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X,\",\n \"indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last elements s += batch_group_size if\",\n \"not None, filter by speaker_id self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root,\",\n \"- np.exp(-(n / N - t / T)**2 / (2 * g *\",\n \"_ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel),\",\n \"l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T, g): W = np.zeros((max_N, max_T),\",\n \"= torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand\",\n \"args[\\\"--reset-optimizer\\\"] # Which model to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"]\",\n \"mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask) mel_loss = (1\",\n \"= join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path, alignment) # Save averaged alignment\",\n \"is not None # Learning rate schedule if hparams.lr_schedule is not None: lr_schedule_f\",\n \"and self.speaker_id is not None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter\",\n \"= FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare\",\n \"by speaker_id # using multi-speaker dataset as a single speaker dataset indices =\",\n \"np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths)\",\n \"pass dv3.audio.save_wav(signal, path) # Target mel spectrogram if mel_outputs is not None: mel_output\",\n \"shift mask decoder_target_mask = decoder_target_mask[:, r:, :] target_mask = target_mask[:, r:, :] else:\",\n \"Predicted spectrogram if linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output))\",\n \"= args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts =\",\n \"len(l) == 5 self.multi_speaker = len(l) == 5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3],\",\n \"save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) #\",\n \"args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path =\",\n \"checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id\",\n \"/= np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment)\",\n \"max_len is None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0) seq_range = torch.arange(0, max_len).long()\",\n \"= hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset,\",\n \"self.frame_lengths = [] self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with\",\n \"clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step +=\",\n \"x in batch] max_input_len = max(input_lengths) target_lengths = [len(x[1]) for x in batch]\",\n \"explanation.\\\", ] import dv3.synthesis synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True)\",\n \"assert max_target_len % downsample_step == 0 # Set 0 for zero beginning padding\",\n \"at the CERN laboratory say they have discovered a new particle.\\\", \\\"There's a\",\n \"with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True)\",\n \"alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep' if info is\",\n \"mask) + (1 - w) * z.mean() else: binary_div = z.mean() return l1_loss,\",\n \"return l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T, g): W = np.zeros((max_N,\",\n \"try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception as\",\n \"if train_postnet: n_priority_freq = int(hparams.priority_freq / (fs * 0.5) * linear_dim) linear_l1_loss, linear_binary_div\",\n \"None else None data_root = args[\\\"--data-root\\\"] if data_root is None: data_root = join(dirname(__file__),\",\n \"args[\\\"--data-root\\\"] if data_root is None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"]\",\n \"# save files as well for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1))\",\n \"g).T return W def train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0,\",\n \"tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Predicted mel spectrogram\",\n \"writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\",\",\n \"text, speaker_id = self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx],\",\n \"batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in\",\n \"state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path,\",\n \"if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step\",\n \"* 255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step)\",\n \"= \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part of\",\n \"is None: batch_group_size = min(batch_size * 32, len(self.lengths)) if batch_group_size % batch_size !=\",\n \"= max_target_len // r // downsample_step # frame positions s, e = 1,\",\n \"0 and global_step % checkpoint_interval == 0: save_states( global_step, writer, mel_outputs, linear_outputs, attn,\",\n \"frame_positions = positions # Downsample mel spectrogram if downsample_step > 1: mel =\",\n \"= X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text, speaker_id = self.X[idx] return text,\",\n \"text in enumerate(texts): signal, alignment, _, mel = synthesis.tts( model, text, p=0, speaker_id=speaker_id,\",\n \"- 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done =\",\n \"Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions = Variable(frame_positions) done = Variable(done) target_lengths =\",\n \"a new particle.\\\", \\\"There's a way to measure the acute emotional intelligence that\",\n \"checkpoint_dir, ismultispeaker) # Update loss.backward() if clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(),\",\n \"e: warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target mel spectrogram if mel_outputs is not\",\n \"}, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim,\",\n \"w * masked_mean(z, mask) + (1 - w) * z.mean() else: binary_div =\",\n \"padding # imitates initial decoder states b_pad = r max_target_len += b_pad *\",\n \"if hparams.lr_schedule is not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr,\",\n \"Downsample mel spectrogram if downsample_step > 1: mel = mel[:, 0::downsample_step, :].contiguous() #\",\n \"= args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model to be trained train_seq2seq =\",\n \"spectrogram if downsample_step > 1: mel = mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths\",\n \"import dv3.audio import dv3.lrschedule import torch from torch.utils import data as data_utils from\",\n \"Exception as e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel, y,\",\n \"global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal,\",\n \"Setup summary writer for tensorboard if log_event_path is None: log_event_path = \\\"log/run-test\\\" +\",\n \"mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask) mel_loss =\",\n \"model from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule import torch from\",\n \"not None and priority_w > 0: if w > 0: priority_loss = w\",\n \"join, expanduser import random import librosa.display from matplotlib import pyplot as plt import\",\n \"done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is not None\",\n \"# Set 0 for zero beginning padding # imitates initial decoder states b_pad\",\n \"Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format(\",\n \"optimizer, reset_optimizer) # Load embedding if load_embedding is not None: print(\\\"Loading embedding from\",\n \"input, target, lengths=None, mask=None, max_len=None): if lengths is None and mask is None:\",\n \"Lengths input_lengths = [len(x[0]) for x in batch] max_input_len = max(input_lengths) target_lengths =\",\n \"as plt import sys import os from tensorboardX import SummaryWriter from matplotlib import\",\n \"# if b_pad > 0: # s, e = s - 1, e\",\n \"= prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) # Predicted spectrogram if linear_outputs is\",\n \"frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat, _ =\",\n \"None # Losses w = hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss, mel_binary_div =\",\n \"== 5 multi_speaker = len(l) == 5 self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]),\",\n \"linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\",\",\n \"\\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try: train(model,\",\n \"= model.cuda() linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr =\",\n \"None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] #\",\n \"train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda() linear_dim = model.linear_dim r = hparams.outputs_per_step\",\n \"train_seq2seq, train_postnet) if global_step > 0 and global_step % hparams.eval_interval == 0: eval_model(global_step,\",\n \"using multi-speaker dataset as a single speaker dataset indices = np.array(speaker_ids) == self.speaker_id\",\n \"W def train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True):\",\n \"window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def load_checkpoint(path, model, optimizer, reset_optimizer): global global_step\",\n \"i + 1)) save_alignment(path, alignment) # Save averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\")\",\n \":], decoder_target_mask) mel_loss = (1 - w) * mel_l1_loss + w * mel_binary_div\",\n \"plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root = data_root\",\n \"multi-speaker dataset as a single speaker dataset indices = np.array(speaker_ids) == self.speaker_id paths\",\n \"losses if train_seq2seq and train_postnet: loss = mel_loss + linear_loss + done_loss elif\",\n \"downsample_step == 0 # Set 0 for zero beginning padding # imitates initial\",\n \"self.Y[idx] def __len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len is None: max_len\",\n \"None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path, model, optimizer,\",\n \"exist_ok=True) # hard coded speaker_ids = [0, 1, 10] if ismultispeaker else [None]\",\n \"import Variable from torch import nn from torch import optim import torch.backends.cudnn as\",\n \"save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step > 0 and\",\n \"kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward,\",\n \"model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\",\",\n \"of data in case for multi-speaker datasets. -h, --help Show this help message\",\n \"leaders at the Group of 20 conference.\\\", \\\"Generative adversarial network or variational auto-encoder.\\\",\",\n \"spectrogram if mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target\",\n \"# Downsample mel spectrogram if downsample_step > 1: mel = mel[:, 0::downsample_step, :].contiguous()\",\n \"from dv3.hparams import hparams, hparams_debug_string fs = hparams.sample_rate global_step = 0 global_epoch =\",\n \"1: # spectrogram-domain mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask\",\n \"features. --checkpoint-dir=<dir> Directory where to save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters\",\n \"if speaker_id is not None else \\\"single\\\" for idx, text in enumerate(texts): signal,\",\n \"import dirname, join from tqdm import tqdm, trange from datetime import datetime #\",\n \"import SummaryWriter from matplotlib import cm from warnings import warn from dv3.hparams import\",\n \"max_target_len, g).T return W def train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None,\",\n \"embedding from checkpoint. --speaker-id=<N> Use specific speaker of data in case for multi-speaker\",\n \"this as a miracle without any physical explanation.\\\", ] import dv3.synthesis synthesis._frontend =\",\n \"attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else soft_mask attn_loss\",\n \"float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if\",\n \"from matplotlib import cm from warnings import warn from dv3.hparams import hparams, hparams_debug_string\",\n \"batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in batch])\",\n \"os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path, alignment) #\",\n \"optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model): state\",\n \"sequence_length.data.max() batch_size = sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand\",\n \"load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding if load_embedding is not None: print(\\\"Loading\",\n \"model. --train-postnet-only Train only postnet model. --restore-parts=<path> Restore part of the model. --log-event-path=<name>\",\n \"Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand\",\n \"masked_mean(y, mask): # (B, T, D) mask_ = mask.expand_as(y) return (y * mask_).sum()\",\n \"_frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class\",\n \"batch_group_size = self.batch_group_size s, e = 0, 0 for i in range(len(indices) //\",\n \"def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] #\",\n \"dataset indices = np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) #\",\n \"speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id # using multi-speaker\",\n \"(B, T, 1) if mask is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B,\",\n \"set later def _pad(seq, max_len, constant_values=0): return np.pad(seq, (0, max_len - len(seq)), mode='constant',\",\n \"downsample_step > 1: mel = mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths = input_lengths.long().numpy()\",\n \"is not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding if load_embedding is\",\n \"ismultispeaker) # Update loss.backward() if clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh)\",\n \"== 0 self.permutate = permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size\",\n \"spectrogram if mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted\",\n \"global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step += 1 running_loss += loss.data[0] averaged_loss =\",\n \"hparams.eval_interval == 0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) # Update loss.backward() if clip_thresh\",\n \"text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet: y = y.cuda() mel = mel.cuda() done,\",\n \"averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss /\",\n \"data as data_utils from torch.autograd import Variable from torch import nn from torch\",\n \"-y * y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w > 0: binary_div =\",\n \"priority_loss = l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) l1_loss = (1 - priority_w)\",\n \"# Combine losses if train_seq2seq and train_postnet: loss = mel_loss + linear_loss +\",\n \"global_step) # Predicted spectrogram if linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram\",\n \"= torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r // downsample_step # frame positions s,\",\n \"y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step\",\n \"= prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) # Predicted audio signal signal =\",\n \"k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command\",\n \"/ (fs * 0.5) * linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r, :],\",\n \"dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead,\",\n \"target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) * linear_l1_loss + w *\",\n \"model.state_dict() valid_state_dict = {k: v for k, v in state.items() if k in\",\n \"not None: xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png')\",\n \"mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker else None\",\n \"return W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B = len(input_lengths) max_input_len = input_lengths.max()\",\n \"checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda() linear_dim =\",\n \"> 0: # decoder output domain mask decoder_target_mask = sequence_mask( target_lengths / (r\",\n \"to measure the acute emotional intelligence that has never gone out of style.\\\",\",\n \"logit(y_hat) z = -y * y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w >\",\n \") return model def load_checkpoint(path, model, optimizer, reset_optimizer): global global_step global global_epoch print(\\\"Load\",\n \"for zero beginning padding # imitates initial decoder states b_pad = r max_target_len\",\n \"len(input_lengths) max_input_len = input_lengths.max() W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b in\",\n \"from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule import torch from torch.utils\",\n \"not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path, model,\",\n \"- 1) input_length = input_lengths[idx] # Alignment # Multi-hop attention if attn is\",\n \"global_step += 1 running_loss += loss.data[0] averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per\",\n \"hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step,\",\n \"Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio path =\",\n \"+ 1)) save_alignment(path, alignment) # Save averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir,\",\n \"cm from warnings import warn from dv3.hparams import hparams, hparams_debug_string fs = hparams.sample_rate\",\n \"model from checkpoint path. --checkpoint-postnet=<path> Restore postnet model from checkpoint path. --train-seq2seq-only Train\",\n \"batch_group_size assert batch_group_size % batch_size == 0 self.permutate = permutate def __iter__(self): indices\",\n \"0 self.permutate = permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size s,\",\n \"= mel_loss + done_loss elif train_postnet: loss = linear_loss # attention if train_seq2seq\",\n \"is not None l1_loss = w * masked_l1(y_hat, y, mask=mask) + (1 -\",\n \"hparams.downsample_step current_lr = init_lr binary_criterion = nn.BCELoss() assert train_seq2seq or train_postnet global global_step,\",\n \"linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels,\",\n \"not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step)\",\n \"= data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model =\",\n \"= alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step)\",\n \"in case for multi-speaker datasets. -h, --help Show this help message and exit\",\n \"> 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step)\",\n \"1 # if b_pad > 0: # s, e = s - 1,\",\n \"print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" # Presets if hparams.preset is not None and\",\n \"matplotlib import pyplot as plt import sys import os from tensorboardX import SummaryWriter\",\n \"cast numpy.int64 to int self.frame_lengths = list(map(int, self.frame_lengths)) return paths def collect_features(self, path):\",\n \"g=0.2): B = len(input_lengths) max_input_len = input_lengths.max() W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32)\",\n \"attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet)\",\n \"self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None): if lengths is\",\n \"measure the acute emotional intelligence that has never gone out of style.\\\", \\\"President\",\n \"TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids = None self.multi_speaker =\",\n \"torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict = {k: v for k, v in state.items()\",\n \"train_seq2seq and train_postnet: suffix = \\\"\\\" m = model elif train_seq2seq: suffix =\",\n \"model def load_checkpoint(path, model, optimizer, reset_optimizer): global global_step global global_epoch print(\\\"Load checkpoint from:\",\n \"def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X, Mel, Y): self.X =\",\n \":priority_bin], y[:, :, :priority_bin]) else: priority_loss = l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin])\",\n \"g * g)) return W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B = len(input_lengths)\",\n \"style.\\\", \\\"President Trump met with other leaders at the Group of 20 conference.\\\",\",\n \":] target_mask = target_mask[:, r:, :] else: decoder_target_mask, target_mask = None, None #\",\n \"done: if train_seq2seq: done_loss = binary_criterion(done_hat, done) # linear: if train_postnet: n_priority_freq =\",\n \"for param_group in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() # Used for Position encoding\",\n \"checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab,\",\n \"= np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len,\",\n \"else: text = args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32),\",\n \"decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection,\",\n \"elif train_seq2seq: loss = mel_loss + done_loss elif train_postnet: loss = linear_loss #\",\n \"+ linear_loss + done_loss elif train_seq2seq: loss = mel_loss + done_loss elif train_postnet:\",\n \"max_input_len), dtype=np.float32) for b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T\",\n \"= join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as f: lines = f.readlines() l =\",\n \"= frame_positions.cuda() if train_postnet: y = y.cuda() mel = mel.cuda() done, target_lengths =\",\n \"soft_mask.cuda() if use_cuda else soft_mask attn_loss = (attn * soft_mask).mean() loss += attn_loss\",\n \"= mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker else\",\n \"\\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part of the\",\n \"W = np.zeros((max_N, max_T), dtype=np.float32) for n in range(N): for t in range(T):\",\n \"= np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths =\",\n \"max_input_len = input_lengths.max() W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b in range(B):\",\n \"is None l1_loss = l1(y_hat, y) # Priority L1 loss if priority_bin is\",\n \"- max_target_len % r assert max_target_len % r == 0 if max_target_len %\",\n \"mel_binary_div # done: if train_seq2seq: done_loss = binary_criterion(done_hat, done) # linear: if train_postnet:\",\n \"lengths=None, mask=None, max_len=None): if lengths is None and mask is None: raise RuntimeError(\\\"Should\",\n \"optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\",\",\n \"\\\\ + (1 - w) * l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) else:\",\n \"coded texts = [ \\\"Scientists at the CERN laboratory say they have discovered\",\n \"from numba import jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import join,\",\n \"sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float()\",\n \"mask=mask) \\\\ + (1 - w) * l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin])\",\n \"Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step)\",\n \"if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\",\",\n \"* mel_l1_loss + w * mel_binary_div # done: if train_seq2seq: done_loss = binary_criterion(done_hat,\",\n \"print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary writer for tensorboard if\",\n \"torch.log(1 - x + eps) def masked_mean(y, mask): # (B, T, D) mask_\",\n \"0: max_target_len += downsample_step - max_target_len % downsample_step assert max_target_len % downsample_step ==\",\n \"if ismultispeaker else None # Create mask if we use masked loss if\",\n \"N - t / T)**2 / (2 * g * g)) return W\",\n \"import torch.backends.cudnn as cudnn from torch.utils import data as data_utils from torch.utils.data.sampler import\",\n \"0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend = None #\",\n \"int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root,\",\n \"if self.multi_speaker: text, speaker_id = self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else: return\",\n \"assert hparams.name == \\\"deepvoice3\\\" # Presets if hparams.preset is not None and hparams.preset\",\n \"i in range(len(indices) // batch_group_size): s = i * batch_group_size e = s\",\n \"checkpoint_interval == 0: save_states( global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir)\",\n \"= optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step,\",\n \"given. --checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint path. --checkpoint-postnet=<path> Restore postnet model from\",\n \"auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have accepted this as a miracle without any\",\n \"event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader, optimizer, writer,\",\n \"max_N, T, max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32) for n in range(N):\",\n \"(fs * 0.5) * linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r, :], y[:,\",\n \"logit(x, eps=1e-8): return torch.log(x + eps) - torch.log(1 - x + eps) def\",\n \"np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) y_batch = torch.FloatTensor(c) # text\",\n \"0)], mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment, path, info=None): fig, ax = plt.subplots()\",\n \"max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32) for n in range(N): for t\",\n \"nn.BCELoss() assert train_seq2seq or train_postnet global global_step, global_epoch while global_epoch < nepochs: running_loss\",\n \"in state.items() if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args\",\n \"+= downsample_step - max_target_len % downsample_step assert max_target_len % downsample_step == 0 #\",\n \"to int self.frame_lengths = list(map(int, self.frame_lengths)) return paths def collect_features(self, path): return np.load(path)\",\n \"ax = plt.subplots() im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel\",\n \"plot_alignment(alignment, path, info=None): fig, ax = plt.subplots() im = ax.imshow( alignment, aspect='auto', origin='lower',\",\n \"5 self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col],\",\n \"/ (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip against freq axis\",\n \"save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and train_postnet: suffix =\",\n \"lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) ==\",\n \"multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in batch]) else: speaker_ids = None return\",\n \":, :priority_bin], y[:, :, :priority_bin], mask=mask) \\\\ + (1 - w) * l1(y_hat[:,\",\n \"lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in\",\n \"= w * masked_l1(y_hat, y, mask=mask) + (1 - w) * l1(y_hat, y)\",\n \"None linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat = None, None, None # Losses\",\n \"global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\",\",\n \"elif train_postnet: assert speaker_ids is None linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat =\",\n \"elif train_seq2seq: suffix = \\\"_seq2seq\\\" m = model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\"\",\n \"mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight #\",\n \"masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight # L1 loss if\",\n \"if global_step > 0 and global_step % hparams.eval_interval == 0: eval_model(global_step, writer, model,\",\n \"mel_loss + linear_loss + done_loss elif train_seq2seq: loss = mel_loss + done_loss elif\",\n \"Filter by speaker_id # using multi-speaker dataset as a single speaker dataset if\",\n \"0::downsample_step, :].contiguous() # Lengths input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r //\",\n \"mel_outputs, linear_outputs, attn, done_hat = model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif\",\n \"eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if\",\n \"int(speaker_id) if speaker_id is not None else None data_root = args[\\\"--data-root\\\"] if data_root\",\n \"hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z = -y\",\n \"args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train both if not specified if not train_seq2seq\",\n \"b_pad = r max_target_len += b_pad * downsample_step a = np.array([_pad(x[0], max_input_len) for\",\n \"step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and train_postnet: suffix = \\\"\\\" m\",\n \"optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts,\",\n \"param_group['lr'] = current_lr optimizer.zero_grad() # Used for Position encoding text_positions, frame_positions = positions\",\n \"zero beginning padding # imitates initial decoder states b_pad = r max_target_len +=\",\n \"+= 1 running_loss += loss.data[0] averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\",\",\n \"with other leaders at the Group of 20 conference.\\\", \\\"Generative adversarial network or\",\n \"- w) * linear_l1_loss + w * linear_binary_div # Combine losses if train_seq2seq\",\n \"model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else\",\n \"// downsample_step - 1), max_decoder_target_len, constant_values=1) for x in batch]) done = torch.FloatTensor(done).unsqueeze(-1)\",\n \"model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1))\",\n \"= None self.multi_speaker = False # If not None, filter by speaker_id self.speaker_id\",\n \"binary_criterion(done_hat, done) # linear: if train_postnet: n_priority_freq = int(hparams.priority_freq / (fs * 0.5)\",\n \"self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id # using\",\n \"only postnet model. --restore-parts=<path> Restore part of the model. --log-event-path=<name> Log event path.\",\n \"seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1)\",\n \"- max_target_len % downsample_step assert max_target_len % downsample_step == 0 # Set 0\",\n \"global_step) # Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try:\",\n \":-r, :], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w)\",\n \"mask=None, max_len=None): if lengths is None and mask is None: raise RuntimeError(\\\"Should provide\",\n \"rate schedule if hparams.lr_schedule is not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr =\",\n \"torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1]) // r //\",\n \"checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None\",\n \"= batch_size if batch_group_size is None: batch_group_size = min(batch_size * 32, len(self.lengths)) if\",\n \"if attn is not None and attn.dim() == 4: for i, alignment in\",\n \"// self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last elements s\",\n \"optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\":\",\n \"Binary divergence loss if hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits =\",\n \"it 3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices\",\n \"p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource):\",\n \"for seq2seq text-to-speech synthesis model. usage: train.py [options] options: --data-root=<dir> Directory contains preprocessed\",\n \"= hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion = nn.BCELoss() assert train_seq2seq\",\n \"= \\\"_postnet\\\" m = model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state =\",\n \"D) mask_ = mask.expand_as(y) return (y * mask_).sum() / mask_.sum() def spec_loss(y_hat, y,\",\n \"batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model = build_model() if use_cuda:\",\n \"try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except\",\n \"mel_l1_loss + w * mel_binary_div # done: if train_seq2seq: done_loss = binary_criterion(done_hat, done)\",\n \"max_len = sequence_length.data.max() batch_size = sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size,\",\n \"return loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step =\",\n \"= done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker else None # Create mask\",\n \"ismultispeaker else None if use_cuda: if train_seq2seq: x = x.cuda() text_positions = text_positions.cuda()\",\n \"model, text, p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) # Alignment path = join(eval_output_dir,\",\n \"global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio\",\n \"# harded coded texts = [ \\\"Scientists at the CERN laboratory say they\",\n \"# Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval)\",\n \"path. --checkpoint-postnet=<path> Restore postnet model from checkpoint path. --train-seq2seq-only Train only seq2seq model.\",\n \"torch.log(1 + torch.exp(y_hat_logits)) if w > 0: binary_div = w * masked_mean(z, mask)\",\n \"= model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is\",\n \"= args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id is not None else None data_root\",\n \"speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None else \\\"single\\\"\",\n \"--data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir> Directory where to save model checkpoints [default:\",\n \"exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path, alignment) # Save\",\n \"None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet,\",\n \"import data as data_utils from torch.autograd import Variable from torch import nn from\",\n \"dv3.lrschedule import torch from torch.utils import data as data_utils from torch.autograd import Variable\",\n \"train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask\",\n \"= 'Decoder timestep' if info is not None: xlabel += '\\\\n\\\\n' + info\",\n \"random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last elements s += batch_group_size\",\n \"is None mel_outputs, attn, done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths)\",\n \"for multi-speaker datasets. -h, --help Show this help message and exit \\\"\\\"\\\" from\",\n \"self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts return texts, speaker_ids else:\",\n \"% r == 0 if max_target_len % downsample_step != 0: max_target_len += downsample_step\",\n \"text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None\",\n \"speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None else \\\"single\\\" for idx,\",\n \"= hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :],\",\n \"> 0: binary_div = w * masked_mean(z, mask) + (1 - w) *\",\n \"model, checkpoint_dir, ismultispeaker) # Update loss.backward() if clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm(\",\n \"if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in batch]) else: speaker_ids = None\",\n \"l1_loss = l1(y_hat, y) # Priority L1 loss if priority_bin is not None\",\n \"= list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths\",\n \"hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" # Presets if hparams.preset is not None\",\n \"is not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs)\",\n \"else [None] for speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not\",\n \"if multi_speaker and self.speaker_id is not None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines))\",\n \"3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices =\",\n \"speaker_id is not None else None data_root = args[\\\"--data-root\\\"] if data_root is None:\",\n \"any physical explanation.\\\", ] import dv3.synthesis synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\")\",\n \"= y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8):\",\n \"m = model elif train_seq2seq: suffix = \\\"_seq2seq\\\" m = model.seq2seq elif train_postnet:\",\n \"priority_loss # Binary divergence loss if hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_()) else:\",\n \"input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat, _ = model.seq2seq(\",\n \"1 - np.exp(-(n / N - t / T)**2 / (2 * g\",\n \"Trump met with other leaders at the Group of 20 conference.\\\", \\\"Generative adversarial\",\n \"timestep' if info is not None: xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder\",\n \"target_lengths = [len(x[1]) for x in batch] max_target_len = max(target_lengths) if max_target_len %\",\n \"def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B = len(input_lengths) max_input_len = input_lengths.max() W =\",\n \"linear_outputs = None elif train_postnet: assert speaker_ids is None linear_outputs = model.postnet(mel) mel_outputs,\",\n \"mask_, target * mask_) return loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r\",\n \"preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) #\",\n \"len(batch[0]) == 4 # Lengths input_lengths = [len(x[0]) for x in batch] max_input_len\",\n \"constant_values=1) for x in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3]\",\n \"not None else None data_root = args[\\\"--data-root\\\"] if data_root is None: data_root =\",\n \"_NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None): self.data_root = data_root self.col = col self.frame_lengths\",\n \"*args): if self.multi_speaker: text, speaker_id = args else: text = args[0] seq =\",\n \"# imitates initial decoder states b_pad = r max_target_len += b_pad * downsample_step\",\n \"r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) * linear_l1_loss +\",\n \"assert mask is None l1_loss = l1(y_hat, y) # Priority L1 loss if\",\n \"use masked loss if hparams.masked_loss_weight > 0: # decoder output domain mask decoder_target_mask\",\n \"None: xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close()\",\n \"batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm)\",\n \"1, 10] if ismultispeaker else [None] for speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id)\",\n \"linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) # Predicted audio signal\",\n \"- t / T)**2 / (2 * g * g)) return W def\",\n \"> 0: priority_loss = w * masked_l1( y_hat[:, :, :priority_bin], y[:, :, :priority_bin],\",\n \"print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training postnet model\\\") else: assert False, \\\"must be\",\n \"0: if w > 0: priority_loss = w * masked_l1( y_hat[:, :, :priority_bin],\",\n \"= prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) # Target spectrogram if linear_outputs is\",\n \"# Learning rate schedule if hparams.lr_schedule is not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule)\",\n \"writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh\",\n \"model from checkpoint path if given. --checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint path.\",\n \"writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx,\",\n \"None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved\",\n \"args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"]\",\n \"text, speaker_id = args else: text = args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if\",\n \"def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root,\",\n \"w) * z.mean() else: binary_div = z.mean() return l1_loss, binary_div @jit(nopython=True) def guided_attention(N,\",\n \"freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def load_checkpoint(path, model, optimizer, reset_optimizer):\",\n \"20 conference.\\\", \\\"Generative adversarial network or variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have\",\n \"def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size\",\n \"= False _frontend = None # to be set later def _pad(seq, max_len,\",\n \"l1(y_hat, y) else: assert mask is None l1_loss = l1(y_hat, y) # Priority\",\n \"# spectrogram-domain mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask #\",\n \"- len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x = np.pad(x, [(b_pad, max_len\",\n \"plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root\",\n \"if ismultispeaker else [None] for speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id\",\n \"in range(N): for t in range(T): W[n, t] = 1 - np.exp(-(n /\",\n \"random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e]\",\n \"norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step += 1 running_loss += loss.data[0]\",\n \"range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W def train(model, data_loader,\",\n \"- w) * l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) else: priority_loss = l1(y_hat[:,\",\n \":, :priority_bin]) else: priority_loss = l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) l1_loss =\",\n \"mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask # shift mask\",\n \"train both if not specified if not train_seq2seq and not train_postnet: print(\\\"Training whole\",\n \"input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step >\",\n \"without any physical explanation.\\\", ] import dv3.synthesis synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir,\",\n \"Sort by lengths 2. Pick a small patch and randomize it 3. Permutate\",\n \"optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is not None: print(\\\"Load optimizer state from {}\\\".format(path))\",\n \"r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions,\",\n \"from checkpoint path. --checkpoint-postnet=<path> Restore postnet model from checkpoint path. --train-seq2seq-only Train only\",\n \"writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step += 1 running_loss += loss.data[0] averaged_loss = running_loss\",\n \"= col self.frame_lengths = [] self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root,\",\n \"exit \\\"\\\"\\\" from docopt import docopt import sys from os.path import dirname, join\",\n \"= hparams.sample_rate global_step = 0 global_epoch = 0 use_cuda = torch.cuda.is_available() if use_cuda:\",\n \"{}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict = {k: v for k,\",\n \"torch.utils import data as data_utils from torch.autograd import Variable from torch import nn\",\n \"= list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines))\",\n \"= speaker_ids is not None # Learning rate schedule if hparams.lr_schedule is not\",\n \"= nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None): if lengths is None\",\n \"speaker_ids is None mel_outputs, attn, done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions,\",\n \"FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size)\",\n \"if train_seq2seq: x = x.cuda() text_positions = text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet:\",\n \"checkpoint[\\\"optimizer\\\"] if optimizer_state is not None: print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step\",\n \"assert batch_group_size % batch_size == 0 self.permutate = permutate def __iter__(self): indices =\",\n \"* linear_l1_loss + w * linear_binary_div # Combine losses if train_seq2seq and train_postnet:\",\n \"checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is not None:\",\n \"texts = [ \\\"Scientists at the CERN laboratory say they have discovered a\",\n \"= MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight # L1 loss if w\",\n \"max_decoder_target_len = max_target_len // r // downsample_step # frame positions s, e =\",\n \"# The deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule\",\n \"{}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def save_states(global_step,\",\n \"5 multi_speaker = len(l) == 5 self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines))\",\n \"model) # Setup summary writer for tensorboard if log_event_path is None: log_event_path =\",\n \"max_decoder_target_len, constant_values=1) for x in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids =\",\n \"None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) #\",\n \"global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if\",\n \"global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer:\",\n \"eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): # harded coded texts = [ \\\"Scientists at\",\n \"+ 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # save files as well\",\n \"None # Learning rate schedule if hparams.lr_schedule is not None: lr_schedule_f = getattr(dv3.lrschedule,\",\n \"global_step) global_step += 1 running_loss += loss.data[0] averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss\",\n \"nn.L1Loss() w = hparams.masked_loss_weight # L1 loss if w > 0: assert mask\",\n \"e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1]) // r // downsample_step\",\n \"checkpoint path if given. --checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint path. --checkpoint-postnet=<path> Restore\",\n \"reset_optimizer) # Load embedding if load_embedding is not None: print(\\\"Loading embedding from {}\\\".format(load_embedding))\",\n \"x + eps) def masked_mean(y, mask): # (B, T, D) mask_ = mask.expand_as(y)\",\n \"\\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str),\",\n \"2. Pick a small patch and randomize it 3. Permutate mini-batchs \\\"\\\"\\\" def\",\n \"downsample_step assert max_target_len % downsample_step == 0 # Set 0 for zero beginning\",\n \"/ (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain mask target_mask\",\n \"Combine losses if train_seq2seq and train_postnet: loss = mel_loss + linear_loss + done_loss\",\n \"list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id # using multi-speaker dataset as\",\n \"* mask_).sum() / mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss()\",\n \"y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) * linear_l1_loss\",\n \"(B, T, D) mask_ = mask.expand_as(input) loss = self.criterion(input * mask_, target *\",\n \"from warnings import warn from dv3.hparams import hparams, hparams_debug_string fs = hparams.sample_rate global_step\",\n \"max_len=None): if max_len is None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0) seq_range =\",\n \"Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" # Presets if hparams.preset\",\n \"is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input)\",\n \"# alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text, speaker_id =\",\n \"__len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len is None: max_len = sequence_length.data.max()\",\n \"# Predicted mel spectrogram if mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output\",\n \"if optimizer_state is not None: print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step =\",\n \"speaker_ids = speaker_ids.cuda() if ismultispeaker else None # Create mask if we use\",\n \"Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z = -y * y_hat_logits + torch.log(1 +\",\n \"None, None # Losses w = hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss, mel_binary_div\",\n \"np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str),\",\n \"librosa.display from matplotlib import pyplot as plt import sys import os from tensorboardX\",\n \"permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size is None: batch_group_size\",\n \"= nn.L1Loss() w = hparams.masked_loss_weight # L1 loss if w > 0: assert\",\n \"done = np.array([_pad(np.zeros(len(x[1]) // r // downsample_step - 1), max_decoder_target_len, constant_values=1) for x\",\n \"global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step)\",\n \"- 1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done\",\n \"+ 1 # if b_pad > 0: # s, e = s -\",\n \"suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state,\",\n \"+ eps) def masked_mean(y, mask): # (B, T, D) mask_ = mask.expand_as(y) return\",\n \"model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if\",\n \"255) def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): # harded coded texts = [\",\n \"acute emotional intelligence that has never gone out of style.\\\", \\\"President Trump met\",\n \"= mask.expand_as(input) loss = self.criterion(input * mask_, target * mask_) return loss /\",\n \"not None: indices = np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False\",\n \"= Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return\",\n \"Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker else None if use_cuda: if train_seq2seq: x\",\n \"if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not\",\n \"multi_speaker = len(l) == 5 self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths\",\n \"checkpoint:\\\", checkpoint_path) def build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels,\",\n \"data_root, speaker_id=None): self.data_root = data_root self.speaker_ids = None self.multi_speaker = False # If\",\n \"step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram -\",\n \"os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids = [0, 1, 10] if ismultispeaker else\",\n \"/ mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1 =\",\n \"map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda\",\n \"loss = self.criterion(input * mask_, target * mask_) return loss / mask_.sum() def\",\n \"None return x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids def\",\n \"speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None else \\\"single\\\" for idx, text\",\n \"hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model) # Load\",\n \"(2 * g * g)) return W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B\",\n \"optimizer_state is not None: print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"]\",\n \"checkpoint_dir=None): print(\\\"Save intermediate states at step {}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx\",\n \"clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda() linear_dim = model.linear_dim r =\",\n \"--train-seq2seq-only Train only seq2seq model. --train-postnet-only Train only postnet model. --restore-parts=<path> Restore part\",\n \"\\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) #\",\n \"0 if max_target_len % downsample_step != 0: max_target_len += downsample_step - max_target_len %\",\n \"return torch.log(x + eps) - torch.log(1 - x + eps) def masked_mean(y, mask):\",\n \"checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts\",\n \"patch and randomize it 3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None,\",\n \"mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) # Target\",\n \"model, optimizer, reset_optimizer) # Load embedding if load_embedding is not None: print(\\\"Loading embedding\",\n \"and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask =\",\n \"/ (len(data_loader)))) global_epoch += 1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet):\",\n \"texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]),\",\n \"particle.\\\", \\\"There's a way to measure the acute emotional intelligence that has never\",\n \"key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part\",\n \"for x in batch] max_input_len = max(input_lengths) target_lengths = [len(x[1]) for x in\",\n \"init_lr binary_criterion = nn.BCELoss() assert train_seq2seq or train_postnet global global_step, global_epoch while global_epoch\",\n \"@jit(nopython=True) def guided_attention(N, max_N, T, max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32) for\",\n \"if speaker_id is not None else None data_root = args[\\\"--data-root\\\"] if data_root is\",\n \"optim import torch.backends.cudnn as cudnn from torch.utils import data as data_utils from torch.utils.data.sampler\",\n \"1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class\",\n \"sequence_mask(sequence_length, max_len=None): if max_len is None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0) seq_range\",\n \"preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format(\",\n \"speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths\",\n \"len(l) == 5 multi_speaker = len(l) == 5 self.frame_lengths = list( map(lambda l:\",\n \"writer, model, checkpoint_dir, ismultispeaker): # harded coded texts = [ \\\"Scientists at the\",\n \"mask): # (B, T, D) mask_ = mask.expand_as(y) return (y * mask_).sum() /\",\n \"--checkpoint=<path> Restore model from checkpoint path if given. --checkpoint-seq2seq=<path> Restore seq2seq model from\",\n \"train_seq2seq, train_postnet): if train_seq2seq and train_postnet: suffix = \\\"\\\" m = model elif\",\n \"warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save\",\n \"train.py [options] options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir> Directory where to save\",\n \"\\\"There's a way to measure the acute emotional intelligence that has never gone\",\n \"% hparams.eval_interval == 0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) # Update loss.backward() if\",\n \"at step {}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) -\",\n \"plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids\",\n \"// batch_group_size): s = i * batch_group_size e = s + batch_group_size random.shuffle(indices[s:e])\",\n \"tqdm, trange from datetime import datetime # The deepvoice3 model from dv3.deepvoice3_pytorch import\",\n \"{}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state\",\n \"model from checkpoint path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only Train only postnet\",\n \"writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step += 1 running_loss +=\",\n \"linear_l1_loss + w * linear_binary_div # Combine losses if train_seq2seq and train_postnet: loss\",\n \"max_T), dtype=np.float32) for n in range(N): for t in range(T): W[n, t] =\",\n \"(np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip against freq axis return\",\n \"self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64\",\n \"freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): #\",\n \"= None elif train_postnet: assert speaker_ids is None linear_outputs = model.postnet(mel) mel_outputs, attn,\",\n \"model. --restore-parts=<path> Restore part of the model. --log-event-path=<name> Log event path. --reset-optimizer Reset\",\n \"len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x = np.pad(x, [(b_pad, max_len -\",\n \"done_hat = None, None, None # Losses w = hparams.binary_divergence_weight # mel: if\",\n \"batch]) else: speaker_ids = None return x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions),\",\n \"for x in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths =\",\n \"suffix = \\\"_postnet\\\" m = model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state\",\n \"= Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else soft_mask attn_loss = (attn *\",\n \"y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8): return\",\n \"spectrogram if linear_outputs is not None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target\",\n \"--checkpoint-postnet=<path> Restore postnet model from checkpoint path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only\",\n \"decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else soft_mask\",\n \"train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm,\",\n \"if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is not\",\n \"alignment) # Save averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path =\",\n \"checkpoint. --speaker-id=<N> Use specific speaker of data in case for multi-speaker datasets. -h,\",\n \"else: target_mask = decoder_target_mask # shift mask decoder_target_mask = decoder_target_mask[:, r:, :] target_mask\",\n \"while global_epoch < nepochs: running_loss = 0. for step, (x, input_lengths, mel, y,\",\n \"mask is None l1_loss = l1(y_hat, y) # Priority L1 loss if priority_bin\",\n \"train_postnet: assert speaker_ids is None linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat = None,\",\n \"be set later def _pad(seq, max_len, constant_values=0): return np.pad(seq, (0, max_len - len(seq)),\",\n \"paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64 to\",\n \"synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) # Alignment path =\",\n \"Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler(\",\n \"path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str)\",\n \"state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model def\",\n \"batch_group_size is None: batch_group_size = min(batch_size * 32, len(self.lengths)) if batch_group_size % batch_size\",\n \"input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at step {}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths))\",\n \"decoder_target_mask[:, r:, :] target_mask = target_mask[:, r:, :] else: decoder_target_mask, target_mask = None,\",\n \"alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step,\",\n \"downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain mask target_mask = sequence_mask( target_lengths,\",\n \"b_pad * downsample_step a = np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch\",\n \"self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size is None: batch_group_size =\",\n \"text_positions, frame_positions = positions # Downsample mel spectrogram if downsample_step > 1: mel\",\n \"batch_size == 0 self.permutate = permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size =\",\n \"spectrogram if linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted\",\n \"# Load checkpoints if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if\",\n \"len(l) == 5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids =\",\n \"constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x = np.pad(x, [(b_pad, max_len - len(x) -\",\n \"np.flip(spectrogram, axis=1) # flip against freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step,\",\n \"torch.exp(y_hat_logits)) if w > 0: binary_div = w * masked_mean(z, mask) + (1\",\n \"(0, max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x = np.pad(x,\",\n \"mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step,\",\n \"X, Mel, Y): self.X = X self.Mel = Mel self.Y = Y #\",\n \"input_lengths, mel, y, positions, done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker =\",\n \"dv3.audio import dv3.lrschedule import torch from torch.utils import data as data_utils from torch.autograd\",\n \"axis=1) # flip against freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer,\",\n \"frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None elif\",\n \"self.data_root = data_root self.speaker_ids = None self.multi_speaker = False # If not None,\",\n \"path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag,\",\n \"= sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask # shift mask decoder_target_mask =\",\n \"self.Y = Y # alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker:\",\n \"1), max_input_len) for x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len\",\n \"a small patch and randomize it 3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths,\",\n \"global_epoch < nepochs: running_loss = 0. for step, (x, input_lengths, mel, y, positions,\",\n \"target_mask = None, None # Apply model if train_seq2seq and train_postnet: mel_outputs, linear_outputs,\",\n \"done_loss elif train_postnet: loss = linear_loss # attention if train_seq2seq and hparams.use_guided_attention: soft_mask\",\n \"model) # Load checkpoints if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer)\",\n \"checkpoints]. --hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path> Restore model from checkpoint path if\",\n \"Load checkpoints if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path\",\n \"paths)) if multi_speaker and self.speaker_id is not None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]),\",\n \"= [len(x[1]) for x in batch] max_target_len = max(target_lengths) if max_target_len % r\",\n \"0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort by lengths 2. Pick\",\n \"linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) # Predicted\",\n \"from torch.utils import data as data_utils from torch.autograd import Variable from torch import\",\n \"discovered a new particle.\\\", \\\"There's a way to measure the acute emotional intelligence\",\n \"if linear_outputs is not None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear\",\n \"numpy as np from numba import jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource from\",\n \"= np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length = input_lengths[idx] #\",\n \"global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() # Used for\",\n \"init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer, global_step,\",\n \"as data_utils from torch.autograd import Variable from torch import nn from torch import\",\n \"mask.expand_as(y) return (y * mask_).sum() / mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0):\",\n \"% downsample_step == 0 # Set 0 for zero beginning padding # imitates\",\n \"cudnn.benchmark = False _frontend = None # to be set later def _pad(seq,\",\n \"current_lr, global_step) global_step += 1 running_loss += loss.data[0] averaged_loss = running_loss / (len(data_loader))\",\n \"eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids = [0, 1,\",\n \"Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target mel spectrogram if mel_outputs\",\n \"len(x[0]) + 1), max_input_len) for x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len\",\n \"met with other leaders at the Group of 20 conference.\\\", \\\"Generative adversarial network\",\n \"10] if ismultispeaker else [None] for speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if\",\n \"target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask # shift mask decoder_target_mask = decoder_target_mask[:, r:,\",\n \"speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout,\",\n \"_load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3\",\n \"join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment,\",\n \"synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids\",\n \"eps=1e-8): return torch.log(x + eps) - torch.log(1 - x + eps) def masked_mean(y,\",\n \"global_epoch += 1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq\",\n \"import warn from dv3.hparams import hparams, hparams_debug_string fs = hparams.sample_rate global_step = 0\",\n \"print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend,\",\n \"= model elif train_seq2seq: suffix = \\\"_seq2seq\\\" m = model.seq2seq elif train_postnet: suffix\",\n \"5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids = list(map(lambda l:\",\n \"use_cuda: model = model.cuda() linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step\",\n \"done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker else None # Create mask if\",\n \"attn_loss if global_step > 0 and global_step % checkpoint_interval == 0: save_states( global_step,\",\n \"global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient\",\n \"global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram)\",\n \"dv3.synthesis synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded\",\n \"= mel_loss + linear_loss + done_loss elif train_seq2seq: loss = mel_loss + done_loss\",\n \"not None: mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step)\",\n \"hparams.masked_loss_weight # L1 loss if w > 0: assert mask is not None\",\n \"soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if\",\n \"speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx,\",\n \"from torch.autograd import Variable from torch import nn from torch import optim import\",\n \"join(self.data_root, f), paths)) if multi_speaker and self.speaker_id is not None: speaker_ids = list(map(lambda\",\n \"hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr'] =\",\n \"= {k: v for k, v in state.items() if k in model_dict} model_dict.update(valid_state_dict)\",\n \"\\\"\\\" m = model elif train_seq2seq: suffix = \\\"_seq2seq\\\" m = model.seq2seq elif\",\n \"\\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker = len(batch[0]) == 4\",\n \":priority_bin], y[:, :, :priority_bin]) l1_loss = (1 - priority_w) * l1_loss + priority_w\",\n \"= indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last elements s += batch_group_size if s\",\n \"text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for x in batch], dtype=np.int) text_positions\",\n \"y, positions, done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is\",\n \"in range(len(indices) // batch_group_size): s = i * batch_group_size e = s +\",\n \"255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) #\",\n \"seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self):\",\n \"emotional intelligence that has never gone out of style.\\\", \\\"President Trump met with\",\n \"None: indices = np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return\",\n \"= model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel), -1,\",\n \"w > 0: assert mask is not None l1_loss = w * masked_l1(y_hat,\",\n \"e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done\",\n \"model elif train_seq2seq: suffix = \\\"_seq2seq\\\" m = model.seq2seq elif train_postnet: suffix =\",\n \"mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\",\",\n \"done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs =\",\n \"0.5) * linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:, :-r, :], y[:, r:, :],\",\n \"if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask)\",\n \"checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda() linear_dim = model.linear_dim\",\n \"as cudnn from torch.utils import data as data_utils from torch.utils.data.sampler import Sampler import\",\n \"exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\"\",\n \"warn from dv3.hparams import hparams, hparams_debug_string fs = hparams.sample_rate global_step = 0 global_epoch\",\n \"args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"]\",\n \"== \\\"deepvoice3\\\" # Presets if hparams.preset is not None and hparams.preset != \\\"\\\":\",\n \"import optim import torch.backends.cudnn as cudnn from torch.utils import data as data_utils from\",\n \"\\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1))\",\n \"downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std,\",\n \"(1 - priority_w) * l1_loss + priority_w * priority_loss # Binary divergence loss\",\n \"False, \\\"must be specified wrong args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert\",\n \"parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir,\",\n \"[] self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\")\",\n \"np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to\",\n \"self.multi_speaker = False # If not None, filter by speaker_id self.speaker_id = speaker_id\",\n \"= 0, 0 for i in range(len(indices) // batch_group_size): s = i *\",\n \"spec_loss( linear_outputs[:, :-r, :], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1\",\n \"and mask is None: raise RuntimeError(\\\"Should provide either lengths or mask\\\") # (B,\",\n \"for k, v in state.items() if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__\",\n \"PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort by lengths 2. Pick a small patch\",\n \"with open(meta, \\\"rb\\\") as f: lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l)\",\n \"checkpoint_dir, ismultispeaker): # harded coded texts = [ \\\"Scientists at the CERN laboratory\",\n \"Variable(text_positions) frame_positions = Variable(frame_positions) done = Variable(done) target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids)\",\n \"mel, y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if\",\n \"mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None elif train_postnet: assert speaker_ids is\",\n \"optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step > 0 and global_step %\",\n \"options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir> Directory where to save model checkpoints\",\n \"--speaker-id=<N> Use specific speaker of data in case for multi-speaker datasets. -h, --help\",\n \"len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len is None: max_len = sequence_length.data.max() batch_size =\",\n \"Alignment # Multi-hop attention if attn is not None and attn.dim() == 4:\",\n \"y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at step {}\\\".format(global_step)) # idx = np.random.randint(0,\",\n \"is not None: xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path,\",\n \"the acute emotional intelligence that has never gone out of style.\\\", \\\"President Trump\",\n \"linear_outputs[:, :-r, :], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 -\",\n \"lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model)\",\n \"running_loss += loss.data[0] averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch)\",\n \"never gone out of style.\\\", \\\"President Trump met with other leaders at the\",\n \"intermediate states at step {}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx = min(1,\",\n \"line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path\",\n \"!= \\\"\\\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset\",\n \"priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight # L1\",\n \"= sequence_length.data.max() batch_size = sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len)\",\n \"path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path, alignment) # Save averaged\",\n \"from tqdm import tqdm, trange from datetime import datetime # The deepvoice3 model\",\n \"-1, mel.size(-1)) linear_outputs = None elif train_postnet: assert speaker_ids is None linear_outputs =\",\n \"# flip against freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model,\",\n \"True, True if train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training postnet model\\\") else:\",\n \"param_group in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() # Used for Position encoding text_positions,\",\n \"None self.multi_speaker = False # If not None, filter by speaker_id self.speaker_id =\",\n \"priority_w) * l1_loss + priority_w * priority_loss # Binary divergence loss if hparams.binary_divergence_weight\",\n \"target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker else None # Create\",\n \"collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model = build_model() if use_cuda: model = model.cuda()\",\n \"x def plot_alignment(alignment, path, info=None): fig, ax = plt.subplots() im = ax.imshow( alignment,\",\n \"= ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep' if\",\n \"hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion = nn.BCELoss() assert train_seq2seq or\",\n \"+ torch.log(1 + torch.exp(y_hat_logits)) if w > 0: binary_div = w * masked_mean(z,\",\n \"mel = mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy()\",\n \"self.batch_size = batch_size if batch_group_size is None: batch_group_size = min(batch_size * 32, len(self.lengths))\",\n \"audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step))\",\n \"# Which model to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] #\",\n \"# Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset\",\n \"model.train() ismultispeaker = speaker_ids is not None # Learning rate schedule if hparams.lr_schedule\",\n \"global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step)\",\n \"len(x) - b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment, path, info=None):\",\n \"if train_seq2seq and train_postnet: loss = mel_loss + linear_loss + done_loss elif train_seq2seq:\",\n \"save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states at\",\n \"1), max_decoder_target_len, constant_values=1) for x in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids\",\n \"\\\"\\\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\":\",\n \"fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep' if info is not None: xlabel +=\",\n \"# L1 loss if w > 0: assert mask is not None l1_loss\",\n \"torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore\",\n \"constant_values=0) return x def plot_alignment(alignment, path, info=None): fig, ax = plt.subplots() im =\",\n \"prepare_spec_image(mel), global_step) # Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path)\",\n \"if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch,\",\n \"checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"]\",\n \"self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) #\",\n \"is None linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat = None, None, None #\",\n \"# (B, T, D) mask_ = mask.expand_as(y) return (y * mask_).sum() / mask_.sum()\",\n \"if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is not None: print(\\\"Load optimizer\",\n \"# Binary divergence loss if hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits\",\n \"def train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if\",\n \"attn, done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs\",\n \"0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z = -y * y_hat_logits\",\n \"# Permutate batches if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] =\",\n \"]. --checkpoint=<path> Restore model from checkpoint path if given. --checkpoint-seq2seq=<path> Restore seq2seq model\",\n \"at the Group of 20 conference.\\\", \\\"Generative adversarial network or variational auto-encoder.\\\", \\\"Please\",\n \"w > 0: priority_loss = w * masked_l1( y_hat[:, :, :priority_bin], y[:, :,\",\n \"\\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model to be trained\",\n \"return texts def collect_features(self, *args): if self.multi_speaker: text, speaker_id = args else: text\",\n \"batch_group_size if s < len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices) class\",\n \"+ done_loss elif train_postnet: loss = linear_loss # attention if train_seq2seq and hparams.use_guided_attention:\",\n \"return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False)\",\n \"l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker and\",\n \"Variable(done) target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker else None if use_cuda:\",\n \"= Variable(speaker_ids) if ismultispeaker else None if use_cuda: if train_seq2seq: x = x.cuda()\",\n \"linear_binary_div = spec_loss( linear_outputs[:, :-r, :], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss\",\n \"writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal,\",\n \"__iter__(self): indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e = 0, 0 for\",\n \"if mask is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_\",\n \"init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() # Used\",\n \":priority_bin], y[:, :, :priority_bin], mask=mask) \\\\ + (1 - w) * l1(y_hat[:, :,\",\n \"= lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad()\",\n \"Y # alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text, speaker_id\",\n \"class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially\",\n \"data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model\",\n \"max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input) loss = self.criterion(input * mask_,\",\n \"s - 1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) #\",\n \"global_step) if clip_thresh > 0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step)\",\n \"4 # Lengths input_lengths = [len(x[0]) for x in batch] max_input_len = max(input_lengths)\",\n \"= SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs,\",\n \"in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T)\",\n \"loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step = hparams.downsample_step\",\n \"mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size\",\n \"path) try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception\",\n \"from torch.utils.data.sampler import Sampler import numpy as np from numba import jit from\",\n \"dirname, join from tqdm import tqdm, trange from datetime import datetime # The\",\n \"batch], dtype=np.float32) y_batch = torch.FloatTensor(c) # text positions text_positions = np.array([_pad(np.arange(1, len(x[0]) +\",\n \"global_epoch while global_epoch < nepochs: running_loss = 0. for step, (x, input_lengths, mel,\",\n \"indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e = 0, 0 for i\",\n \"if downsample_step > 1: # spectrogram-domain mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else:\",\n \"max_target_len += r - max_target_len % r assert max_target_len % r == 0\",\n \"target_lengths[b], max_target_len, g).T return W def train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None,\",\n \"* 255), global_step) # Predicted mel spectrogram if mel_outputs is not None: mel_output\",\n \"is not None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id\",\n \"alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Mel\",\n \"model if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn, done_hat = model( x, mel,\",\n \"not None: print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch =\",\n \"None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer,\",\n \"self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last elements s +=\",\n \"log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model to be trained train_seq2seq\",\n \"= (1 - w) * linear_l1_loss + w * linear_binary_div # Combine losses\",\n \"// r // downsample_step # frame positions s, e = 1, max_decoder_target_len +\",\n \"= True, True if train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training postnet model\\\")\",\n \"specified if not train_seq2seq and not train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet =\",\n \"checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and train_postnet: suffix = \\\"\\\" m =\",\n \"speaker_ids = Variable(speaker_ids) if ismultispeaker else None if use_cuda: if train_seq2seq: x =\",\n \"contains preprocessed features. --checkpoint-dir=<dir> Directory where to save model checkpoints [default: checkpoints]. --hparams=<parmas>\",\n \"signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try:\",\n \"dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in batch],\",\n \"target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{},\",\n \"optimizer, reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load\",\n \"return (y * mask_).sum() / mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1\",\n \"a single speaker dataset indices = np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths\",\n \"join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment)\",\n \"global_step = 0 global_epoch = 0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark =\",\n \"batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size is\",\n \"if use_cuda: model = model.cuda() linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step =\",\n \"= (1 - priority_w) * l1_loss + priority_w * priority_loss # Binary divergence\",\n \"\\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\":\",\n \"b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment, path, info=None): fig, ax\",\n \"if train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training postnet model\\\") else: assert False,\",\n \"say they have discovered a new particle.\\\", \\\"There's a way to measure the\",\n \"clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet)\",\n \"\\\"Generative adversarial network or variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have accepted this\",\n \"done flags done = np.array([_pad(np.zeros(len(x[1]) // r // downsample_step - 1), max_decoder_target_len, constant_values=1)\",\n \"None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) #\",\n \"done) # linear: if train_postnet: n_priority_freq = int(hparams.priority_freq / (fs * 0.5) *\",\n \"= list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker and self.speaker_id is not None:\",\n \"restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer,\",\n \"n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size,\",\n \"self.col = col self.frame_lengths = [] self.speaker_id = speaker_id def collect_files(self): meta =\",\n \"seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class\",\n \"in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is not None # Learning rate schedule\",\n \"is not None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary writer\",\n \"writer for tensorboard if log_event_path is None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\",\",\n \"Filter by speaker_id # using multi-speaker dataset as a single speaker dataset indices\",\n \"paths def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource,\",\n \"0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if\",\n \"i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag,\",\n \"is not None: indices = np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker =\",\n \"not train_seq2seq and not train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet = True, True\",\n \"None elif train_postnet: assert speaker_ids is None linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat\",\n \"reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None elif train_postnet: assert speaker_ids\",\n \"args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"]\",\n \"dv3.audio.save_wav(signal, path) # Target mel spectrogram if mel_outputs is not None: mel_output =\",\n \"0. for step, (x, input_lengths, mel, y, positions, done, target_lengths, speaker_ids) \\\\ in\",\n \"Mel, Y): self.X = X self.Mel = Mel self.Y = Y # alias\",\n \"decoder states b_pad = r max_target_len += b_pad * downsample_step a = np.array([_pad(x[0],\",\n \"speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None):\",\n \"# Used for Position encoding text_positions, frame_positions = positions # Downsample mel spectrogram\",\n \"v in state.items() if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\":\",\n \"args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id is\",\n \"builder import dv3.audio import dv3.lrschedule import torch from torch.utils import data as data_utils\",\n \"dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except\",\n \"(per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1 def save_checkpoint(model,\",\n \"if we use masked loss if hparams.masked_loss_weight > 0: # decoder output domain\",\n \"return texts, speaker_ids else: return texts def collect_features(self, *args): if self.multi_speaker: text, speaker_id\",\n \"states b_pad = r max_target_len += b_pad * downsample_step a = np.array([_pad(x[0], max_input_len)\",\n \"float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]),\",\n \"super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort by lengths\",\n \"physical explanation.\\\", ] import dv3.synthesis synthesis._frontend = _frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir,\",\n \"Load embedding from checkpoint. --speaker-id=<N> Use specific speaker of data in case for\",\n \"import hparams, hparams_debug_string fs = hparams.sample_rate global_step = 0 global_epoch = 0 use_cuda\",\n \"max_target_len += b_pad * downsample_step a = np.array([_pad(x[0], max_input_len) for x in batch],\",\n \"frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup\",\n \"return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root,\",\n \"x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r //\",\n \"if clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\",\",\n \"k, v in state.items() if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ ==\",\n \"alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text, speaker_id = self.X[idx]\",\n \"of 20 conference.\\\", \\\"Generative adversarial network or variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some\",\n \"done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in batch]) else:\",\n \"self.multi_speaker: text, speaker_id = args else: text = args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob)\",\n \"downsample_step = hparams.downsample_step multi_speaker = len(batch[0]) == 4 # Lengths input_lengths = [len(x[0])\",\n \"= _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32)\",\n \"running_loss = 0. for step, (x, input_lengths, mel, y, positions, done, target_lengths, speaker_ids)\",\n \"text positions text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for x in batch],\",\n \"= input_lengths.max() W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b in range(B): W[b]\",\n \"% batch_size self.batch_group_size = batch_group_size assert batch_group_size % batch_size == 0 self.permutate =\",\n \"positions # Downsample mel spectrogram if downsample_step > 1: mel = mel[:, 0::downsample_step,\",\n \"indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions X =\",\n \"* g)) return W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B = len(input_lengths) max_input_len\",\n \"speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort by lengths 2. Pick a\",\n \"# mel: if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:, r:,\",\n \"- b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment, path, info=None): fig,\",\n \"\\\"\\\"\\\"Trainining script for seq2seq text-to-speech synthesis model. usage: train.py [options] options: --data-root=<dir> Directory\",\n \"def load_checkpoint(path, model, optimizer, reset_optimizer): global global_step global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path))\",\n \"train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet = True, True if train_seq2seq: print(\\\"Training seq2seq\",\n \"class PyTorchDataset(object): def __init__(self, X, Mel, Y): self.X = X self.Mel = Mel\",\n \"os.path import dirname, join from tqdm import tqdm, trange from datetime import datetime\",\n \"range(len(indices) // batch_group_size): s = i * batch_group_size e = s + batch_group_size\",\n \"= 1, max_decoder_target_len + 1 # if b_pad > 0: # s, e\",\n \"as e: warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target mel spectrogram if mel_outputs is\",\n \"input_length = input_lengths[idx] # Alignment # Multi-hop attention if attn is not None\",\n \"max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask # shift mask decoder_target_mask = decoder_target_mask[:, r:, :]\",\n \"r == 0 if max_target_len % downsample_step != 0: max_target_len += downsample_step -\",\n \"frontend, builder import dv3.audio import dv3.lrschedule import torch from torch.utils import data as\",\n \"Used for Position encoding text_positions, frame_positions = positions # Downsample mel spectrogram if\",\n \"= args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id\",\n \"summary writer for tensorboard if log_event_path is None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\"\",\n \"== 5 self.multi_speaker = len(l) == 5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines))\",\n \"return text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return\",\n \"running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch\",\n \"0 global_epoch = 0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend\",\n \"% downsample_step != 0: max_target_len += downsample_step - max_target_len % downsample_step assert max_target_len\",\n \"\\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path, alignment) # Save averaged alignment alignment_dir =\",\n \"= args[\\\"--checkpoint-postnet\\\"] load_embedding = args[\\\"--load-embedding\\\"] checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id =\",\n \"= x.cuda() text_positions = text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet: y = y.cuda()\",\n \"signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs,\",\n \"1) input_length = input_lengths[idx] # Alignment # Multi-hop attention if attn is not\",\n \"in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None else \\\"single\\\" for\",\n \"global_step, global_epoch while global_epoch < nepochs: running_loss = 0. for step, (x, input_lengths,\",\n \"> 0 and global_step % checkpoint_interval == 0: save_states( global_step, writer, mel_outputs, linear_outputs,\",\n \"join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids = [0, 1, 10] if\",\n \"* l1_loss + priority_w * priority_loss # Binary divergence loss if hparams.binary_divergence_weight <=\",\n \"ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep' if info\",\n \"for i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1)\",\n \"- x + eps) def masked_mean(y, mask): # (B, T, D) mask_ =\",\n \"random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X, Mel,\",\n \"save_states( global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer,\",\n \"for x in batch]) else: speaker_ids = None return x_batch, input_lengths, mel_batch, y_batch,\",\n \"not specified if not train_seq2seq and not train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet\",\n \"target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is not None #\",\n \"// 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask,\",\n \"train_seq2seq, train_postnet = True, True if train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training\",\n \"hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend)\",\n \"mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input) loss =\",\n \"self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as\",\n \"in batch] max_input_len = max(input_lengths) target_lengths = [len(x[1]) for x in batch] max_target_len\",\n \"getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr']\",\n \"numba import jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import join, expanduser\",\n \"mask decoder_target_mask = sequence_mask( target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step >\",\n \"- np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip against freq axis return np.uint8(cm.magma(spectrogram.T)\",\n \"None if use_cuda: if train_seq2seq: x = x.cuda() text_positions = text_positions.cuda() frame_positions =\",\n \"using multi-speaker dataset as a single speaker dataset if self.speaker_id is not None:\",\n \"mask is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ =\",\n \"decoder output domain mask decoder_target_mask = sequence_mask( target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1)\",\n \"seq_range = torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda:\",\n \"[None] for speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None\",\n \"Directory contains preprocessed features. --checkpoint-dir=<dir> Directory where to save model checkpoints [default: checkpoints].\",\n \"Permutate batches if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1,\",\n \"= build_model() if use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1,\",\n \"[len(x[1]) for x in batch] max_target_len = max(target_lengths) if max_target_len % r !=\",\n \"= getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id))\",\n \"[default: ]. --checkpoint=<path> Restore model from checkpoint path if given. --checkpoint-seq2seq=<path> Restore seq2seq\",\n \"flags done = np.array([_pad(np.zeros(len(x[1]) // r // downsample_step - 1), max_decoder_target_len, constant_values=1) for\",\n \"= [0, 1, 10] if ismultispeaker else [None] for speaker_id in speaker_ids: speaker_str\",\n \"def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and train_postnet: suffix\",\n \"= \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Predicted mel spectrogram if\",\n \"well for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path =\",\n \"e = 0, 0 for i in range(len(indices) // batch_group_size): s = i\",\n \"r // downsample_step # Feed data x, mel, y = Variable(x), Variable(mel), Variable(y)\",\n \":priority_bin]) l1_loss = (1 - priority_w) * l1_loss + priority_w * priority_loss #\",\n \"0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) # Update loss.backward() if clip_thresh > 0:\",\n \"optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not\",\n \"hparams_debug_string fs = hparams.sample_rate global_step = 0 global_epoch = 0 use_cuda = torch.cuda.is_available()\",\n \"max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\\",\n \"L1 loss if priority_bin is not None and priority_w > 0: if w\",\n \"model\\\") elif train_postnet: print(\\\"Training postnet model\\\") else: assert False, \\\"must be specified wrong\",\n \"[0, 1] spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram =\",\n \"% checkpoint_interval == 0: save_states( global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths,\",\n \"\\\"\\\"\\\" from docopt import docopt import sys from os.path import dirname, join from\",\n \"5 self.multi_speaker = len(l) == 5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if\",\n \":priority_bin]) else: priority_loss = l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) l1_loss = (1\",\n \"constant_values=0): return np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0):\",\n \"aha, need to cast numpy.int64 to int self.frame_lengths = list(map(int, self.frame_lengths)) return paths\",\n \"from docopt import docopt import sys from os.path import dirname, join from tqdm\",\n \"= guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W def train(model, data_loader, optimizer, writer,\",\n \"None l1_loss = l1(y_hat, y) # Priority L1 loss if priority_bin is not\",\n \"\\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model = getattr(builder, hparams.builder)(\",\n \"permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e = 0,\",\n \"False _frontend = None # to be set later def _pad(seq, max_len, constant_values=0):\",\n \"hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with\",\n \"and hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters\",\n \"way to measure the acute emotional intelligence that has never gone out of\",\n \"= model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\" m = model.postnet checkpoint_path = join(\",\n \"jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import join, expanduser import random\",\n \"= data_root self.speaker_ids = None self.multi_speaker = False # If not None, filter\",\n \"done_loss = binary_criterion(done_hat, done) # linear: if train_postnet: n_priority_freq = int(hparams.priority_freq / (fs\",\n \"in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() # Used for Position encoding text_positions, frame_positions\",\n \"load_embedding is not None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary\",\n \"writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\",\",\n \"masked_l1( y_hat[:, :, :priority_bin], y[:, :, :priority_bin], mask=mask) \\\\ + (1 - w)\",\n \"trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model\",\n \"mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels,\",\n \"info=None): fig, ax = plt.subplots() im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im,\",\n \"that has never gone out of style.\\\", \\\"President Trump met with other leaders\",\n \"PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\")\",\n \"# Multi-hop attention if attn is not None and attn.dim() == 4: for\",\n \"s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm = np.arange(len(indices[:e]) //\",\n \"self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text, speaker_id = self.X[idx] return\",\n \"if train_seq2seq: done_loss = binary_criterion(done_hat, done) # linear: if train_postnet: n_priority_freq = int(hparams.priority_freq\",\n \"from checkpoint path if given. --checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint path. --checkpoint-postnet=<path>\",\n \"nepochs: running_loss = 0. for step, (x, input_lengths, mel, y, positions, done, target_lengths,\",\n \"global_epoch, train_seq2seq, train_postnet) if global_step > 0 and global_step % hparams.eval_interval == 0:\",\n \"and train_postnet: loss = mel_loss + linear_loss + done_loss elif train_seq2seq: loss =\",\n \"torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand =\",\n \"path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs) except\",\n \"CERN laboratory say they have discovered a new particle.\\\", \\\"There's a way to\",\n \"\\\"train.txt\\\") with open(meta, \\\"rb\\\") as f: lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert\",\n \"torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r // downsample_step # frame positions s, e\",\n \"= len(batch[0]) == 4 # Lengths input_lengths = [len(x[0]) for x in batch]\",\n \"+= attn_loss if global_step > 0 and global_step % checkpoint_interval == 0: save_states(\",\n \"# https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part of the model from: {}\\\".format(path)) state\",\n \"as f: lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or\",\n \"class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def\",\n \"0: # s, e = s - 1, e - 1 frame_positions =\",\n \"mel = synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) # Alignment\",\n \"os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag =\",\n \"prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T)\",\n \"and exit \\\"\\\"\\\" from docopt import docopt import sys from os.path import dirname,\",\n \"batch_group_size -= batch_group_size % batch_size self.batch_group_size = batch_group_size assert batch_group_size % batch_size ==\",\n \"= linear_loss # attention if train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2),\",\n \"= mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) # Predicted spectrogram\",\n \"= Variable(frame_positions) done = Variable(done) target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids) if ismultispeaker\",\n \"torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size is None: batch_group_size = min(batch_size * 32,\",\n \"if self.multi_speaker: text, speaker_id = args else: text = args[0] seq = _frontend.text_to_sequence(text,\",\n \"Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else soft_mask attn_loss = (attn * soft_mask).mean()\",\n \"if given. --checkpoint-seq2seq=<path> Restore seq2seq model from checkpoint path. --checkpoint-postnet=<path> Restore postnet model\",\n \"{}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq,\",\n \"model = model.cuda() linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr\",\n \"\\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader,\",\n \"f: join(self.data_root, f), paths)) if multi_speaker and self.speaker_id is not None: speaker_ids =\",\n \"or len(l) == 5 self.multi_speaker = len(l) == 5 texts = list(map(lambda l:\",\n \"max_target_len % r != 0: max_target_len += r - max_target_len % r assert\",\n \"speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id is not None else None\",\n \"signal, alignment, _, mel = synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False) signal /=\",\n \"float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step)\",\n \"def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root,\",\n \"= docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path\",\n \"= join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i\",\n \"priority_w > 0: if w > 0: priority_loss = w * masked_l1( y_hat[:,\",\n \"optimizer.zero_grad() # Used for Position encoding text_positions, frame_positions = positions # Downsample mel\",\n \"= target_mask[:, r:, :] else: decoder_target_mask, target_mask = None, None # Apply model\",\n \"def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as f: lines =\",\n \"1, max_decoder_target_len + 1 # if b_pad > 0: # s, e =\",\n \"global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step,\",\n \"# Create mask if we use masked loss if hparams.masked_loss_weight > 0: #\",\n \"= int(speaker_id) if speaker_id is not None else None data_root = args[\\\"--data-root\\\"] if\",\n \"mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32)\",\n \"checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default: ]. --checkpoint=<path> Restore model from checkpoint\",\n \"None else \\\"single\\\" for idx, text in enumerate(texts): signal, alignment, _, mel =\",\n \"PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset = PyTorchDataset(X, Mel, Y)\",\n \"max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain mask target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1)\",\n \"if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args = docopt(__doc__)\",\n \"postnet model from checkpoint path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only Train only\",\n \"(len(data_loader)))) global_epoch += 1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if\",\n \":-r, :], mel[:, r:, :], decoder_target_mask) mel_loss = (1 - w) * mel_l1_loss\",\n \"spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx,\",\n \"print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim,\",\n \"model. usage: train.py [options] options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir> Directory where\",\n \"{}, step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram\",\n \"if mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel\",\n \"linear_outputs, attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq,\",\n \"target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker else None # Create mask if we\",\n \"def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler\",\n \"timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root =\",\n \"step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model = getattr(builder,\",\n \"from torch import optim import torch.backends.cudnn as cudnn from torch.utils import data as\",\n \"batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker = len(batch[0]) == 4 #\",\n \"max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda()\",\n \"n_priority_freq = int(hparams.priority_freq / (fs * 0.5) * linear_dim) linear_l1_loss, linear_binary_div = spec_loss(\",\n \"1] spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram,\",\n \"writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1 def\",\n \"def save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate states\",\n \"target_mask[:, r:, :] else: decoder_target_mask, target_mask = None, None # Apply model if\",\n \"speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step)\",\n \"(B, T, D) mask_ = mask.expand_as(y) return (y * mask_).sum() / mask_.sum() def\",\n \"writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step)\",\n \"[len(x[0]) for x in batch] max_input_len = max(input_lengths) target_lengths = [len(x[1]) for x\",\n \"__len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X, Mel, Y): self.X = X\",\n \"multi_speaker = len(batch[0]) == 4 # Lengths input_lengths = [len(x[0]) for x in\",\n \"sys import os from tensorboardX import SummaryWriter from matplotlib import cm from warnings\",\n \"print(\\\"Restore part of the model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict()\",\n \"len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length = input_lengths[idx] # Alignment #\",\n \"list(map(int, self.frame_lengths)) return paths def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self,\",\n \"+ torch.exp(y_hat_logits)) if w > 0: binary_div = w * masked_mean(z, mask) +\",\n \"(1 - w) * z.mean() else: binary_div = z.mean() return l1_loss, binary_div @jit(nopython=True)\",\n \"batch_size self.batch_group_size = batch_group_size assert batch_group_size % batch_size == 0 self.permutate = permutate\",\n \"decoder_target_mask # shift mask decoder_target_mask = decoder_target_mask[:, r:, :] target_mask = target_mask[:, r:,\",\n \"not None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary writer for\",\n \"the model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict = {k:\",\n \"hparams.sample_rate global_step = 0 global_epoch = 0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark\",\n \"= X self.Mel = Mel self.Y = Y # alias self.multi_speaker = X.file_data_source.multi_speaker\",\n \"for n in range(N): for t in range(T): W[n, t] = 1 -\",\n \"linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat = None, None, None # Losses w\",\n \"hparams.lr_schedule is not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step,\",\n \"l1_loss + priority_w * priority_loss # Binary divergence loss if hparams.binary_divergence_weight <= 0:\",\n \"None, None, None # Losses w = hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss,\",\n \":] else: decoder_target_mask, target_mask = None, None # Apply model if train_seq2seq and\",\n \"checkpoint path. --checkpoint-postnet=<path> Restore postnet model from checkpoint path. --train-seq2seq-only Train only seq2seq\",\n \"self.batch_group_size s, e = 0, 0 for i in range(len(indices) // batch_group_size): s\",\n \"def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker = len(batch[0])\",\n \"= np.flip(spectrogram, axis=1) # flip against freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def\",\n \"w > 0: binary_div = w * masked_mean(z, mask) + (1 - w)\",\n \"self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines))\",\n \"print(\\\"Training whole model\\\") train_seq2seq, train_postnet = True, True if train_seq2seq: print(\\\"Training seq2seq model\\\")\",\n \"Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format(\",\n \"priority_bin is not None and priority_w > 0: if w > 0: priority_loss\",\n \"Train only postnet model. --restore-parts=<path> Restore part of the model. --log-event-path=<name> Log event\",\n \"= Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset\",\n \"self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len is\",\n \"train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train both if not specified if\",\n \"betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model) #\",\n \"if self.speaker_id is not None: indices = np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices])\",\n \"frame_positions = Variable(frame_positions) done = Variable(done) target_lengths = Variable(target_lengths) speaker_ids = Variable(speaker_ids) if\",\n \"== \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path\",\n \"= np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) y_batch = torch.FloatTensor(c) #\",\n \"return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): # harded coded\",\n \"postnet model. --restore-parts=<path> Restore part of the model. --log-event-path=<name> Log event path. --reset-optimizer\",\n \"if train_seq2seq and train_postnet: suffix = \\\"\\\" m = model elif train_seq2seq: suffix\",\n \"self.multi_speaker = len(l) == 5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker:\",\n \"\\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path)\",\n \"writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception as e:\",\n \"spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) # Predicted audio signal signal\",\n \"linear spectrogram\\\", spectrogram, global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /=\",\n \"train_postnet: loss = mel_loss + linear_loss + done_loss elif train_seq2seq: loss = mel_loss\",\n \"= decoder_target_mask[:, r:, :] target_mask = target_mask[:, r:, :] else: decoder_target_mask, target_mask =\",\n \"hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints\",\n \"Restore seq2seq model from checkpoint path. --checkpoint-postnet=<path> Restore postnet model from checkpoint path.\",\n \"Learning rate schedule if hparams.lr_schedule is not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr\",\n \"mel[:, r:, :], decoder_target_mask) mel_loss = (1 - w) * mel_l1_loss + w\",\n \"provide either lengths or mask\\\") # (B, T, 1) if mask is None:\",\n \"ax=ax) xlabel = 'Decoder timestep' if info is not None: xlabel += '\\\\n\\\\n'\",\n \"len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X, Mel, Y): self.X = X self.Mel =\",\n \"speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def load_checkpoint(path, model, optimizer,\",\n \"Position encoding text_positions, frame_positions = positions # Downsample mel spectrogram if downsample_step >\",\n \"against freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker):\",\n \"None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer\",\n \"= binary_criterion(done_hat, done) # linear: if train_postnet: n_priority_freq = int(hparams.priority_freq / (fs *\",\n \"optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model\",\n \"import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4)))\",\n \"speaker_ids is not None # Learning rate schedule if hparams.lr_schedule is not None:\",\n \"info is not None: xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout()\",\n \"--log-event-path=<name> Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N>\",\n \"# Filter by speaker_id # using multi-speaker dataset as a single speaker dataset\",\n \"for speaker_id in speaker_ids: speaker_str = \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None else\",\n \"and global_step % hparams.eval_interval == 0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) # Update\",\n \"mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment, path, info=None): fig, ax = plt.subplots() im\",\n \"fast=False) signal /= np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str))\",\n \"data_utils from torch.utils.data.sampler import Sampler import numpy as np from numba import jit\",\n \"np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions)\",\n \"- 1), max_decoder_target_len, constant_values=1) for x in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker:\",\n \"1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # save files as well for\",\n \"torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]),\",\n \"input_lengths[idx] # Alignment # Multi-hop attention if attn is not None and attn.dim()\",\n \"done_hat = model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids\",\n \"save files as well for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir,\",\n \"sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model = build_model() if use_cuda: model =\",\n \"Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and\",\n \"out of style.\\\", \\\"President Trump met with other leaders at the Group of\",\n \"train_seq2seq or train_postnet global global_step, global_epoch while global_epoch < nepochs: running_loss = 0.\",\n \"sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask # shift mask decoder_target_mask = decoder_target_mask[:,\",\n \"= getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups:\",\n \"spectrogram = np.flip(spectrogram, axis=1) # flip against freq axis return np.uint8(cm.magma(spectrogram.T) * 255)\",\n \"# Apply model if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn, done_hat = model(\",\n \"global_step) # Predicted mel spectrogram if mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy()\",\n \"loss if priority_bin is not None and priority_w > 0: if w >\",\n \"writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer,\",\n \"part of the model. --log-event-path=<name> Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load\",\n \"encoding text_positions, frame_positions = positions # Downsample mel spectrogram if downsample_step > 1:\",\n \"nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq,\",\n \"f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 multi_speaker\",\n \"data_root self.speaker_ids = None self.multi_speaker = False # If not None, filter by\",\n \"mask=mask) + (1 - w) * l1(y_hat, y) else: assert mask is None\",\n \"x = x.cuda() text_positions = text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet: y =\",\n \"writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram, global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal\",\n \"X self.Mel = Mel self.Y = Y # alias self.multi_speaker = X.file_data_source.multi_speaker def\",\n \"or variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have accepted this as a miracle\",\n \"Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception as e: warn(str(e))\",\n \"else: y_hat_logits = logit(y_hat) z = -y * y_hat_logits + torch.log(1 + torch.exp(y_hat_logits))\",\n \"= self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e = 0, 0 for i in\",\n \"0 for zero beginning padding # imitates initial decoder states b_pad = r\",\n \"* masked_l1(y_hat, y, mask=mask) + (1 - w) * l1(y_hat, y) else: assert\",\n \"+ str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train!\",\n \"raise RuntimeError(\\\"Should provide either lengths or mask\\\") # (B, T, 1) if mask\",\n \"FileDataSource from os.path import join, expanduser import random import librosa.display from matplotlib import\",\n \"imitates initial decoder states b_pad = r max_target_len += b_pad * downsample_step a\",\n \"checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer, global_step, checkpoint_dir,\",\n \"this help message and exit \\\"\\\"\\\" from docopt import docopt import sys from\",\n \"__init__(self, data_root, col, speaker_id=None): self.data_root = data_root self.col = col self.frame_lengths = []\",\n \"lines)) if self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id\",\n \"downsample_step != 0: max_target_len += downsample_step - max_target_len % downsample_step assert max_target_len %\",\n \"+ done_loss elif train_seq2seq: loss = mel_loss + done_loss elif train_postnet: loss =\",\n \"train_postnet): if train_seq2seq and train_postnet: suffix = \\\"\\\" m = model elif train_seq2seq:\",\n \"small patch and randomize it 3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16,\",\n \"train_seq2seq and train_postnet: loss = mel_loss + linear_loss + done_loss elif train_seq2seq: loss\",\n \"state.items() if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args =\",\n \"Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted\",\n \"list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker\",\n \"(seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def\",\n \"\\\"Some have accepted this as a miracle without any physical explanation.\\\", ] import\",\n \"if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\",\",\n \"mask_) return loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step\",\n \"np.zeros((max_N, max_T), dtype=np.float32) for n in range(N): for t in range(T): W[n, t]\",\n \"None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary writer for tensorboard\",\n \"return model def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data =\",\n \"= torch.FloatTensor(c) # text positions text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for\",\n \"torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in batch]) else: speaker_ids =\",\n \"global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs) except Exception as e: warn(str(e))\",\n \"# Train! try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq,\",\n \"def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None,\",\n \"// downsample_step # frame positions s, e = 1, max_decoder_target_len + 1 #\",\n \"= np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c\",\n \"checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step > 0\",\n \"= None, None, None # Losses w = hparams.binary_divergence_weight # mel: if train_seq2seq:\",\n \"\\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Mel writer.add_image(\\\"(Eval) Predicted mel\",\n \"num_workers=hparams.num_workers, sampler=sampler, collate_fn=collate_fn, pin_memory=hparams.pin_memory) print(\\\"dataloader_prepared\\\") # Model model = build_model() if use_cuda: model\",\n \"divergence loss if hparams.binary_divergence_weight <= 0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat)\",\n \"if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn, done_hat = model( x, mel, speaker_ids=speaker_ids,\",\n \"of style.\\\", \\\"President Trump met with other leaders at the Group of 20\",\n \"Hyper parameters [default: ]. --checkpoint=<path> Restore model from checkpoint path if given. --checkpoint-seq2seq=<path>\",\n \"[(b_pad, max_len - len(x) - b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return x def\",\n \"* mel_binary_div # done: if train_seq2seq: done_loss = binary_criterion(done_hat, done) # linear: if\",\n \"the Group of 20 conference.\\\", \\\"Generative adversarial network or variational auto-encoder.\\\", \\\"Please call\",\n \"= join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path,\",\n \"= np.pad(x, [(b_pad, max_len - len(x) - b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return\",\n \"w = hparams.masked_loss_weight # L1 loss if w > 0: assert mask is\",\n \"if w > 0: assert mask is not None l1_loss = w *\",\n \"log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer =\",\n \"w * masked_l1(y_hat, y, mask=mask) + (1 - w) * l1(y_hat, y) else:\",\n \"if max_len is None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0) seq_range = torch.arange(0,\",\n \"+ eps) - torch.log(1 - x + eps) def masked_mean(y, mask): # (B,\",\n \"# linear: if train_postnet: n_priority_freq = int(hparams.priority_freq / (fs * 0.5) * linear_dim)\",\n \"x.cuda() text_positions = text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet: y = y.cuda() mel\",\n \"max_target_len // r // downsample_step # frame positions s, e = 1, max_decoder_target_len\",\n \"* z.mean() else: binary_div = z.mean() return l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N,\",\n \"= \\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None else \\\"single\\\" for idx, text in\",\n \"if lengths is None and mask is None: raise RuntimeError(\\\"Should provide either lengths\",\n \"global_step global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not\",\n \"len(input_lengths) - 1) input_length = input_lengths[idx] # Alignment # Multi-hop attention if attn\",\n \"# Target mel spectrogram if mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy() mel_output\",\n \"Train only seq2seq model. --train-postnet-only Train only postnet model. --restore-parts=<path> Restore part of\",\n \"= state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part of the model from:\",\n \"--restore-parts=<path> Restore part of the model. --log-event-path=<name> Log event path. --reset-optimizer Reset optimizer.\",\n \"priority_w * priority_loss # Binary divergence loss if hparams.binary_divergence_weight <= 0: binary_div =\",\n \"output domain mask decoder_target_mask = sequence_mask( target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if\",\n \"len(l) == 5 self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda\",\n \"= args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else:\",\n \"# Lengths input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r // downsample_step #\",\n \"linear_loss + done_loss elif train_seq2seq: loss = mel_loss + done_loss elif train_postnet: loss\",\n \"= optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts is not None:\",\n \"= len(l) == 5 texts = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids\",\n \"os.path import join, expanduser import random import librosa.display from matplotlib import pyplot as\",\n \"value_projection=hparams.value_projection, ) return model def load_checkpoint(path, model, optimizer, reset_optimizer): global global_step global global_epoch\",\n \"train_postnet: y = y.cuda() mel = mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids\",\n \"getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step,\",\n \"l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f:\",\n \"averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment\",\n \"sequence_mask( target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain\",\n \"optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model,\",\n \"mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask) mel_loss = (1 - w) *\",\n \"attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0,\",\n \"epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers,\",\n \"binary_criterion = nn.BCELoss() assert train_seq2seq or train_postnet global global_step, global_epoch while global_epoch <\",\n \"None # to be set later def _pad(seq, max_len, constant_values=0): return np.pad(seq, (0,\",\n \"[options] options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir> Directory where to save model\",\n \"(r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain mask target_mask =\",\n \"lines)) paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root, f),\",\n \"= (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) #\",\n \"= args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id is not None\",\n \"import os from tensorboardX import SummaryWriter from matplotlib import cm from warnings import\",\n \"linear: if train_postnet: n_priority_freq = int(hparams.priority_freq / (fs * 0.5) * linear_dim) linear_l1_loss,\",\n \"postnet model\\\") else: assert False, \\\"must be specified wrong args\\\" # Override hyper\",\n \"None: mel_output = mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) #\",\n \"dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root,\",\n \"= 0. for step, (x, input_lengths, mel, y, positions, done, target_lengths, speaker_ids) \\\\\",\n \"loss = mel_loss + linear_loss + done_loss elif train_seq2seq: loss = mel_loss +\",\n \"1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags\",\n \"# [0, 1] spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram\",\n \"lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 multi_speaker = len(l) ==\",\n \"for step, (x, input_lengths, mel, y, positions, done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)):\",\n \"sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__()\",\n \"= max(target_lengths) if max_target_len % r != 0: max_target_len += r - max_target_len\",\n \"= model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if checkpoint_restore_parts\",\n \"seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand\",\n \"= torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1]) // r\",\n \"input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d\",\n \"save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): #\",\n \"a miracle without any physical explanation.\\\", ] import dv3.synthesis synthesis._frontend = _frontend eval_output_dir\",\n \"except Exception as e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel,\",\n \"train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt:\",\n \"target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32)\",\n \"args else: text = args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq,\",\n \"max_target_len % downsample_step assert max_target_len % downsample_step == 0 # Set 0 for\",\n \"else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path)\",\n \"= Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions = Variable(frame_positions) done = Variable(done)\",\n \"return W def train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True,\",\n \"seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq,\",\n \"t / T)**2 / (2 * g * g)) return W def guided_attentions(input_lengths,\",\n \"if global_step > 0 and global_step % checkpoint_interval == 0: save_states( global_step, writer,\",\n \"from nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import join, expanduser import random import\",\n \"# Save averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir,\",\n \"in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W def train(model,\",\n \"data_root, speaker_id=None): super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource,\",\n \"model\\\") train_seq2seq, train_postnet = True, True if train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet:\",\n \"x in batch] max_target_len = max(target_lengths) if max_target_len % r != 0: max_target_len\",\n \"float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step)\",\n \"FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler\",\n \"(1 - w) * l1(y_hat, y) else: assert mask is None l1_loss =\",\n \"None l1_loss = w * masked_l1(y_hat, y, mask=mask) + (1 - w) *\",\n \"y, mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight\",\n \"import frontend, builder import dv3.audio import dv3.lrschedule import torch from torch.utils import data\",\n \"linear_binary_div # Combine losses if train_seq2seq and train_postnet: loss = mel_loss + linear_loss\",\n \"if hparams.preset is not None and hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset] import\",\n \"c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) y_batch = torch.FloatTensor(c)\",\n \"model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) #\",\n \"else soft_mask attn_loss = (attn * soft_mask).mean() loss += attn_loss if global_step >\",\n \"now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format(\",\n \"spectrogram, global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path\",\n \"\\\"rb\\\") as f: lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4\",\n \"import data as data_utils from torch.utils.data.sampler import Sampler import numpy as np from\",\n \"_, mel = synthesis.tts( model, text, p=0, speaker_id=speaker_id, fast=False) signal /= np.max(np.abs(signal)) #\",\n \"hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel =\",\n \"= int(hparams.priority_freq / (fs * 0.5) * linear_dim) linear_l1_loss, linear_binary_div = spec_loss( linear_outputs[:,\",\n \"mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None mel_outputs, attn,\",\n \"json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions X\",\n \"dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None):\",\n \"0: batch_group_size -= batch_group_size % batch_size self.batch_group_size = batch_group_size assert batch_group_size % batch_size\",\n \"perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle\",\n \"Load embedding if load_embedding is not None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model)\",\n \"text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str))\",\n \"else None if use_cuda: if train_seq2seq: x = x.cuda() text_positions = text_positions.cuda() frame_positions\",\n \"T, D) mask_ = mask.expand_as(y) return (y * mask_).sum() / mask_.sum() def spec_loss(y_hat,\",\n \"= Mel self.Y = Y # alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx):\",\n \"idx, text in enumerate(texts): signal, alignment, _, mel = synthesis.tts( model, text, p=0,\",\n \"None: raise RuntimeError(\\\"Should provide either lengths or mask\\\") # (B, T, 1) if\",\n \"writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) # Target spectrogram if linear_outputs is not None:\",\n \"tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # save\",\n \"= min(1, len(input_lengths) - 1) input_length = input_lengths[idx] # Alignment # Multi-hop attention\",\n \"global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\",\",\n \"l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[3], lines)) if self.multi_speaker: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter\",\n \"speaker_ids = None return x_batch, input_lengths, mel_batch, y_batch, \\\\ (text_positions, frame_positions), done, target_lengths,\",\n \"if use_cuda: if train_seq2seq: x = x.cuda() text_positions = text_positions.cuda() frame_positions = frame_positions.cuda()\",\n \"SummaryWriter from matplotlib import cm from warnings import warn from dv3.hparams import hparams,\",\n \"not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq,\",\n \"is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output,\",\n \"+ w * mel_binary_div # done: if train_seq2seq: done_loss = binary_criterion(done_hat, done) #\",\n \"print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"]\",\n \"train_postnet: loss = linear_loss # attention if train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths,\",\n \"return iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X, Mel, Y):\",\n \"= model.postnet(mel) mel_outputs, attn, done_hat = None, None, None # Losses w =\",\n \"col self.frame_lengths = [] self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\")\",\n \"self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X) def\",\n \"call Stella.\\\", \\\"Some have accepted this as a miracle without any physical explanation.\\\",\",\n \"import random import librosa.display from matplotlib import pyplot as plt import sys import\",\n \"mask_ = mask.expand_as(y) return (y * mask_).sum() / mask_.sum() def spec_loss(y_hat, y, mask,\",\n \"= \\\"\\\" m = model elif train_seq2seq: suffix = \\\"_seq2seq\\\" m = model.seq2seq\",\n \"train_postnet: print(\\\"Training postnet model\\\") else: assert False, \\\"must be specified wrong args\\\" #\",\n \"batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b =\",\n \"max_target_len % downsample_step != 0: max_target_len += downsample_step - max_target_len % downsample_step assert\",\n \"batch_group_size % batch_size != 0: batch_group_size -= batch_group_size % batch_size self.batch_group_size = batch_group_size\",\n \"docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path =\",\n \"torch.utils import data as data_utils from torch.utils.data.sampler import Sampler import numpy as np\",\n \"datasets. -h, --help Show this help message and exit \\\"\\\"\\\" from docopt import\",\n \"0: binary_div = w * masked_mean(z, mask) + (1 - w) * z.mean()\",\n \"# Losses w = hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss(\",\n \"{}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch = checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model):\",\n \"= args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding =\",\n \"batch_group_size % batch_size == 0 self.permutate = permutate def __iter__(self): indices = self.sorted_indices.clone()\",\n \"adversarial network or variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have accepted this as\",\n \"if w > 0: binary_div = w * masked_mean(z, mask) + (1 -\",\n \"if downsample_step > 1: mel = mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths =\",\n \"None and priority_w > 0: if w > 0: priority_loss = w *\",\n \"W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W def train(model, data_loader, optimizer,\",\n \"# using multi-speaker dataset as a single speaker dataset if self.speaker_id is not\",\n \"signal /= np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path,\",\n \"if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def\",\n \"attn, done_hat = model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert\",\n \"seq_range_expand = Variable(seq_range_expand) if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand)\",\n \"# (B, T, 1) if mask is None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) #\",\n \"-= batch_group_size % batch_size self.batch_group_size = batch_group_size assert batch_group_size % batch_size == 0\",\n \"# Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir,\",\n \"checkpoint_restore_parts = args[\\\"--restore-parts\\\"] speaker_id = args[\\\"--speaker-id\\\"] speaker_id = int(speaker_id) if speaker_id is not\",\n \"super(MelSpecDataSource, self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0,\",\n \"lengths 2. Pick a small patch and randomize it 3. Permutate mini-batchs \\\"\\\"\\\"\",\n \"= \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # save files\",\n \"not None else \\\"single\\\" for idx, text in enumerate(texts): signal, alignment, _, mel\",\n \"x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions = Variable(frame_positions)\",\n \"and randomize it 3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True):\",\n \"prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram))\",\n \"# Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]),\",\n \"event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use specific\",\n \"<reponame>drat/Neural-Voice-Cloning-With-Few-Samples \\\"\\\"\\\"Trainining script for seq2seq text-to-speech synthesis model. usage: train.py [options] options: --data-root=<dir>\",\n \"torch.backends.cudnn as cudnn from torch.utils import data as data_utils from torch.utils.data.sampler import Sampler\",\n \"cudnn from torch.utils import data as data_utils from torch.utils.data.sampler import Sampler import numpy\",\n \"+ w * linear_binary_div # Combine losses if train_seq2seq and train_postnet: loss =\",\n \"\\\"deepvoice3\\\" # Presets if hparams.preset is not None and hparams.preset != \\\"\\\": preset\",\n \"sample_rate=fs) except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target mel spectrogram\",\n \"model. --log-event-path=<name> Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding from checkpoint.\",\n \"Variable(speaker_ids) if ismultispeaker else None if use_cuda: if train_seq2seq: x = x.cuda() text_positions\",\n \"from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) # Setup summary writer for tensorboard if log_event_path is\",\n \"input_lengths.max() W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b in range(B): W[b] =\",\n \"json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend\",\n \"help message and exit \\\"\\\"\\\" from docopt import docopt import sys from os.path\",\n \"model_dict = model.state_dict() valid_state_dict = {k: v for k, v in state.items() if\",\n \"y.cuda() mel = mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if\",\n \"reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding\",\n \"hparams.name == \\\"deepvoice3\\\" # Presets if hparams.preset is not None and hparams.preset !=\",\n \"torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for\",\n \"{k: v for k, v in state.items() if k in model_dict} model_dict.update(valid_state_dict) model.load_state_dict(model_dict)\",\n \"def collect_features(self, *args): if self.multi_speaker: text, speaker_id = args else: text = args[0]\",\n \"from datetime import datetime # The deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder\",\n \"writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]),\",\n \"mel, y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions = Variable(frame_positions) done\",\n \":], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) * linear_l1_loss + w\",\n \"multi-speaker dataset as a single speaker dataset if self.speaker_id is not None: indices\",\n \"\\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path =\",\n \"in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad) for x\",\n \"interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep' if info is not None: xlabel\",\n \"global_step) def logit(x, eps=1e-8): return torch.log(x + eps) - torch.log(1 - x +\",\n \"+= batch_group_size if s < len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices)\",\n \"train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) print(\\\"Finished\\\")\",\n \"min(1, len(input_lengths) - 1) input_length = input_lengths[idx] # Alignment # Multi-hop attention if\",\n \"== 4 or len(l) == 5 multi_speaker = len(l) == 5 self.frame_lengths =\",\n \"for x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2], max_target_len, b_pad=b_pad)\",\n \"= guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda\",\n \"if w > 0: priority_loss = w * masked_l1( y_hat[:, :, :priority_bin], y[:,\",\n \"y) # Priority L1 loss if priority_bin is not None and priority_w >\",\n \"paths = list(map(lambda l: l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root, f), paths))\",\n \"Show this help message and exit \\\"\\\"\\\" from docopt import docopt import sys\",\n \"= sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss,\",\n \"states at step {}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths)\",\n \"Variable from torch import nn from torch import optim import torch.backends.cudnn as cudnn\",\n \"for x in batch], dtype=np.float32) y_batch = torch.FloatTensor(c) # text positions text_positions =\",\n \"iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X, Mel, Y): self.X\",\n \"not None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step)\",\n \"model, optimizer, global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step > 0 and global_step\",\n \"be specified wrong args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name ==\",\n \"list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64 to int self.frame_lengths\",\n \"for x in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for\",\n \"speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort by\",\n \"255), global_step) # save files as well for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i\",\n \"global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet:\",\n \"expanduser import random import librosa.display from matplotlib import pyplot as plt import sys\",\n \"writer = SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval,\",\n \"max_target_len += downsample_step - max_target_len % downsample_step assert max_target_len % downsample_step == 0\",\n \"= decoder_target_mask # shift mask decoder_target_mask = decoder_target_mask[:, r:, :] target_mask = target_mask[:,\",\n \"Train! try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet)\",\n \"= current_lr optimizer.zero_grad() # Used for Position encoding text_positions, frame_positions = positions #\",\n \"mel_batch, y_batch, \\\\ (text_positions, frame_positions), done, target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M')\",\n \"class MaskedL1Loss(nn.Module): def __init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target,\",\n \"FileSourceDataset, FileDataSource from os.path import join, expanduser import random import librosa.display from matplotlib\",\n \"randomize it 3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths,\",\n \"mask is None: raise RuntimeError(\\\"Should provide either lengths or mask\\\") # (B, T,\",\n \"multi_speaker and self.speaker_id is not None: speaker_ids = list(map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) #\",\n \"else: assert mask is None l1_loss = l1(y_hat, y) # Priority L1 loss\",\n \"range(N): for t in range(T): W[n, t] = 1 - np.exp(-(n / N\",\n \"model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 +\",\n \"mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs = None elif train_postnet: assert speaker_ids is None linear_outputs\",\n \"hparams.save_optimizer_state else None torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, },\",\n \"LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized\",\n \"z = -y * y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w > 0:\",\n \"dtype=np.float32) y_batch = torch.FloatTensor(c) # text positions text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1),\",\n \"np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length = input_lengths[idx] # Alignment\",\n \"hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker = len(batch[0]) == 4 # Lengths input_lengths =\",\n \"mask_ = mask.expand_as(input) loss = self.criterion(input * mask_, target * mask_) return loss\",\n \"return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self, X, Mel, Y): self.X = X self.Mel\",\n \"binary_div = z.mean() return l1_loss, binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T, g):\",\n \"global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target mel\",\n \"self).__init__(data_root, 1, speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id)\",\n \"= join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment,\",\n \"in range(T): W[n, t] = 1 - np.exp(-(n / N - t /\",\n \"spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask) mel_loss = (1 - w)\",\n \"seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand < seq_length_expand).float() class MaskedL1Loss(nn.Module): def\",\n \"domain mask decoder_target_mask = sequence_mask( target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step\",\n \"Losses w = hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:,\",\n \"np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip against freq axis return np.uint8(cm.magma(spectrogram.T) *\",\n \"model.postnet(mel) mel_outputs, attn, done_hat = None, None, None # Losses w = hparams.binary_divergence_weight\",\n \"wrong args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\" #\",\n \"torch.save({ \\\"state_dict\\\": m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\",\",\n \"padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding,\",\n \"import nn from torch import optim import torch.backends.cudnn as cudnn from torch.utils import\",\n \"x in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x\",\n \"data_root is None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer =\",\n \"PyTorchDataset(object): def __init__(self, X, Mel, Y): self.X = X self.Mel = Mel self.Y\",\n \"time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram - np.min(spectrogram)) /\",\n \"lengths is None and mask is None: raise RuntimeError(\\\"Should provide either lengths or\",\n \"args) checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"]\",\n \"Sampler import numpy as np from numba import jit from nnmnkwii.datasets import FileSourceDataset,\",\n \"**hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() # Used for Position\",\n \"hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda()\",\n \"target_mask = sequence_mask( target_lengths, max_len=y.size(1)).unsqueeze(-1) else: target_mask = decoder_target_mask # shift mask decoder_target_mask\",\n \"np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last elements\",\n \"return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None): self.data_root = data_root\",\n \"(y * mask_).sum() / mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1 =\",\n \"self.speaker_id is not None: indices = np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker\",\n \"\\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def build_model():\",\n \"batch] max_target_len = max(target_lengths) if max_target_len % r != 0: max_target_len += r\",\n \"\\\"multispeaker{}\\\".format(speaker_id) if speaker_id is not None else \\\"single\\\" for idx, text in enumerate(texts):\",\n \"speaker dataset if self.speaker_id is not None: indices = np.array(speaker_ids) == self.speaker_id texts\",\n \"train_postnet = True, True if train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training postnet\",\n \"if log_event_path is None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event\",\n \"self.criterion(input * mask_, target * mask_) return loss / mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create\",\n \"train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]),\",\n \"!= 0: max_target_len += r - max_target_len % r assert max_target_len % r\",\n \"in batch] max_target_len = max(target_lengths) if max_target_len % r != 0: max_target_len +=\",\n \"(len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader)))) global_epoch += 1\",\n \"by speaker_id self.speaker_id = speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta,\",\n \"w = hparams.binary_divergence_weight # mel: if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r,\",\n \"for Position encoding text_positions, frame_positions = positions # Downsample mel spectrogram if downsample_step\",\n \"--load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use specific speaker of data in case\",\n \"is not None and priority_w > 0: if w > 0: priority_loss =\",\n \"# idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length =\",\n \"b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b)\",\n \"y_hat[:, :, :priority_bin], y[:, :, :priority_bin], mask=mask) \\\\ + (1 - w) *\",\n \"= permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e =\",\n \"def build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size //\",\n \"model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion = nn.BCELoss()\",\n \"Permutate mini-batchs \\\"\\\"\\\" def __init__(self, lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths))\",\n \"max_input_len, target_lengths[b], max_target_len, g).T return W def train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None,\",\n \"max_target_len % r == 0 if max_target_len % downsample_step != 0: max_target_len +=\",\n \"np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Predicted mel spectrogram if mel_outputs is not\",\n \"global_step > 0 and global_step % hparams.eval_interval == 0: eval_model(global_step, writer, model, checkpoint_dir,\",\n \"max_target_len = max(target_lengths) if max_target_len % r != 0: max_target_len += r -\",\n \"speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as f: lines\",\n \"train_seq2seq: print(\\\"Training seq2seq model\\\") elif train_postnet: print(\\\"Training postnet model\\\") else: assert False, \\\"must\",\n \"forward(self, input, target, lengths=None, mask=None, max_len=None): if lengths is None and mask is\",\n \"batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset = PyTorchDataset(X, Mel, Y) data_loader =\",\n \"# done flags done = np.array([_pad(np.zeros(len(x[1]) // r // downsample_step - 1), max_decoder_target_len,\",\n \"global_step % checkpoint_interval == 0: save_states( global_step, writer, mel_outputs, linear_outputs, attn, mel, y,\",\n \"import docopt import sys from os.path import dirname, join from tqdm import tqdm,\",\n \"if use_cuda: cudnn.benchmark = False _frontend = None # to be set later\",\n \"1. Sort by lengths 2. Pick a small patch and randomize it 3.\",\n \"last elements s += batch_group_size if s < len(indices): random.shuffle(indices[s:]) return iter(indices) def\",\n \"ismultispeaker): # harded coded texts = [ \\\"Scientists at the CERN laboratory say\",\n \"_pad_2d(x, max_len, b_pad=0): x = np.pad(x, [(b_pad, max_len - len(x) - b_pad), (0,\",\n \"target_lengths, max_target_len, g=0.2): B = len(input_lengths) max_input_len = input_lengths.max() W = np.zeros((B, max_target_len,\",\n \"mel: if train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :],\",\n \"mask decoder_target_mask = decoder_target_mask[:, r:, :] target_mask = target_mask[:, r:, :] else: decoder_target_mask,\",\n \"writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) # Predicted spectrogram if linear_outputs is not None:\",\n \"(attn * soft_mask).mean() loss += attn_loss if global_step > 0 and global_step %\",\n \"int(speaker_id) else: return np.asarray(seq, dtype=np.int32) class _NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None): self.data_root\",\n \"sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand = seq_range.unsqueeze(0).expand(batch_size, max_len) seq_range_expand = Variable(seq_range_expand) if\",\n \"is not None: linear_output = linear_outputs[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Predicted linear spectrogram\\\", spectrogram,\",\n \"= input_lengths[idx] # Alignment # Multi-hop attention if attn is not None and\",\n \"mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch,\",\n \"os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root,\",\n \"definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id))\",\n \"print(\\\"dataloader_prepared\\\") # Model model = build_model() if use_cuda: model = model.cuda() optimizer =\",\n \"the model. --log-event-path=<name> Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding from\",\n \"f: lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l)\",\n \"def __len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len is None: max_len =\",\n \"= FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths,\",\n \"batch_size if batch_group_size is None: batch_group_size = min(batch_size * 32, len(self.lengths)) if batch_group_size\",\n \"self.X = X self.Mel = Mel self.Y = Y # alias self.multi_speaker =\",\n \"= args[\\\"--data-root\\\"] if data_root is None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path =\",\n \"downsample_step - 1), max_decoder_target_len, constant_values=1) for x in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if\",\n \"not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is not None: print(\\\"Load optimizer state\",\n \"= running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss, global_epoch) print(\\\"Loss: {}\\\".format(running_loss / (len(data_loader))))\",\n \"+ priority_w * priority_loss # Binary divergence loss if hparams.binary_divergence_weight <= 0: binary_div\",\n \"clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step) writer.add_scalar(\\\"mel\",\n \"l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id # using multi-speaker dataset as a\",\n \"dataset if self.speaker_id is not None: indices = np.array(speaker_ids) == self.speaker_id texts =\",\n \"we use masked loss if hparams.masked_loss_weight > 0: # decoder output domain mask\",\n \"{}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir,\",\n \"dv3.hparams import hparams, hparams_debug_string fs = hparams.sample_rate global_step = 0 global_epoch = 0\",\n \"print(\\\"Los event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader, optimizer,\",\n \"== 0 if max_target_len % downsample_step != 0: max_target_len += downsample_step - max_target_len\",\n \"np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs)\",\n \"text = args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id)\",\n \"_pad(seq, max_len, constant_values=0): return np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x,\",\n \"x, mel, speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None mel_outputs,\",\n \"join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path, alignment) # Save averaged alignment alignment_dir\",\n \"writer.add_image(\\\"(Eval) Predicted mel spectrogram text{}_{}\\\".format(idx, speaker_str), prepare_spec_image(mel), global_step) # Audio path = join(eval_output_dir,\",\n \"import jit from nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import join, expanduser import\",\n \"np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): # harded coded texts\",\n \"a way to measure the acute emotional intelligence that has never gone out\",\n \"= input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r // downsample_step # Feed data x,\",\n \"0: writer.add_scalar(\\\"gradient norm\\\", grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step += 1 running_loss\",\n \"= logit(y_hat) z = -y * y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w\",\n \"from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict = {k: v for\",\n \"network or variational auto-encoder.\\\", \\\"Please call Stella.\\\", \\\"Some have accepted this as a\",\n \"idx, speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255),\",\n \"valid_state_dict = {k: v for k, v in state.items() if k in model_dict}\",\n \"-h, --help Show this help message and exit \\\"\\\"\\\" from docopt import docopt\",\n \"part of the model from: {}\\\".format(path)) state = torch.load(path)[\\\"state_dict\\\"] model_dict = model.state_dict() valid_state_dict\",\n \"s += batch_group_size if s < len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return\",\n \"= np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts return\",\n \"!= 0: max_target_len += downsample_step - max_target_len % downsample_step assert max_target_len % downsample_step\",\n \"mask_).sum() / mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1\",\n \"= r max_target_len += b_pad * downsample_step a = np.array([_pad(x[0], max_input_len) for x\",\n \"= join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal, global_step, sample_rate=fs) except Exception\",\n \"train_seq2seq and not train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet = True, True if\",\n \"hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper parameters with preset \\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend =\",\n \"SummaryWriter(log_dir=log_event_path) # Train! try: train(model, data_loader, optimizer, writer, init_lr=hparams.initial_learning_rate, checkpoint_dir=checkpoint_dir, checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh,\",\n \"collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as f: lines = f.readlines()\",\n \"sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset = PyTorchDataset(X,\",\n \"speaker dataset indices = np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices])\",\n \"= torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\",\",\n \"masked_mean(z, mask) + (1 - w) * z.mean() else: binary_div = z.mean() return\",\n \"def _pad_2d(x, max_len, b_pad=0): x = np.pad(x, [(b_pad, max_len - len(x) - b_pad),\",\n \"r max_target_len += b_pad * downsample_step a = np.array([_pad(x[0], max_input_len) for x in\",\n \"global_step % hparams.eval_interval == 0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) # Update loss.backward()\",\n \"converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input, max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, )\",\n \"model to be trained train_seq2seq = args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train both\",\n \"either lengths or mask\\\") # (B, T, 1) if mask is None: mask\",\n \"assert speaker_ids is None mel_outputs, attn, done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions,\",\n \"min(batch_size * 32, len(self.lengths)) if batch_group_size % batch_size != 0: batch_group_size -= batch_group_size\",\n \"global_step, idx, speaker_str)) save_alignment(path, alignment) tag = \\\"eval_averaged_alignment_{}_{}\\\".format(idx, speaker_str) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) *\",\n \"tensorboard if log_event_path is None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los\",\n \"def __getitem__(self, idx): if self.multi_speaker: text, speaker_id = self.X[idx] return text, self.Mel[idx], self.Y[idx],\",\n \"is not None: load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path,\",\n \"spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x + eps) - torch.log(1 - x\",\n \"embedding if load_embedding is not None: print(\\\"Loading embedding from {}\\\".format(load_embedding)) _load_embedding(load_embedding, model) #\",\n \"torch from torch.utils import data as data_utils from torch.autograd import Variable from torch\",\n \"w) * l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) else: priority_loss = l1(y_hat[:, :,\",\n \"l1 = nn.L1Loss() w = hparams.masked_loss_weight # L1 loss if w > 0:\",\n \"None and hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override hyper\",\n \"case for multi-speaker datasets. -h, --help Show this help message and exit \\\"\\\"\\\"\",\n \"embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels,\",\n \"# Update loss.backward() if clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step()\",\n \"alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment,\",\n \"reset_optimizer) if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is\",\n \"= False # If not None, filter by speaker_id self.speaker_id = speaker_id def\",\n \"eps) - torch.log(1 - x + eps) def masked_mean(y, mask): # (B, T,\",\n \"m = model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\" m = model.postnet checkpoint_path =\",\n \"B = len(input_lengths) max_input_len = input_lengths.max() W = np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for\",\n \"indices = np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha,\",\n \"optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and train_postnet: suffix = \\\"\\\"\",\n \"for tensorboard if log_event_path is None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\")\",\n \"alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255),\",\n \"they have discovered a new particle.\\\", \\\"There's a way to measure the acute\",\n \"= [len(x[0]) for x in batch] max_input_len = max(input_lengths) target_lengths = [len(x[1]) for\",\n \"l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) else: priority_loss = l1(y_hat[:, :, :priority_bin], y[:,\",\n \"False return texts return texts, speaker_ids else: return texts def collect_features(self, *args): if\",\n \"elif train_postnet: suffix = \\\"_postnet\\\" m = model.postnet checkpoint_path = join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step,\",\n \"\\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # Predicted mel spectrogram if mel_outputs\",\n \"self.speaker_ids = None self.multi_speaker = False # If not None, filter by speaker_id\",\n \"Pick a small patch and randomize it 3. Permutate mini-batchs \\\"\\\"\\\" def __init__(self,\",\n \"gone out of style.\\\", \\\"President Trump met with other leaders at the Group\",\n \"signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\", signal,\",\n \"window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def load_checkpoint(path, model, optimizer, reset_optimizer): global\",\n \"= dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio\",\n \"load_checkpoint(checkpoint_postnet_path, model.postnet, optimizer, reset_optimizer) if checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer)\",\n \"= soft_mask.cuda() if use_cuda else soft_mask attn_loss = (attn * soft_mask).mean() loss +=\",\n \"if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1)\",\n \"= len(l) == 5 self.frame_lengths = list( map(lambda l: int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[2]), lines)) paths =\",\n \"speaker_id = self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx]\",\n \"np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip against freq\",\n \"y_hat_logits = logit(y_hat) z = -y * y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if\",\n \"# done: if train_seq2seq: done_loss = binary_criterion(done_hat, done) # linear: if train_postnet: n_priority_freq\",\n \"matplotlib import cm from warnings import warn from dv3.hparams import hparams, hparams_debug_string fs\",\n \"data x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions =\",\n \"= torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch\",\n \"= checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\"\",\n \"aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep' if info is not\",\n \"as a single speaker dataset if self.speaker_id is not None: indices = np.array(speaker_ids)\",\n \"suffix = \\\"\\\" m = model elif train_seq2seq: suffix = \\\"_seq2seq\\\" m =\",\n \"= list(map(int, self.frame_lengths)) return paths def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def\",\n \"in batch]) done = torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in\",\n \"done, target_lengths, speaker_ids def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path,\",\n \"lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 self.multi_speaker = len(l) ==\",\n \"batches if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm,\",\n \"0: priority_loss = w * masked_l1( y_hat[:, :, :priority_bin], y[:, :, :priority_bin], mask=mask)\",\n \"checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is\",\n \"checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path is not\",\n \"y[:, :, :priority_bin]) l1_loss = (1 - priority_w) * l1_loss + priority_w *\",\n \"mask_.sum() def spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss()\",\n \"to cast numpy.int64 to int self.frame_lengths = list(map(int, self.frame_lengths)) return paths def collect_features(self,\",\n \"linear_loss # attention if train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma)\",\n \"optimizer.param_groups: param_group['lr'] = current_lr optimizer.zero_grad() # Used for Position encoding text_positions, frame_positions =\",\n \"loss\\\", float(mel_loss.data[0]), global_step) writer.add_scalar(\\\"mel_l1_loss\\\", float(mel_l1_loss.data[0]), global_step) writer.add_scalar(\\\"mel_binary_div_loss\\\", float(mel_binary_div.data[0]), global_step) if train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]),\",\n \"* soft_mask).mean() loss += attn_loss if global_step > 0 and global_step % checkpoint_interval\",\n \"import librosa.display from matplotlib import pyplot as plt import sys import os from\",\n \"model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=( hparams.adam_beta1, hparams.adam_beta2), eps=hparams.adam_eps, weight_decay=hparams.weight_decay) if\",\n \"None # Apply model if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn, done_hat =\",\n \"speaker_id = int(speaker_id) if speaker_id is not None else None data_root = args[\\\"--data-root\\\"]\",\n \"0: save_states( global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir) save_checkpoint( model,\",\n \"def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram):\",\n \"dtype=np.int) x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1],\",\n \"= data_root self.col = col self.frame_lengths = [] self.speaker_id = speaker_id def collect_files(self):\",\n \"whole model\\\") train_seq2seq, train_postnet = True, True if train_seq2seq: print(\\\"Training seq2seq model\\\") elif\",\n \"coded speaker_ids = [0, 1, 10] if ismultispeaker else [None] for speaker_id in\",\n \"checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step > 0 and global_step % hparams.eval_interval ==\",\n \"writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq and\",\n \"if sequence_length.is_cuda: seq_range_expand = seq_range_expand.cuda() seq_length_expand = sequence_length.unsqueeze(1) \\\\ .expand_as(seq_range_expand) return (seq_range_expand <\",\n \"mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) # Predicted spectrogram if linear_outputs\",\n \"mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() // r\",\n \"col, speaker_id=None): self.data_root = data_root self.col = col self.frame_lengths = [] self.speaker_id =\",\n \"__init__(self, X, Mel, Y): self.X = X self.Mel = Mel self.Y = Y\",\n \"1: mel = mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths = input_lengths.long().numpy() decoder_lengths =\",\n \"spec_loss(y_hat, y, mask, priority_bin=None, priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w =\",\n \"guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B = len(input_lengths) max_input_len = input_lengths.max() W = np.zeros((B,\",\n \"* 32, len(self.lengths)) if batch_group_size % batch_size != 0: batch_group_size -= batch_group_size %\",\n \"sample_rate=fs) except Exception as e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs, attn,\",\n \"files as well for now alignment_dir = join(checkpoint_dir, \\\"alignment_layer{}\\\".format(i + 1)) os.makedirs(alignment_dir, exist_ok=True)\",\n \"def guided_attention(N, max_N, T, max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32) for n\",\n \"= torch.LongTensor([x[3] for x in batch]) else: speaker_ids = None return x_batch, input_lengths,\",\n \"r:, :], decoder_target_mask) mel_loss = (1 - w) * mel_l1_loss + w *\",\n \"Save averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step))\",\n \"positions text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for x in batch], dtype=np.int)\",\n \"writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) # save files as well for now\",\n \"sys from os.path import dirname, join from tqdm import tqdm, trange from datetime\",\n \"downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion = nn.BCELoss() assert train_seq2seq or train_postnet\",\n \"if use_cuda else soft_mask attn_loss = (attn * soft_mask).mean() loss += attn_loss if\",\n \"checkpoint[\\\"global_epoch\\\"] return model def _load_embedding(path, model): state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data\",\n \"= \\\"_seq2seq\\\" m = model.seq2seq elif train_postnet: suffix = \\\"_postnet\\\" m = model.postnet\",\n \"train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn, done_hat = model( x, mel, speaker_ids=speaker_ids, text_positions=text_positions,\",\n \"is not None: print(\\\"Load optimizer state from {}\\\".format(path)) optimizer.load_state_dict(checkpoint[\\\"optimizer\\\"]) global_step = checkpoint[\\\"global_step\\\"] global_epoch\",\n \"--help Show this help message and exit \\\"\\\"\\\" from docopt import docopt import\",\n \"speaker_ids = torch.LongTensor([x[3] for x in batch]) else: speaker_ids = None return x_batch,\",\n \"assert max_target_len % r == 0 if max_target_len % downsample_step != 0: max_target_len\",\n \"model.cuda() linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr\",\n \"loss if w > 0: assert mask is not None l1_loss = w\",\n \"w * linear_binary_div # Combine losses if train_seq2seq and train_postnet: loss = mel_loss\",\n \"np.array([_pad(np.zeros(len(x[1]) // r // downsample_step - 1), max_decoder_target_len, constant_values=1) for x in batch])\",\n \"save_alignment(path, alignment) # Save averaged alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path\",\n \"eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) # Update loss.backward() if clip_thresh > 0: grad_norm\",\n \"enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i + 1) writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) *\",\n \"== 4: for i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag = \\\"alignment_layer{}\\\".format(i\",\n \"def __init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids = None self.multi_speaker = False\",\n \"fig, ax = plt.subplots() im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax)\",\n \"checkpoint_interval=hparams.checkpoint_interval, nepochs=hparams.nepochs, clip_thresh=hparams.clip_thresh, train_seq2seq=train_seq2seq, train_postnet=train_postnet) except KeyboardInterrupt: save_checkpoint( model, optimizer, global_step, checkpoint_dir, global_epoch,\",\n \"= join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model\",\n \"= hparams.downsample_step current_lr = init_lr binary_criterion = nn.BCELoss() assert train_seq2seq or train_postnet global\",\n \"open(meta, \\\"rb\\\") as f: lines = f.readlines() l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) ==\",\n \"linear_binary_div.data[0]), global_step) if train_seq2seq and hparams.use_guided_attention: writer.add_scalar(\\\"attn_loss\\\", float(attn_loss.data[0]), global_step) if clip_thresh > 0:\",\n \"global global_step, global_epoch while global_epoch < nepochs: running_loss = 0. for step, (x,\",\n \"1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention, use_decoder_state_for_postnet_input=hparams.use_decoder_state_for_postnet_input,\",\n \"__getitem__(self, idx): if self.multi_speaker: text, speaker_id = self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id\",\n \"spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x + eps) - torch.log(1 -\",\n \"self.frame_lengths)) return paths def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root,\",\n \"'\\\\n\\\\n' + info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def\",\n \"signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path) # Target\",\n \"grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq:\",\n \"signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted\",\n \":, :priority_bin], y[:, :, :priority_bin]) l1_loss = (1 - priority_w) * l1_loss +\",\n \"= 0 global_epoch = 0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark = False\",\n \"class _NPYDataSource(FileDataSource): def __init__(self, data_root, col, speaker_id=None): self.data_root = data_root self.col = col\",\n \"len(self.lengths)) if batch_group_size % batch_size != 0: batch_group_size -= batch_group_size % batch_size self.batch_group_size\",\n \"# text positions text_positions = np.array([_pad(np.arange(1, len(x[0]) + 1), max_input_len) for x in\",\n \"= s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm = np.arange(len(indices[:e])\",\n \"schedule if hparams.lr_schedule is not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f(\",\n \"writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda()\",\n \"checkpoint_seq2seq_path is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not None:\",\n \"spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1)\",\n \"r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion = nn.BCELoss() assert\",\n \"% downsample_step assert max_target_len % downsample_step == 0 # Set 0 for zero\",\n \"linear spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x + eps) - torch.log(1\",\n \"pass def save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None): print(\\\"Save intermediate\",\n \"Multi-hop attention if attn is not None and attn.dim() == 4: for i,\",\n \"if checkpoint_path is not None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding if\",\n \"= lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 self.multi_speaker = len(l)\",\n \"n in range(N): for t in range(T): W[n, t] = 1 - np.exp(-(n\",\n \"accepted this as a miracle without any physical explanation.\\\", ] import dv3.synthesis synthesis._frontend\",\n \"tqdm(enumerate(data_loader)): model.train() ismultispeaker = speaker_ids is not None # Learning rate schedule if\",\n \"max_positions=hparams.max_positions, speaker_embedding_weight_std=hparams.speaker_embedding_weight_std, freeze_embedding=hparams.freeze_embedding, window_ahead=hparams.window_ahead, window_backward=hparams.window_backward, key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def load_checkpoint(path, model,\",\n \"r = hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker = len(batch[0]) == 4 # Lengths\",\n \"model = build_model() if use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate, betas=(\",\n \"max_target_len, g=0.2): B = len(input_lengths) max_input_len = input_lengths.max() W = np.zeros((B, max_target_len, max_input_len),\",\n \"w) * linear_l1_loss + w * linear_binary_div # Combine losses if train_seq2seq and\",\n \"= torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size is None: batch_group_size = min(batch_size *\",\n \"None, None # Apply model if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn, done_hat\",\n \"np.zeros((B, max_target_len, max_input_len), dtype=np.float32) for b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b],\",\n \"y[:, :, :priority_bin]) else: priority_loss = l1(y_hat[:, :, :priority_bin], y[:, :, :priority_bin]) l1_loss\",\n \"args[\\\"--train-postnet-only\\\"] # train both if not specified if not train_seq2seq and not train_postnet:\",\n \"return len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len is None: max_len = sequence_length.data.max() batch_size\",\n \"# Handle last elements s += batch_group_size if s < len(indices): random.shuffle(indices[s:]) return\",\n \"= sequence_mask( target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: #\",\n \"= 0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend = None\",\n \"/ mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker\",\n \"= join(eval_output_dir, \\\"step{:09d}_text{}_{}_predicted.wav\\\".format( global_step, idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio signal\",\n \"masked_l1(y_hat, y, mask=mask) + (1 - w) * l1(y_hat, y) else: assert mask\",\n \"+ info plt.xlabel(xlabel) plt.ylabel('Encoder timestep') plt.tight_layout() plt.savefig(path, format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self,\",\n \"import Sampler import numpy as np from numba import jit from nnmnkwii.datasets import\",\n \"= mel[:, 0::downsample_step, :].contiguous() # Lengths input_lengths = input_lengths.long().numpy() decoder_lengths = target_lengths.long().numpy() //\",\n \"current_lr = init_lr binary_criterion = nn.BCELoss() assert train_seq2seq or train_postnet global global_step, global_epoch\",\n \"--train-postnet-only Train only postnet model. --restore-parts=<path> Restore part of the model. --log-event-path=<name> Log\",\n \"return self.X[idx], self.Mel[idx], self.Y[idx] def __len__(self): return len(self.X) def sequence_mask(sequence_length, max_len=None): if max_len\",\n \"is None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand\",\n \"mel spectrogram if downsample_step > 1: mel = mel[:, 0::downsample_step, :].contiguous() # Lengths\",\n \"guided_attention(N, max_N, T, max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32) for n in\",\n \"attention if train_seq2seq and hparams.use_guided_attention: soft_mask = guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask =\",\n \"def masked_mean(y, mask): # (B, T, D) mask_ = mask.expand_as(y) return (y *\",\n \"m.state_dict(), \\\"optimizer\\\": optimizer_state, \\\"global_step\\\": step, \\\"global_epoch\\\": epoch, }, checkpoint_path) print(\\\"Saved checkpoint:\\\", checkpoint_path) def\",\n \"self.batch_size)[perm, :].view(-1) # Handle last elements s += batch_group_size if s < len(indices):\",\n \"mask_.sum() def collate_fn(batch): \\\"\\\"\\\"Create batch\\\"\\\"\\\" r = hparams.outputs_per_step downsample_step = hparams.downsample_step multi_speaker =\",\n \"None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group\",\n \"W[n, t] = 1 - np.exp(-(n / N - t / T)**2 /\",\n \":, :priority_bin]) l1_loss = (1 - priority_w) * l1_loss + priority_w * priority_loss\",\n \"linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def logit(x,\",\n \"None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0) seq_range = torch.arange(0, max_len).long() seq_range_expand =\",\n \"= join( checkpoint_dir, \\\"checkpoint_step{:09d}{}.pth\\\".format(global_step, suffix)) optimizer_state = optimizer.state_dict() if hparams.save_optimizer_state else None torch.save({\",\n \"max_target_len % downsample_step == 0 # Set 0 for zero beginning padding #\",\n \"speaker_ids is None linear_outputs = model.postnet(mel) mel_outputs, attn, done_hat = None, None, None\",\n \"linear_outputs is not None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\",\",\n \"model): state = torch.load(path)[\\\"state_dict\\\"] key = \\\"seq2seq.encoder.embed_tokens.weight\\\" model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def\",\n \"e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs, attn, mel, y, input_lengths, checkpoint_dir=None):\",\n \"n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size // 2 + 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx,\",\n \"RuntimeError(\\\"Should provide either lengths or mask\\\") # (B, T, 1) if mask is\",\n \"batch_group_size % batch_size self.batch_group_size = batch_group_size assert batch_group_size % batch_size == 0 self.permutate\",\n \"getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel\",\n \"def plot_alignment(alignment, path, info=None): fig, ax = plt.subplots() im = ax.imshow( alignment, aspect='auto',\",\n \"mel.size(-1)) linear_outputs = None elif train_postnet: assert speaker_ids is None linear_outputs = model.postnet(mel)\",\n \"len(l) == 4 or len(l) == 5 multi_speaker = len(l) == 5 self.frame_lengths\",\n \"format='png') plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids =\",\n \"max_input_len) for x in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths = torch.LongTensor(input_lengths) target_lengths\",\n \"if checkpoint_restore_parts is not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path is\",\n \"Input dataset definitions X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y =\",\n \"mel_outputs, attn, done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape\",\n \"= args else: text = args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return\",\n \"= target_lengths.long().numpy() // r // downsample_step # Feed data x, mel, y =\",\n \"--reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use specific speaker of\",\n \"mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) # reshape mel_outputs = mel_outputs.view(len(mel), -1, mel.size(-1)) linear_outputs =\",\n \"\\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer = args[\\\"--reset-optimizer\\\"] # Which model to be\",\n \"= sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input) loss = self.criterion(input\",\n \"have accepted this as a miracle without any physical explanation.\\\", ] import dv3.synthesis\",\n \"or mask\\\") # (B, T, 1) if mask is None: mask = sequence_mask(lengths,\",\n \"key_projection=hparams.key_projection, value_projection=hparams.value_projection, ) return model def load_checkpoint(path, model, optimizer, reset_optimizer): global global_step global\",\n \"= 1 - np.exp(-(n / N - t / T)**2 / (2 *\",\n \"seq2seq model from checkpoint path. --checkpoint-postnet=<path> Restore postnet model from checkpoint path. --train-seq2seq-only\",\n \"max_input_len) for x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len //\",\n \"mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x = np.pad(x, [(b_pad, max_len - len(x)\",\n \"preprocessed features. --checkpoint-dir=<dir> Directory where to save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper\",\n \"def time_string(): return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format(\",\n \"0 and global_step % hparams.eval_interval == 0: eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker) #\",\n \"len(batch), max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1]) // r // downsample_step -\",\n \"text_positions = text_positions.cuda() frame_positions = frame_positions.cuda() if train_postnet: y = y.cuda() mel =\",\n \"s < len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def\",\n \"1 running_loss += loss.data[0] averaged_loss = running_loss / (len(data_loader)) writer.add_scalar(\\\"loss (per epoch)\\\", averaged_loss,\",\n \"load_checkpoint(path, model, optimizer, reset_optimizer): global global_step global global_epoch print(\\\"Load checkpoint from: {}\\\".format(path)) checkpoint\",\n \"not None: lr_schedule_f = getattr(dv3.lrschedule, hparams.lr_schedule) current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for\",\n \"specified wrong args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name == \\\"deepvoice3\\\"\",\n \"f), paths)) if multi_speaker and self.speaker_id is not None: speaker_ids = list(map(lambda l:\",\n \"0: # decoder output domain mask decoder_target_mask = sequence_mask( target_lengths / (r *\",\n \"plt.close() class TextDataSource(FileDataSource): def __init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids = None\",\n \"\\\"President Trump met with other leaders at the Group of 20 conference.\\\", \\\"Generative\",\n \"= self.criterion(input * mask_, target * mask_) return loss / mask_.sum() def collate_fn(batch):\",\n \"current_lr = lr_schedule_f( init_lr, global_step, **hparams.lr_schedule_kwargs) for param_group in optimizer.param_groups: param_group['lr'] = current_lr\",\n \"= torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend = None # to be\",\n \"<= 0: binary_div = Variable(y.data.new(1).zero_()) else: y_hat_logits = logit(y_hat) z = -y *\",\n \"1).T) * 255), global_step) # Predicted mel spectrogram if mel_outputs is not None:\",\n \"checkpoint_path) def build_model(): model = getattr(builder, hparams.builder)( n_speakers=hparams.n_speakers, speaker_embed_dim=hparams.speaker_embed_dim, n_vocab=_frontend.n_vocab, embed_dim=hparams.text_embed_dim, mel_dim=hparams.num_mels, linear_dim=hparams.fft_size\",\n \"= spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask) mel_loss = (1 -\",\n \"global_step, checkpoint_dir, global_epoch, train_seq2seq, train_postnet) if global_step > 0 and global_step % hparams.eval_interval\",\n \"torch.log(x + eps) - torch.log(1 - x + eps) def masked_mean(y, mask): #\",\n \"0: assert mask is not None l1_loss = w * masked_l1(y_hat, y, mask=mask)\",\n \"model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if optimizer_state is not None: print(\\\"Load\",\n \"- len(x) - b_pad), (0, 0)], mode=\\\"constant\\\", constant_values=0) return x def plot_alignment(alignment, path,\",\n \"l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 multi_speaker =\",\n \"usage: train.py [options] options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir> Directory where to\",\n \"# If not None, filter by speaker_id self.speaker_id = speaker_id def collect_files(self): meta\",\n \"int(l.decode(\\\"utf-8\\\").split(\\\"|\\\")[-1]), lines)) # Filter by speaker_id # using multi-speaker dataset as a single\",\n \"* 255) def eval_model(global_step, writer, model, checkpoint_dir, ismultispeaker): # harded coded texts =\",\n \"* g * g)) return W def guided_attentions(input_lengths, target_lengths, max_target_len, g=0.2): B =\",\n \"% r assert max_target_len % r == 0 if max_target_len % downsample_step !=\",\n \"mel_outputs, attn, done_hat = None, None, None # Losses w = hparams.binary_divergence_weight #\",\n \"g): W = np.zeros((max_N, max_T), dtype=np.float32) for n in range(N): for t in\",\n \"Update loss.backward() if clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() #\",\n \"= speaker_id def collect_files(self): meta = join(self.data_root, \\\"train.txt\\\") with open(meta, \\\"rb\\\") as f:\",\n \"== 0 # Set 0 for zero beginning padding # imitates initial decoder\",\n \"self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None): if lengths\",\n \"prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) # Target spectrogram if linear_outputs is not\",\n \"deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule import torch\",\n \"None mel_outputs, attn, done_hat, _ = model.seq2seq( x, mel, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) #\",\n \"mel_outputs is not None: mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\",\",\n \"frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch), max_decoder_target_len) # done flags done = np.array([_pad(np.zeros(len(x[1]) //\",\n \"lengths or mask\\\") # (B, T, 1) if mask is None: mask =\",\n \"else \\\"single\\\" for idx, text in enumerate(texts): signal, alignment, _, mel = synthesis.tts(\",\n \"soft_mask attn_loss = (attn * soft_mask).mean() loss += attn_loss if global_step > 0\",\n \"32, len(self.lengths)) if batch_group_size % batch_size != 0: batch_group_size -= batch_group_size % batch_size\",\n \"is not None and attn.dim() == 4: for i, alignment in enumerate(attn): alignment\",\n \"alignment alignment_dir = join(checkpoint_dir, \\\"alignment_ave\\\") os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_alignment.png\\\".format(global_step)) alignment =\",\n \"data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler): \\\"\\\"\\\"Partially randmoized sampler 1. Sort\",\n \"{}\\\".format(global_step)) # idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length\",\n \"global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path =\",\n \"Model model = build_model() if use_cuda: model = model.cuda() optimizer = optim.Adam(model.get_trainable_parameters(), lr=hparams.initial_learning_rate,\",\n \"Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader setup dataset =\",\n \"speaker_ids = [0, 1, 10] if ismultispeaker else [None] for speaker_id in speaker_ids:\",\n \"elif train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat, _ = model.seq2seq( x,\",\n \"return paths def collect_features(self, path): return np.load(path) class MelSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None):\",\n \"mel_output, global_step) # Target spectrogram if linear_outputs is not None: linear_output = y[idx].cpu().data.numpy()\",\n \"binary_div @jit(nopython=True) def guided_attention(N, max_N, T, max_T, g): W = np.zeros((max_N, max_T), dtype=np.float32)\",\n \"a = np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch = torch.LongTensor(a) input_lengths\",\n \"__init__(self): super(MaskedL1Loss, self).__init__() self.criterion = nn.L1Loss(size_average=False) def forward(self, input, target, lengths=None, mask=None, max_len=None):\",\n \"nnmnkwii.datasets import FileSourceDataset, FileDataSource from os.path import join, expanduser import random import librosa.display\",\n \"\\\\\\\"{}\\\\\\\": {}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input\",\n \"None: batch_group_size = min(batch_size * 32, len(self.lengths)) if batch_group_size % batch_size != 0:\",\n \"= nn.BCELoss() assert train_seq2seq or train_postnet global global_step, global_epoch while global_epoch < nepochs:\",\n \"guided_attentions(input_lengths, decoder_lengths, attn.size(-2), g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else\",\n \"np.max(np.abs(signal)) # Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag\",\n \"speaker_str), signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def save_states(global_step, writer,\",\n \"mel_output = mel_outputs[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Predicted mel spectrogram\\\", mel_output, global_step) # Predicted\",\n \"info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram =\",\n \"g=hparams.guided_attention_sigma) soft_mask = Variable(torch.from_numpy(soft_mask)) soft_mask = soft_mask.cuda() if use_cuda else soft_mask attn_loss =\",\n \"* batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate:\",\n \"= np.array(speaker_ids) == self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need\",\n \"loss += attn_loss if global_step > 0 and global_step % checkpoint_interval == 0:\",\n \"None: mask = sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input) loss\",\n \"args[0] seq = _frontend.text_to_sequence(text, p=hparams.replace_pronunciation_prob) if self.multi_speaker: return np.asarray(seq, dtype=np.int32), int(speaker_id) else: return\",\n \"= FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) # Prepare sampler frame_lengths = Mel.file_data_source.frame_lengths\",\n \"s = i * batch_group_size e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate\",\n \"speaker_ids=speaker_ids, text_positions=text_positions, frame_positions=frame_positions, input_lengths=input_lengths) elif train_seq2seq: assert speaker_ids is None mel_outputs, attn, done_hat,\",\n \"r:, :] target_mask = target_mask[:, r:, :] else: decoder_target_mask, target_mask = None, None\",\n \"self.sorted_indices.clone() batch_group_size = self.batch_group_size s, e = 0, 0 for i in range(len(indices)\",\n \"is not None: load_checkpoint(checkpoint_seq2seq_path, model.seq2seq, optimizer, reset_optimizer) if checkpoint_path is not None: load_checkpoint(checkpoint_path,\",\n \"attn.dim() == 4: for i, alignment in enumerate(attn): alignment = alignment[idx].cpu().data.numpy() tag =\",\n \"not train_postnet: print(\\\"Training whole model\\\") train_seq2seq, train_postnet = True, True if train_seq2seq: print(\\\"Training\",\n \"log_event_path is None: log_event_path = \\\"log/run-test\\\" + str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path:\",\n \"= np.arange(len(indices[:e]) // self.batch_size) random.shuffle(perm) indices[:e] = indices[:e].view(-1, self.batch_size)[perm, :].view(-1) # Handle last\",\n \":], y[:, r:, :], target_mask, priority_bin=n_priority_freq, priority_w=hparams.priority_freq_weight) linear_loss = (1 - w) *\",\n \"Feed data x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions) frame_positions\",\n \"= model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion =\",\n \"= positions # Downsample mel spectrogram if downsample_step > 1: mel = mel[:,\",\n \"clip_thresh > 0: grad_norm = torch.nn.utils.clip_grad_norm( model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]),\",\n \"// r // downsample_step # Feed data x, mel, y = Variable(x), Variable(mel),\",\n \"https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part of the model from: {}\\\".format(path)) state =\",\n \"X = FileSourceDataset(TextDataSource(data_root, speaker_id)) Mel = FileSourceDataset(MelSpecDataSource(data_root, speaker_id)) Y = FileSourceDataset(LinearSpecDataSource(data_root, speaker_id)) #\",\n \"idx = np.random.randint(0, len(input_lengths)) idx = min(1, len(input_lengths) - 1) input_length = input_lengths[idx]\",\n \"(x, input_lengths, mel, y, positions, done, target_lengths, speaker_ids) \\\\ in tqdm(enumerate(data_loader)): model.train() ismultispeaker\",\n \"== self.speaker_id paths = list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast\",\n \"checkpoint_dir = args[\\\"--checkpoint-dir\\\"] checkpoint_path = args[\\\"--checkpoint\\\"] checkpoint_seq2seq_path = args[\\\"--checkpoint-seq2seq\\\"] checkpoint_postnet_path = args[\\\"--checkpoint-postnet\\\"] load_embedding\",\n \"lengths, batch_size=16, batch_group_size=None, permutate=True): self.lengths, self.sorted_indices = torch.sort(torch.LongTensor(lengths)) self.batch_size = batch_size if batch_group_size\",\n \"flip against freq axis return np.uint8(cm.magma(spectrogram.T) * 255) def eval_model(global_step, writer, model, checkpoint_dir,\",\n \"t] = 1 - np.exp(-(n / N - t / T)**2 / (2\",\n \"im = ax.imshow( alignment, aspect='auto', origin='lower', interpolation='none') fig.colorbar(im, ax=ax) xlabel = 'Decoder timestep'\",\n \"import sys from os.path import dirname, join from tqdm import tqdm, trange from\",\n \"train_postnet global global_step, global_epoch while global_epoch < nepochs: running_loss = 0. for step,\",\n \"range(T): W[n, t] = 1 - np.exp(-(n / N - t / T)**2\",\n \"is not None and hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset))\",\n \"harded coded texts = [ \\\"Scientists at the CERN laboratory say they have\",\n \"str(datetime.now()).replace(\\\" \\\", \\\"_\\\") print(\\\"Los event path: {}\\\".format(log_event_path)) writer = SummaryWriter(log_dir=log_event_path) # Train! try:\",\n \"signal {}_{}\\\".format(idx, speaker_str), signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def\",\n \"of the model. --log-event-path=<name> Log event path. --reset-optimizer Reset optimizer. --load-embedding=<path> Load embedding\",\n \"{}\\\".format( hparams.preset, json.dumps(preset, indent=4))) _frontend = getattr(frontend, hparams.frontend) os.makedirs(checkpoint_dir, exist_ok=True) # Input dataset\",\n \"= list(np.array(paths)[indices]) self.frame_lengths = list(np.array(self.frame_lengths)[indices]) # aha, need to cast numpy.int64 to int\",\n \"setup dataset = PyTorchDataset(X, Mel, Y) data_loader = data_utils.DataLoader( dataset, batch_size=hparams.batch_size, num_workers=hparams.num_workers, sampler=sampler,\",\n \"\\\"\\\"\\\"Partially randmoized sampler 1. Sort by lengths 2. Pick a small patch and\",\n \"None: linear_output = y[idx].cpu().data.numpy() spectrogram = prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def\",\n \"or train_postnet global global_step, global_epoch while global_epoch < nepochs: running_loss = 0. for\",\n \"else None # Create mask if we use masked loss if hparams.masked_loss_weight >\",\n \"text-to-speech synthesis model. usage: train.py [options] options: --data-root=<dir> Directory contains preprocessed features. --checkpoint-dir=<dir>\",\n \"mel_output, global_step) # Predicted spectrogram if linear_outputs is not None: linear_output = linear_outputs[idx].cpu().data.numpy()\",\n \"max_target_len % r assert max_target_len % r == 0 if max_target_len % downsample_step\",\n \"if __name__ == \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args) checkpoint_dir =\",\n \"return datetime.now().strftime('%Y-%m-%d %H:%M') def save_alignment(path, attn): plot_alignment(attn.T, path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(),\",\n \"= None, None # Apply model if train_seq2seq and train_postnet: mel_outputs, linear_outputs, attn,\",\n \"for b in range(B): W[b] = guided_attention(input_lengths[b], max_input_len, target_lengths[b], max_target_len, g).T return W\",\n \"y_hat_logits + torch.log(1 + torch.exp(y_hat_logits)) if w > 0: binary_div = w *\",\n \"downsample_step a = np.array([_pad(x[0], max_input_len) for x in batch], dtype=np.int) x_batch = torch.LongTensor(a)\",\n \"+ batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm = np.arange(len(indices[:e]) // self.batch_size)\",\n \"+= 1 def save_checkpoint(model, optimizer, step, checkpoint_dir, epoch, train_seq2seq, train_postnet): if train_seq2seq and\",\n \"False # If not None, filter by speaker_id self.speaker_id = speaker_id def collect_files(self):\",\n \"= torch.LongTensor(input_lengths) target_lengths = torch.LongTensor(target_lengths) b = np.array([_pad_2d(x[1], max_target_len, b_pad=b_pad) for x in\",\n \"s, e = 0, 0 for i in range(len(indices) // batch_group_size): s =\",\n \"train_seq2seq: mel_l1_loss, mel_binary_div = spec_loss( mel_outputs[:, :-r, :], mel[:, r:, :], decoder_target_mask) mel_loss\",\n \"grad_norm, global_step) writer.add_scalar(\\\"learning rate\\\", current_lr, global_step) global_step += 1 running_loss += loss.data[0] averaged_loss\",\n \"idx, speaker_str)) dv3.audio.save_wav(signal, path) try: writer.add_audio(\\\"(Eval) Predicted audio signal {}_{}\\\".format(idx, speaker_str), signal, global_step,\",\n \"target, lengths=None, mask=None, max_len=None): if lengths is None and mask is None: raise\",\n \"l.decode(\\\"utf-8\\\").split(\\\"|\\\")[self.col], lines)) paths = list(map(lambda f: join(self.data_root, f), paths)) if multi_speaker and self.speaker_id\",\n \"_frontend eval_output_dir = join(checkpoint_dir, \\\"eval\\\") os.makedirs(eval_output_dir, exist_ok=True) # hard coded speaker_ids = [0,\",\n \"< len(indices): random.shuffle(indices[s:]) return iter(indices) def __len__(self): return len(self.sorted_indices) class PyTorchDataset(object): def __init__(self,\",\n \"Reset optimizer. --load-embedding=<path> Load embedding from checkpoint. --speaker-id=<N> Use specific speaker of data\",\n \"= Y # alias self.multi_speaker = X.file_data_source.multi_speaker def __getitem__(self, idx): if self.multi_speaker: text,\",\n \"train(model, data_loader, optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda:\",\n \"None data_root = args[\\\"--data-root\\\"] if data_root is None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\")\",\n \"= torch.FloatTensor(done).unsqueeze(-1) if multi_speaker: speaker_ids = torch.LongTensor([x[3] for x in batch]) else: speaker_ids\",\n \"e = s - 1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand( len(batch),\",\n \"if data_root is None: data_root = join(dirname(__file__), \\\"data\\\", \\\"ljspeech\\\") log_event_path = args[\\\"--log-event-path\\\"] reset_optimizer\",\n \"\\\"step{:09d}_alignment.png\\\".format(global_step)) alignment = attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) *\",\n \"D) mask_ = mask.expand_as(input) loss = self.criterion(input * mask_, target * mask_) return\",\n \"checkpoint path. --train-seq2seq-only Train only seq2seq model. --train-postnet-only Train only postnet model. --restore-parts=<path>\",\n \"a single speaker dataset if self.speaker_id is not None: indices = np.array(speaker_ids) ==\",\n \"init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model = model.cuda() linear_dim\",\n \"speaker_id) class LinearSpecDataSource(_NPYDataSource): def __init__(self, data_root, speaker_id=None): super(LinearSpecDataSource, self).__init__(data_root, 0, speaker_id) class PartialyRandomizedSimilarTimeLengthSampler(Sampler):\",\n \"mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) # Target spectrogram if\",\n \"if max_target_len % downsample_step != 0: max_target_len += downsample_step - max_target_len % downsample_step\",\n \"l = lines[0].decode(\\\"utf-8\\\").split(\\\"|\\\") assert len(l) == 4 or len(l) == 5 self.multi_speaker =\",\n \"% r != 0: max_target_len += r - max_target_len % r assert max_target_len\",\n \"assert mask is not None l1_loss = w * masked_l1(y_hat, y, mask=mask) +\",\n \"1)) os.makedirs(alignment_dir, exist_ok=True) path = join(alignment_dir, \\\"step{:09d}_layer_{}_alignment.png\\\".format( global_step, i + 1)) save_alignment(path, alignment)\",\n \"\\\"must be specified wrong args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"]) print(hparams_debug_string()) assert hparams.name\",\n \"y, mask=mask) + (1 - w) * l1(y_hat, y) else: assert mask is\",\n \"t in range(T): W[n, t] = 1 - np.exp(-(n / N - t\",\n \"max_target_len, b_pad=b_pad) for x in batch], dtype=np.float32) mel_batch = torch.FloatTensor(b) c = np.array([_pad_2d(x[2],\",\n \"= np.zeros((max_N, max_T), dtype=np.float32) for n in range(N): for t in range(T): W[n,\",\n \"not None and hparams.preset != \\\"\\\": preset = hparams.presets[hparams.preset] import json hparams.parse_json(json.dumps(preset)) print(\\\"Override\",\n \"= mel[idx].cpu().data.numpy() mel_output = prepare_spec_image(dv3.audio._denormalize(mel_output)) writer.add_image(\\\"Target mel spectrogram\\\", mel_output, global_step) # Target spectrogram\",\n \"def __init__(self, X, Mel, Y): self.X = X self.Mel = Mel self.Y =\",\n \"max(target_lengths) if max_target_len % r != 0: max_target_len += r - max_target_len %\",\n \"mel = mel.cuda() done, target_lengths = done.cuda(), target_lengths.cuda() speaker_ids = speaker_ids.cuda() if ismultispeaker\",\n \"def sequence_mask(sequence_length, max_len=None): if max_len is None: max_len = sequence_length.data.max() batch_size = sequence_length.size(0)\",\n \"s, e = s - 1, e - 1 frame_positions = torch.arange(s, e).long().unsqueeze(0).expand(\",\n \"# Presets if hparams.preset is not None and hparams.preset != \\\"\\\": preset =\",\n \"audio signal\\\", signal, global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass dv3.audio.save_wav(signal, path)\",\n \"optimizer, writer, init_lr=0.002, checkpoint_dir=None, checkpoint_interval=None, nepochs=None, clip_thresh=1.0, train_seq2seq=True, train_postnet=True): if use_cuda: model =\",\n \"\\\"Please call Stella.\\\", \\\"Some have accepted this as a miracle without any physical\",\n \"datetime import datetime # The deepvoice3 model from dv3.deepvoice3_pytorch import frontend, builder import\",\n \":].view(-1) # Handle last elements s += batch_group_size if s < len(indices): random.shuffle(indices[s:])\",\n \"__init__(self, data_root, speaker_id=None): self.data_root = data_root self.speaker_ids = None self.multi_speaker = False #\",\n \"# (B, T, D) mask_ = mask.expand_as(input) loss = self.criterion(input * mask_, target\",\n \"self.multi_speaker = False return texts return texts, speaker_ids else: return texts def collect_features(self,\",\n \"= prepare_spec_image(dv3.audio._denormalize(linear_output)) writer.add_image(\\\"Target linear spectrogram\\\", spectrogram, global_step) def logit(x, eps=1e-8): return torch.log(x +\",\n \"dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r // downsample_step # frame\",\n \"priority_w=0): masked_l1 = MaskedL1Loss() l1 = nn.L1Loss() w = hparams.masked_loss_weight # L1 loss\",\n \"np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts return texts,\",\n \"# Feed data x, mel, y = Variable(x), Variable(mel), Variable(y) text_positions = Variable(text_positions)\",\n \"indices = np.array(speaker_ids) == self.speaker_id texts = list(np.array(texts)[indices]) self.multi_speaker = False return texts\",\n \"1).T) * 255), global_step) # save files as well for now alignment_dir =\",\n \"# Alignment path = join(eval_output_dir, \\\"step{:09d}_text{}_{}_alignment.png\\\".format( global_step, idx, speaker_str)) save_alignment(path, alignment) tag =\",\n \"sampler frame_lengths = Mel.file_data_source.frame_lengths sampler = PartialyRandomizedSimilarTimeLengthSampler( frame_lengths, batch_size=hparams.batch_size) # Dataset and Dataloader\",\n \"model.get_trainable_parameters(), clip_thresh) optimizer.step() # Logs writer.add_scalar(\\\"loss\\\", float(loss.data[0]), global_step) if train_seq2seq: writer.add_scalar(\\\"done_loss\\\", float(done_loss.data[0]), global_step)\",\n \"train_postnet: writer.add_scalar(\\\"linear_loss\\\", float(linear_loss.data[0]), global_step) writer.add_scalar(\\\"linear_l1_loss\\\", float(linear_l1_loss.data[0]), global_step) writer.add_scalar(\\\"linear_binary_div_loss\\\", float( linear_binary_div.data[0]), global_step) if train_seq2seq\",\n \"target_lengths / (r * downsample_step), max_len=mel.size(1)).unsqueeze(-1) if downsample_step > 1: # spectrogram-domain mask\",\n \"from: {}\\\".format(path)) checkpoint = torch.load(path) model.load_state_dict(checkpoint[\\\"state_dict\\\"]) if not reset_optimizer: optimizer_state = checkpoint[\\\"optimizer\\\"] if\",\n \"= self.X[idx] return text, self.Mel[idx], self.Y[idx], speaker_id else: return self.X[idx], self.Mel[idx], self.Y[idx] def\",\n \"for x in batch], dtype=np.int) text_positions = torch.LongTensor(text_positions) max_decoder_target_len = max_target_len // r\",\n \"'Decoder timestep' if info is not None: xlabel += '\\\\n\\\\n' + info plt.xlabel(xlabel)\",\n \"global_step, i + 1)) save_alignment(path, alignment) # Save averaged alignment alignment_dir = join(checkpoint_dir,\",\n \"spectrogram\\\", spectrogram, global_step) # Predicted audio signal signal = dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal))\",\n \"model.seq2seq.encoder.embed_tokens.weight.data = state[key] # https://discuss.pytorch.org/t/how-to-load-part-of-pre-trained-model/1113/3 def restore_parts(path, model): print(\\\"Restore part of the model\",\n \"path) # Target mel spectrogram if mel_outputs is not None: mel_output = mel[idx].cpu().data.numpy()\",\n \"speaker of data in case for multi-speaker datasets. -h, --help Show this help\",\n \"--checkpoint-dir=<dir> Directory where to save model checkpoints [default: checkpoints]. --hparams=<parmas> Hyper parameters [default:\",\n \"self.batch_group_size = batch_group_size assert batch_group_size % batch_size == 0 self.permutate = permutate def\",\n \"seq2seq model\\\") elif train_postnet: print(\\\"Training postnet model\\\") else: assert False, \\\"must be specified\",\n \"% batch_size == 0 self.permutate = permutate def __iter__(self): indices = self.sorted_indices.clone() batch_group_size\",\n \"- w) * l1(y_hat, y) else: assert mask is None l1_loss = l1(y_hat,\",\n \"from torch.utils import data as data_utils from torch.utils.data.sampler import Sampler import numpy as\",\n \"model_dict.update(valid_state_dict) model.load_state_dict(model_dict) if __name__ == \\\"__main__\\\": args = docopt(__doc__) print(\\\"Command line args:\\\\n\\\", args)\",\n \"global_epoch = 0 use_cuda = torch.cuda.is_available() if use_cuda: cudnn.benchmark = False _frontend =\",\n \"= args[\\\"--train-seq2seq-only\\\"] train_postnet = args[\\\"--train-postnet-only\\\"] # train both if not specified if not\",\n \"dv3.deepvoice3_pytorch import frontend, builder import dv3.audio import dv3.lrschedule import torch from torch.utils import\",\n \"linear_dim = model.linear_dim r = hparams.outputs_per_step downsample_step = hparams.downsample_step current_lr = init_lr binary_criterion\",\n \"import pyplot as plt import sys import os from tensorboardX import SummaryWriter from\",\n \"path, info=\\\"{}, {}, step={}\\\".format( hparams.builder, time_string(), global_step)) def prepare_spec_image(spectrogram): # [0, 1] spectrogram\",\n \"attn.mean(0)[idx].cpu().data.numpy() save_alignment(path, alignment) tag = \\\"averaged_alignment\\\" writer.add_image(tag, np.uint8(cm.viridis(np.flip(alignment, 1).T) * 255), global_step) #\",\n \"+ 1, r=hparams.outputs_per_step, downsample_step=hparams.downsample_step, padding_idx=hparams.padding_idx, dropout=hparams.dropout, kernel_size=hparams.kernel_size, encoder_channels=hparams.encoder_channels, decoder_channels=hparams.decoder_channels, converter_channels=hparams.converter_channels, use_memory_mask=hparams.use_memory_mask, trainable_positional_encodings=hparams.trainable_positional_encodings, force_monotonic_attention=hparams.force_monotonic_attention,\",\n \"is not None: restore_parts(checkpoint_restore_parts, model) # Load checkpoints if checkpoint_postnet_path is not None:\",\n \"def prepare_spec_image(spectrogram): # [0, 1] spectrogram = (spectrogram - np.min(spectrogram)) / (np.max(spectrogram) -\",\n \"Create mask if we use masked loss if hparams.masked_loss_weight > 0: # decoder\",\n \"return np.pad(seq, (0, max_len - len(seq)), mode='constant', constant_values=constant_values) def _pad_2d(x, max_len, b_pad=0): x\",\n \"(spectrogram - np.min(spectrogram)) / (np.max(spectrogram) - np.min(spectrogram)) spectrogram = np.flip(spectrogram, axis=1) # flip\",\n \"dv3.audio.inv_spectrogram(linear_output.T) signal /= np.max(np.abs(signal)) path = join(checkpoint_dir, \\\"step{:09d}_predicted.wav\\\".format( global_step)) try: writer.add_audio(\\\"Predicted audio signal\\\",\",\n \"else: assert False, \\\"must be specified wrong args\\\" # Override hyper parameters hparams.parse(args[\\\"--hparams\\\"])\",\n \"None: load_checkpoint(checkpoint_path, model, optimizer, reset_optimizer) # Load embedding if load_embedding is not None:\",\n \"def logit(x, eps=1e-8): return torch.log(x + eps) - torch.log(1 - x + eps)\",\n \"frame_positions.cuda() if train_postnet: y = y.cuda() mel = mel.cuda() done, target_lengths = done.cuda(),\",\n \"speaker_id # using multi-speaker dataset as a single speaker dataset indices = np.array(speaker_ids)\",\n \"sequence_mask(lengths, max_len).unsqueeze(-1) # (B, T, D) mask_ = mask.expand_as(input) loss = self.criterion(input *\",\n \"e = s + batch_group_size random.shuffle(indices[s:e]) # Permutate batches if self.permutate: perm =\",\n \"frame_positions = frame_positions.cuda() if train_postnet: y = y.cuda() mel = mel.cuda() done, target_lengths\",\n \"assert len(l) == 4 or len(l) == 5 self.multi_speaker = len(l) == 5\"\n]","isConversational":false}}},{"rowIdx":2691,"cells":{"ngram":{"kind":"list like","value":["screen.blit(write(\"Loading index... %fs\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as","= pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size = 40):","(255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0))","pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img),","255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img =","- start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) #","recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000)","faces, capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(),","= find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture,","os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock clock = pygame.time.Clock() # Init","/usr/bin/python2 import time start = time.time() import pygame, numpy import pygame.camera # Init","\"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock clock = pygame.time.Clock() # Init camera","True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount:","== 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif","if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture)","= recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip()","(time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img)))","for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s","fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40): if font_size not in","= [] screen.blit(write(\"Loading index... %fs\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir,","time.time() import pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN)","index.insert(description, img) screen.blit(write(\"Loading images... %fs\" % (time.time() - start)), (0,50)) pygame.display.flip() for img","def main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for event","in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox","line = line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in","= font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI import","main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for event in","import pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic","found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen","line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\" % (time.time() - start)), (0,50)) pygame.display.flip() for","# 30s def main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60)","open(os.path.join(imgdir, \"index.tsv\")) as f: for line in f: line = line.strip().split(\"\\t\") img =","0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face)","= 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown =","else \"images\" photo_samples = [] screen.blit(write(\"Loading index... %fs\" % (time.time() - start)), (0,0))","= time.time() import pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0),","not cameras: pygame.quit() print \"No cameras found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start()","font_size not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour","(time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f: for line in","0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if","screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown =","in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data","elif countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown","index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount","\"images\" photo_samples = [] screen.blit(write(\"Loading index... %fs\" % (time.time() - start)), (0,0)) pygame.display.flip()","10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10","# Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font","# Init font pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\",","line in f: line = line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n)","find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255,","cameras: pygame.quit() print \"No cameras found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() #","sys.argv[1] if len(sys.argv) > 1 else \"images\" photo_samples = [] screen.blit(write(\"Loading index... %fs\"","event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for","% (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\",","images... %fs\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")):","description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\" %","pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186 fonts","= 16, 117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40):","40)} def font(font_size = 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\",","= pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face =","pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height())","= numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree()","recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \"images\" photo_samples = []","= pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit()","faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip()","Init AI import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture","countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -=","font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size =","f: line = line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n","117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40): if font_size","pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock clock","camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen,","while True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return","for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0,","n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\" % (time.time() - start)), (0,50))","for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock clock =","face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def","f: for line in f: line = line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description","AI import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture =","pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) !=","Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255,","Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if","write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init","fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size","screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour =","#! /usr/bin/python2 import time start = time.time() import pygame, numpy import pygame.camera #","def font(font_size = 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size)","pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size","pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print","(0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount = 0","- start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f: for line in f:","return font(font_size).render(str(text), True, colour) # Init AI import recognition import sys, os def","# Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras()","pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount = 0 while","countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get():","pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186","font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI import recognition","print \"No cameras found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def","Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \"No cameras","1 else \"images\" photo_samples = [] screen.blit(write(\"Loading index... %fs\" % (time.time() - start)),","os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture),","True, colour) # Init AI import recognition import sys, os def find_faces(pygame_capture): capture","font_size) return fonts[font_size] def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text),","pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data =","os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images...","faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2)","capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index =","= len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown),","for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image()","len(sys.argv) > 1 else \"images\" photo_samples = [] screen.blit(write(\"Loading index... %fs\" % (time.time()","# Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255,","0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown","186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40): if font_size not","pygame.camera.list_cameras() if not cameras: pygame.quit() print \"No cameras found, exiting!\" sys.exit(1) camera =","import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) #","40): return font(font_size).render(str(text), True, colour) # Init AI import recognition import sys, os","\"No cameras found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture,","(255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces)","if not cameras: pygame.quit() print \"No cameras found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0])","= line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]])","fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face)","capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir =","recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0,","pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size = 40): return","pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init","16, 117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40): if","def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255),","screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for","sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return","return fonts[font_size] def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True,","font(font_size).render(str(text), True, colour) # Init AI import recognition import sys, os def find_faces(pygame_capture):","display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour","== 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1","def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) #","font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI import recognition import","not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour =","(0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock","(pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in","colour) # Init AI import recognition import sys, os def find_faces(pygame_capture): capture =","img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown =","if font_size not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text,","pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces:","pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\")","screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10","screen.blit(write(\"Loading images... %fs\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir,","= {40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40): if font_size not in fonts:","= camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(),","line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description,","pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \"No cameras found, exiting!\"","screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount =","camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \"No cameras found,","pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount = 0 while True:","%fs\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir,","recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen)","pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117,","Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186 fonts =","fonts[font_size] def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour)","in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock clock = pygame.time.Clock() #","= os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading","len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces)","or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown ==","= recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \"images\" photo_samples =","> 1 else \"images\" photo_samples = [] screen.blit(write(\"Loading index... %fs\" % (time.time() -","index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \"images\" photo_samples","capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \"images\"","lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get(): if event.type in","photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init()","sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0,","len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown == 0:","\"thumbnails\", img))) # Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras","bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture,","numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\" % (time.time() -","= pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(),","clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture =","= pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16,","img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init clock clock = pygame.time.Clock()","= 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size]","pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40): if font_size not in fonts: fonts[font_size] =","# Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \"No","pygame.quit() print \"No cameras found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop","screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face","in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour = font_colour,","colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI","import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture,","recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else","start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \"thumbnails\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \"thumbnails\", img))) # Init","pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in","cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \"No cameras found, exiting!\" sys.exit(1)","255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img","= numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1]","capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height())","rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture","pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f: for line in f: line = line.strip().split(\"\\t\")","img) screen.blit(write(\"Loading images... %fs\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in","start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f: for line in f: line","0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip()","10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__ == \"__main__\":","1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) >","= pygame.camera.list_cameras() if not cameras: pygame.quit() print \"No cameras found, exiting!\" sys.exit(1) camera","40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def","recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit()","= numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\" % (time.time()","\"index.tsv\")) as f: for line in f: line = line.strip().split(\"\\t\") img = os.path.join(imgdir,","clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not","faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if","= 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__ ==","event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces,","10 lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get(): if event.type","in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main():","find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index","img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img)","Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False) # Init font pygame.font.init()","fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size =","return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1","= sys.argv[1] if len(sys.argv) > 1 else \"images\" photo_samples = [] screen.blit(write(\"Loading index...","[] screen.blit(write(\"Loading index... %fs\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\"))","with open(os.path.join(imgdir, \"index.tsv\")) as f: for line in f: line = line.strip().split(\"\\t\") img","= 40): return font(font_size).render(str(text), True, colour) # Init AI import recognition import sys,","in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\" % (time.time() - start)), (0,50)) pygame.display.flip()","cameras found, exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces):","clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras:","camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0,","0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces)","capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or","len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0))","numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\"Magic Mirror\") #pygame.mouse.set_visible(False)","False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown","img))) # Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras =","30s def main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for","= 10 lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get(): if","countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__","def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture","exiting!\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen =","True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture","lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces)","return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect","capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect =","!= lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data,","import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1)","bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True,","face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) #","if len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount =","numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if","fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face","(0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount","face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0))","bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0),","rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) ==","else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__ == \"__main__\": main()","font(font_size = 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\"Futura.ttc\", font_size) return","0 while True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN):","camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(),","Init font pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)}","# Init AI import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture))","start = time.time() import pygame, numpy import pygame.camera # Init display screen =","in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect,","index... %fs\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f:","= pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture =","= 0 while True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT,","pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture,","if len(sys.argv) > 1 else \"images\" photo_samples = [] screen.blit(write(\"Loading index... %fs\" %","{40: pygame.font.Font(\"Futura.ttc\", 40)} def font(font_size = 40): if font_size not in fonts: fonts[font_size]","as f: for line in f: line = line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0])","pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()),","in f: line = line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for","import time start = time.time() import pygame, numpy import pygame.camera # Init display","bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False)","= pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces)","0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv)","% (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f: for line","photo_samples = [] screen.blit(write(\"Loading index... %fs\" % (time.time() - start)), (0,0)) pygame.display.flip() with","numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir","for line in f: line = line.strip().split(\"\\t\") img = os.path.join(imgdir, line[0]) description =","for n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\" % (time.time() - start)),","pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture,","faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown","countdown = 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown","#pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186 fonts = {40:","time start = time.time() import pygame, numpy import pygame.camera # Init display screen","imgdir = sys.argv[1] if len(sys.argv) > 1 else \"images\" photo_samples = [] screen.blit(write(\"Loading","line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\"Loading images... %fs\"","= index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10","lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10 else:","(0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f: for line in f: line =","%fs\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as f: for","2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0","font pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\"Futura.ttc\", 40)} def"],"string":"[\n \"screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as\",\n \"= pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size = 40):\",\n \"(255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0))\",\n \"pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img),\",\n \"255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img =\",\n \"- start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) #\",\n \"recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000)\",\n \"faces, capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(),\",\n \"= find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture,\",\n \"os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock clock = pygame.time.Clock() # Init\",\n \"/usr/bin/python2 import time start = time.time() import pygame, numpy import pygame.camera # Init\",\n \"\\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock clock = pygame.time.Clock() # Init camera\",\n \"True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount:\",\n \"== 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif\",\n \"if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture)\",\n \"= recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip()\",\n \"(time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img)))\",\n \"for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s\",\n \"fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40): if font_size not in\",\n \"= [] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir,\",\n \"time.time() import pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN)\",\n \"index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() - start)), (0,50)) pygame.display.flip() for img\",\n \"def main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for event\",\n \"in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox\",\n \"line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in\",\n \"= font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI import\",\n \"main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for event in\",\n \"import pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic\",\n \"found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen\",\n \"line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() - start)), (0,50)) pygame.display.flip() for\",\n \"# 30s def main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60)\",\n \"open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line in f: line = line.strip().split(\\\"\\\\t\\\") img =\",\n \"0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face)\",\n \"= 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown =\",\n \"else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() - start)), (0,0))\",\n \"= time.time() import pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0),\",\n \"not cameras: pygame.quit() print \\\"No cameras found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start()\",\n \"font_size not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text, colour\",\n \"(time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line in\",\n \"0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if\",\n \"screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown =\",\n \"in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data\",\n \"elif countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown\",\n \"index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount\",\n \"\\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() - start)), (0,0)) pygame.display.flip()\",\n \"10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10\",\n \"# Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font\",\n \"# Init font pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\",\",\n \"line in f: line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n)\",\n \"find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255,\",\n \"cameras: pygame.quit() print \\\"No cameras found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() #\",\n \"sys.argv[1] if len(sys.argv) > 1 else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\"\",\n \"event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for\",\n \"% (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\",\",\n \"images... %fs\\\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")):\",\n \"description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" %\",\n \"pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186 fonts\",\n \"= 16, 117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40):\",\n \"40)} def font(font_size = 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\",\",\n \"= pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face =\",\n \"pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height())\",\n \"= numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree()\",\n \"recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \\\"images\\\" photo_samples = []\",\n \"= pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit()\",\n \"faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip()\",\n \"Init AI import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture\",\n \"countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -=\",\n \"font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size =\",\n \"f: line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n\",\n \"117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40): if font_size\",\n \"pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock clock\",\n \"camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen,\",\n \"while True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return\",\n \"for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0,\",\n \"n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() - start)), (0,50))\",\n \"for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock clock =\",\n \"face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def\",\n \"f: for line in f: line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description\",\n \"AI import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture =\",\n \"pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) !=\",\n \"Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255,\",\n \"Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if\",\n \"write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init\",\n \"fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size\",\n \"screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour =\",\n \"#! /usr/bin/python2 import time start = time.time() import pygame, numpy import pygame.camera #\",\n \"def font(font_size = 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size)\",\n \"pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size\",\n \"pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print\",\n \"(0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount = 0\",\n \"- start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line in f:\",\n \"return font(font_size).render(str(text), True, colour) # Init AI import recognition import sys, os def\",\n \"# Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras()\",\n \"pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount = 0 while\",\n \"countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get():\",\n \"pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186\",\n \"font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI import recognition\",\n \"print \\\"No cameras found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def\",\n \"Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \\\"No cameras\",\n \"1 else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() - start)),\",\n \"os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture),\",\n \"True, colour) # Init AI import recognition import sys, os def find_faces(pygame_capture): capture\",\n \"font_size) return fonts[font_size] def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text),\",\n \"pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data =\",\n \"os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images...\",\n \"faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2)\",\n \"capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index =\",\n \"= len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown),\",\n \"for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image()\",\n \"len(sys.argv) > 1 else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time()\",\n \"# Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255,\",\n \"0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown\",\n \"186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40): if font_size not\",\n \"pygame.camera.list_cameras() if not cameras: pygame.quit() print \\\"No cameras found, exiting!\\\" sys.exit(1) camera =\",\n \"import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) #\",\n \"40): return font(font_size).render(str(text), True, colour) # Init AI import recognition import sys, os\",\n \"\\\"No cameras found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture,\",\n \"(255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces)\",\n \"if not cameras: pygame.quit() print \\\"No cameras found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0])\",\n \"= line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]])\",\n \"fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face)\",\n \"capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir =\",\n \"recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0,\",\n \"pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size = 40): return\",\n \"pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init\",\n \"16, 117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40): if\",\n \"def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255),\",\n \"screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for\",\n \"sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return\",\n \"return fonts[font_size] def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True,\",\n \"font(font_size).render(str(text), True, colour) # Init AI import recognition import sys, os def find_faces(pygame_capture):\",\n \"display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour\",\n \"== 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1\",\n \"def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) #\",\n \"font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI import recognition import\",\n \"not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text, colour =\",\n \"(0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock\",\n \"(pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in\",\n \"colour) # Init AI import recognition import sys, os def find_faces(pygame_capture): capture =\",\n \"img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown =\",\n \"if font_size not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text,\",\n \"pygame.KEYDOWN): return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces:\",\n \"pygame, numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\")\",\n \"screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10\",\n \"screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir,\",\n \"= {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40): if font_size not in fonts:\",\n \"= camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(),\",\n \"line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description,\",\n \"pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \\\"No cameras found, exiting!\\\"\",\n \"screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount =\",\n \"camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \\\"No cameras found,\",\n \"pygame.time.wait(10*1000) # 30s def main(): countdown = 10 lastFaceCount = 0 while True:\",\n \"%fs\\\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir,\",\n \"recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen)\",\n \"pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117,\",\n \"Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186 fonts =\",\n \"fonts[font_size] def write(text, colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour)\",\n \"in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock clock = pygame.time.Clock() #\",\n \"= os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading\",\n \"len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces)\",\n \"or len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown ==\",\n \"= recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \\\"images\\\" photo_samples =\",\n \"> 1 else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() -\",\n \"index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \\\"images\\\" photo_samples\",\n \"capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else \\\"images\\\"\",\n \"lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get(): if event.type in\",\n \"photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init()\",\n \"sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0,\",\n \"len(faces) != lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown == 0:\",\n \"\\\"thumbnails\\\", img))) # Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras\",\n \"bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture,\",\n \"numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() -\",\n \"= pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0, screen.get_width(),\",\n \"clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture =\",\n \"= pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16,\",\n \"img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init clock clock = pygame.time.Clock()\",\n \"= 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size]\",\n \"pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40): if font_size not in fonts: fonts[font_size] =\",\n \"# Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \\\"No\",\n \"pygame.quit() print \\\"No cameras found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop\",\n \"screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face\",\n \"in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text, colour = font_colour,\",\n \"colour = font_colour, font_size = 40): return font(font_size).render(str(text), True, colour) # Init AI\",\n \"import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture,\",\n \"recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1 else\",\n \"start)), (0,50)) pygame.display.flip() for img in os.listdir(os.path.join(imgdir, \\\"thumbnails\\\")): photo_samples.append(pygame.image.load(os.path.join(imgdir, \\\"thumbnails\\\", img))) # Init\",\n \"pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in\",\n \"cameras = pygame.camera.list_cameras() if not cameras: pygame.quit() print \\\"No cameras found, exiting!\\\" sys.exit(1)\",\n \"255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img\",\n \"= numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1]\",\n \"capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height())\",\n \"rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture\",\n \"pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line in f: line = line.strip().split(\\\"\\\\t\\\")\",\n \"img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() - start)), (0,50)) pygame.display.flip() for img in\",\n \"start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line in f: line\",\n \"0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip()\",\n \"10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__ == \\\"__main__\\\":\",\n \"1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) >\",\n \"= pygame.camera.list_cameras() if not cameras: pygame.quit() print \\\"No cameras found, exiting!\\\" sys.exit(1) camera\",\n \"40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def\",\n \"recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit()\",\n \"= numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time()\",\n \"\\\"index.tsv\\\")) as f: for line in f: line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir,\",\n \"clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not\",\n \"faces) countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if\",\n \"= 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__ ==\",\n \"event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture = camera.get_image() faces,\",\n \"10 lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get(): if event.type\",\n \"in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main():\",\n \"find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index\",\n \"img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img)\",\n \"Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False) # Init font pygame.font.init()\",\n \"fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size =\",\n \"return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv) > 1\",\n \"= sys.argv[1] if len(sys.argv) > 1 else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index...\",\n \"[] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\"))\",\n \"with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line in f: line = line.strip().split(\\\"\\\\t\\\") img\",\n \"= 40): return font(font_size).render(str(text), True, colour) # Init AI import recognition import sys,\",\n \"in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() - start)), (0,50)) pygame.display.flip()\",\n \"cameras found, exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces):\",\n \"clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras = pygame.camera.list_cameras() if not cameras:\",\n \"camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen = pygame.Rect(0, 0,\",\n \"0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces)\",\n \"capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or\",\n \"len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10 else: screen.blit(write(countdown), (0,0))\",\n \"numpy import pygame.camera # Init display screen = pygame.display.set_mode((0,0), pygame.FULLSCREEN) pygame.display.set_caption(\\\"Magic Mirror\\\") #pygame.mouse.set_visible(False)\",\n \"False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown\",\n \"img))) # Init clock clock = pygame.time.Clock() # Init camera pygame.camera.init() cameras =\",\n \"30s def main(): countdown = 10 lastFaceCount = 0 while True: clock.tick(60) for\",\n \"= 10 lastFaceCount = 0 while True: clock.tick(60) for event in pygame.event.get(): if\",\n \"countdown = 10 else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__\",\n \"def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture\",\n \"exiting!\\\" sys.exit(1) camera = pygame.camera.Camera(cameras[0]) camera.start() # Mainloop def recognize(capture, faces): fullscreen =\",\n \"True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN): return capture\",\n \"lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces)\",\n \"return capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect\",\n \"capture = camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect =\",\n \"!= lastFaceCount: countdown = 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data,\",\n \"import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1)\",\n \"bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True,\",\n \"face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) #\",\n \"if len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount =\",\n \"numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if\",\n \"fullscreen = pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face\",\n \"(0,0)) if len(faces) == 0 or len(faces) != lastFaceCount: countdown = 10 lastFaceCount\",\n \"face = recognition.average(recognition.getRepBBox(capture, face) for face in faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0))\",\n \"bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0),\",\n \"rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) ==\",\n \"else: screen.blit(write(countdown), (0,0)) countdown -= 1 pygame.display.flip() pygame.quit() if __name__ == \\\"__main__\\\": main()\",\n \"font(font_size = 40): if font_size not in fonts: fonts[font_size] = pygame.font.Font(\\\"Futura.ttc\\\", font_size) return\",\n \"0 while True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT, pygame.KEYDOWN):\",\n \"camera.get_image() faces, capture_data = find_faces(capture) for bbox in faces: rect = pygame.Rect(bbox.left(), bbox.top(),\",\n \"Init font pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)}\",\n \"# Init AI import recognition import sys, os def find_faces(pygame_capture): capture = numpy.array(pygame.surfarray.pixels3d(pygame_capture))\",\n \"start = time.time() import pygame, numpy import pygame.camera # Init display screen =\",\n \"in faces: rect = pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect,\",\n \"index... %fs\\\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f:\",\n \"= pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture =\",\n \"= 0 while True: clock.tick(60) for event in pygame.event.get(): if event.type in (pygame.QUIT,\",\n \"pygame.Rect(bbox.left(), bbox.top(), bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture,\",\n \"if len(sys.argv) > 1 else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\" %\",\n \"{40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def font(font_size = 40): if font_size not in fonts: fonts[font_size]\",\n \"as f: for line in f: line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0])\",\n \"pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()),\",\n \"in f: line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description = numpy.array([float(n) for\",\n \"import time start = time.time() import pygame, numpy import pygame.camera # Init display\",\n \"bbox.width(), bbox.height()) pygame.draw.rect(capture, (255, 0, 0), rect, 2) capture = pygame.transform.flip(capture, True, False)\",\n \"= pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0 or len(faces)\",\n \"0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir = sys.argv[1] if len(sys.argv)\",\n \"% (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line\",\n \"photo_samples = [] screen.blit(write(\\\"Loading index... %fs\\\" % (time.time() - start)), (0,0)) pygame.display.flip() with\",\n \"numpy.array(pygame.surfarray.pixels3d(pygame_capture)) capture = numpy.swapaxes(capture, 0, 1) return recognition.align.getAllFaceBoundingBoxes(capture), capture index = recognition.MultiBinaryTree() imgdir\",\n \"for line in f: line = line.strip().split(\\\"\\\\t\\\") img = os.path.join(imgdir, line[0]) description =\",\n \"for n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\" % (time.time() - start)),\",\n \"pygame.Rect(0, 0, screen.get_width(), screen.get_height()) pygame.draw.rect(screen, (255, 255, 255), fullscreen) pygame.display.flip() face = recognition.average(recognition.getRepBBox(capture,\",\n \"faces) img = index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown\",\n \"countdown = 10 lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown\",\n \"#pygame.mouse.set_visible(False) # Init font pygame.font.init() font_colour = 16, 117, 186 fonts = {40:\",\n \"time start = time.time() import pygame, numpy import pygame.camera # Init display screen\",\n \"imgdir = sys.argv[1] if len(sys.argv) > 1 else \\\"images\\\" photo_samples = [] screen.blit(write(\\\"Loading\",\n \"line[0]) description = numpy.array([float(n) for n in line[1:]]) index.insert(description, img) screen.blit(write(\\\"Loading images... %fs\\\"\",\n \"= index.nearest(face) screen.blit(pygame.image.load(img), (0,0)) pygame.display.flip() pygame.time.wait(10*1000) # 30s def main(): countdown = 10\",\n \"lastFaceCount = len(faces) elif countdown == 0: recognize(capture_data, faces) countdown = 10 else:\",\n \"(0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for line in f: line =\",\n \"%fs\\\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \\\"index.tsv\\\")) as f: for\",\n \"2) capture = pygame.transform.flip(capture, True, False) screen.blit(pygame.transform.smoothscale(capture, screen.get_size()), (0,0)) if len(faces) == 0\",\n \"font pygame.font.init() font_colour = 16, 117, 186 fonts = {40: pygame.font.Font(\\\"Futura.ttc\\\", 40)} def\"\n]","isConversational":false}}},{"rowIdx":2692,"cells":{"ngram":{"kind":"list like","value":["pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine in Django. \"\"\"","<reponame>plojyon/resolwe<gh_stars>10-100 \"\"\".. Ignore pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine","from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__,","import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__, __url__, __version__,","engine in Django. \"\"\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__,","in Django. \"\"\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__,","dataflow engine in Django. \"\"\" from resolwe.__about__ import ( # noqa: F401 __author__,","source enterprise dataflow engine in Django. \"\"\" from resolwe.__about__ import ( # noqa:","Resolwe ======= Open source enterprise dataflow engine in Django. \"\"\" from resolwe.__about__ import","\"\"\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__,","enterprise dataflow engine in Django. \"\"\" from resolwe.__about__ import ( # noqa: F401","Open source enterprise dataflow engine in Django. \"\"\" from resolwe.__about__ import ( #","D400. ======= Resolwe ======= Open source enterprise dataflow engine in Django. \"\"\" from","======= Resolwe ======= Open source enterprise dataflow engine in Django. \"\"\" from resolwe.__about__","======= Open source enterprise dataflow engine in Django. \"\"\" from resolwe.__about__ import (","Ignore pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine in Django.","resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__, __url__,","( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__, __url__, __version__, )","\"\"\".. Ignore pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine in","Django. \"\"\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__,"],"string":"[\n \"pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine in Django. \\\"\\\"\\\"\",\n \"<reponame>plojyon/resolwe<gh_stars>10-100 \\\"\\\"\\\".. Ignore pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine\",\n \"from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__,\",\n \"import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__, __url__, __version__,\",\n \"engine in Django. \\\"\\\"\\\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__,\",\n \"in Django. \\\"\\\"\\\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__,\",\n \"dataflow engine in Django. \\\"\\\"\\\" from resolwe.__about__ import ( # noqa: F401 __author__,\",\n \"source enterprise dataflow engine in Django. \\\"\\\"\\\" from resolwe.__about__ import ( # noqa:\",\n \"Resolwe ======= Open source enterprise dataflow engine in Django. \\\"\\\"\\\" from resolwe.__about__ import\",\n \"\\\"\\\"\\\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__,\",\n \"enterprise dataflow engine in Django. \\\"\\\"\\\" from resolwe.__about__ import ( # noqa: F401\",\n \"Open source enterprise dataflow engine in Django. \\\"\\\"\\\" from resolwe.__about__ import ( #\",\n \"D400. ======= Resolwe ======= Open source enterprise dataflow engine in Django. \\\"\\\"\\\" from\",\n \"======= Resolwe ======= Open source enterprise dataflow engine in Django. \\\"\\\"\\\" from resolwe.__about__\",\n \"======= Open source enterprise dataflow engine in Django. \\\"\\\"\\\" from resolwe.__about__ import (\",\n \"Ignore pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine in Django.\",\n \"resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__, __url__,\",\n \"( # noqa: F401 __author__, __copyright__, __email__, __license__, __summary__, __title__, __url__, __version__, )\",\n \"\\\"\\\"\\\".. Ignore pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine in\",\n \"Django. \\\"\\\"\\\" from resolwe.__about__ import ( # noqa: F401 __author__, __copyright__, __email__, __license__,\"\n]","isConversational":false}}},{"rowIdx":2693,"cells":{"ngram":{"kind":"list like","value":["sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac']","x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10))","cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show()","sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10,","fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T,","matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png',","j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9))","ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe","+ 1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax","df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index()","shape:', (som_x, som_y)) # %% # Visualizing distance matrix and activation matrix umatrix","fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column =","sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ##","som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian',","audios by metadata (day, hour, ...) # Each node is colorized according to","1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo'","plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p","first element of SOM shape # Second element will be assigned based on","bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png',","'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) #","= df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour']","as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5,","sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% #","class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16,","as pd import pickle import seaborn as sns import sklearn.preprocessing from python_som import","%% [markdown] # # Imports # %% import matplotlib.pyplot as plt # import","n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% #","y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train)","bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios by","based on the ratio between the # first two principal components of the","sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True)","np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] +","# first two principal components of the train dataset som_x: int = 64","som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG')","np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] =","sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train)","df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df = df[df['grupo'] ==","df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping duplicates # %%","type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping duplicates","bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour'","df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax","sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches:","seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' #","# ## Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count']","in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap","# Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f:","sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] =","= som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1,","Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax =","palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}',","pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) #","assigned based on the ratio between the # first two principal components of","%% [markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True)","- df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% #","librosa as lr # import librosa.display as lrdisp import numpy as np import","len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution","type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp","= pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] =","# type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw()","class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16,","df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of sample dates #","import seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1'","som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix and activation","f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y))","data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as","tail of dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) ==","class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()):","sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax =","# ## Visualizing distribution of features # %% for column in df.iloc[:, 3:-1].columns:","# Filtering dataset by 'group' df = df[df['grupo'] == 1] # %% #","cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count())","balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0]))","class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) ==","for and dropping duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True)","np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] =","%% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x:","%% [markdown] # ## Checking for and dropping duplicates # %% # Resetting","som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix","learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random',","df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing","Checking for and dropping duplicates # %% # Resetting index for duplicate analysis","matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T,","## Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] =","df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala',","[markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index","= df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap","# ## Checking for and dropping duplicates # %% # Resetting index for","np import pandas as pd import pickle import seaborn as sns import sklearn.preprocessing","# Testing python-som with audio dataset # %% [markdown] # # Imports #","as lr # import librosa.display as lrdisp import numpy as np import pandas","hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,","[markdown] # # Testing python-som with audio dataset # %% [markdown] # #","pickle import seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX =","print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for","dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som =","by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x,","## Visualizing distribution of audios by metadata (day, hour, ...) # Each node","transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments","distribution of audios by metadata (day, hour, ...) # Each node is colorized","MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) #","= True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'],","two principal components of the train dataset som_x: int = 64 try: with","plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12)","= 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\",","as np import pandas as pd import pickle import seaborn as sns import","hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\",","bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation matrix separately","df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n')","fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column =","pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True","inplace=True) print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'],","hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig","= pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] =","between the # first two principal components of the train dataset som_x: int","Resetting index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True)","of audios by metadata (day, hour, ...) # Each node is colorized according","## Checking for and dropping duplicates # %% # Resetting index for duplicate","as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x,","'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in","value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine')","= False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\",","fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True)","[markdown] # ## Visualizing distribution of features # %% for column in df.iloc[:,","fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)","transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios by metadata","Each node is colorized according to its most frequent label # %% df['days']","# %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T","class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16,","-1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis()","som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j]","plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %%","pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) #",":, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis()","df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore","# %% # Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig,","'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index,","of dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1])","df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac']","%% [markdown] # ## Visualizing distribution of audios by metadata (day, hour, ...)","# Resetting index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'],","# %% [markdown] # # Testing python-som with audio dataset # %% [markdown]","Visualizing distribution of audios by metadata (day, hour, ...) # Each node is","= class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig =","\"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %%","Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i,","# Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2)","hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,","import librosa as lr # import librosa.display as lrdisp import numpy as np","hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',","ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2,","ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9))","duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime,","%% [markdown] # ## Visualizing distribution of sample dates # %% df_tmp =","== 1] # %% # Dropping tail of dataset for class balancing #","verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type:","df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac']","sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing","try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T","by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala',","ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',","n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del","= df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type:","ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() #","its most frequent label # %% df['days'] = df.index.date df['days'] = (df['days'] -","# %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train),","inplace=True) # %% # Filtering dataset by 'group' df = df[df['grupo'] == 1]","cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing","desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() +","df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days)","= sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()),","# Each node is colorized according to its most frequent label # %%","= som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix and","matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r',","# %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(),","[markdown] # ## Checking for and dropping duplicates # %% # Resetting index","df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\",","= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'),","ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2,","train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som","separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight',","lr # import librosa.display as lrdisp import numpy as np import pandas as","plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':,","= df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'],","print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix and activation matrix","inplace=True) # %% [markdown] # ## Visualizing distribution of sample dates # %%","ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column","plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) #","%% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy()","sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap =","df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore","sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour'","distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del","for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig =","frequent label # %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days']","= hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',","and dropping duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates","distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1,","plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis()","for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int)","for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by","# # Testing python-som with audio dataset # %% [markdown] # # Imports","continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\",","som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x,","robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix","bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of features # %%","# %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False","9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis()","som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j]","1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days'","hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100,","%% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':,","## Visualizing distribution of features # %% for column in df.iloc[:, 3:-1].columns: hmap","try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T","sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %%","df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac']))","= sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax","inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by","n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% #","ratio between the # first two principal components of the train dataset som_x:","# df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of sample dates","= 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError:","%% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column]))","FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset # %% df","for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j]","= False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour']","sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap =","# %% [markdown] # # Imports # %% import matplotlib.pyplot as plt #","by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) #","10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count']","cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb')","except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train)","# %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False","= sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% #","ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] #","label # %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] =","column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i,","= plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig","som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) #","ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column","# %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda","som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance","class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()):","# # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index =","pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True","ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of features","in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap","df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax =","matplotlib.pyplot as plt # import librosa as lr # import librosa.display as lrdisp","transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight',","print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for","sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True)","hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white',","palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df,","10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in","print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for","as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %%","...) # Each node is colorized according to its most frequent label #","3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element of SOM shape","fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show()","som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1","palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name']","ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column","Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap =","SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset # %%","sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac']","robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako',","import pandas as pd import pickle import seaborn as sns import sklearn.preprocessing from","df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0))","element will be assigned based on the ratio between the # first two","== 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of","0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\",","plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments =","robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing","Visualizing distribution of features # %% for column in df.iloc[:, 3:-1].columns: hmap =","= pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True)","= False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown]","# type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\",","activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9))","%% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n')","sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)","df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour","bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days'","dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] =","plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig =","anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True)","SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear',","# Second element will be assigned based on the ratio between the #","most frequent label # %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0])","the ratio between the # first two principal components of the train dataset","import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset #","f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0,","True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\",","cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj',","will be assigned based on the ratio between the # first two principal","len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown]","matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2,","som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training","# Imports # %% import matplotlib.pyplot as plt # import librosa as lr","(som_x, som_y)) # %% # Visualizing distance matrix and activation matrix umatrix =","data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %%","filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(),","[markdown] # ## Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count())","'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j","np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except","shape # Second element will be assigned based on the ratio between the","sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown]","Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac']","transparent=True) # %% [markdown] # ## Visualizing distribution of features # %% for","colorized according to its most frequent label # %% df['days'] = df.index.date df['days']","# %% [markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0,","except Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax","'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y))","3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax =","Imports # %% import matplotlib.pyplot as plt # import librosa as lr #","False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] =","'wb') as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:',","ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing","size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train","p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw()","element of SOM shape # Second element will be assigned based on the","ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance","va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0,","= pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] =","'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True)","= df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw()","= True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10))","features # %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:,","f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %%","False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6,","%% [markdown] # # Testing python-som with audio dataset # %% [markdown] #","figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png',","Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax =","df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'],","df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %%","# %% # Filtering dataset by 'group' df = df[df['grupo'] == 1] #","mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y","ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of","hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"],","hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig","dataset by 'group' df = df[df['grupo'] == 1] # %% # Dropping tail","= pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt,","cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw()","numpy as np import pandas as pd import pickle import seaborn as sns","Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name']","sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt =","Defining first element of SOM shape # Second element will be assigned based","\"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() #","sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # #","(p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using","= class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig =","desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name']","fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis()","i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] =","(ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako',","j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue","df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',","= abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True)","the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f:","distribution of features # %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:,","librosa.display as lrdisp import numpy as np import pandas as pd import pickle","components of the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb')","= sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight',","-1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250,","= -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis')","plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments =","= 'som64_u_grupo1' # %% [markdown] # # Loading dataset # %% df =","plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %%","to its most frequent label # %% df['days'] = df.index.date df['days'] = (df['days']","df = df[df['grupo'] == 1] # %% # Dropping tail of dataset for","ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp =","= 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j","pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True,","%% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig","= sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] #","hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig","= hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3),","%% # Dropping tail of dataset for class balancing # tail_size = abs(","SOM shape # Second element will be assigned based on the ratio between","(df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %%","%% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column]))","ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True)","color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1))","tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index,","'rb') as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1],","= plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True)","hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1,","is colorized according to its most frequent label # %% df['days'] = df.index.date","%% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y))","distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T,","\"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() #","scaler.fit_transform(df_train) # %% # Defining first element of SOM shape # Second element","x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments =","'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10,","analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala,","%% # Defining first element of SOM shape # Second element will be","value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for","of the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as","'som64_u_grupo1' # %% [markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv',","# ## Visualizing distribution of audios by metadata (day, hour, ...) # Each","som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True)","sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center',","Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som,","ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',","dropping duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by","0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of sample","df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime',","of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name']","of SOM shape # Second element will be assigned based on the ratio","= df[df['grupo'] == 1] # %% # Dropping tail of dataset for class","(datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding","ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler =","0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler","import numpy as np import pandas as pd import pickle import seaborn as","hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,","hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i,","for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except","p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler","ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0))","try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T","sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime',","i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception:","# %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x,","fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column =","+ 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's","1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first","# Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax","# Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap","except Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax","9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %%","df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac'","Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax =","= scaler.fit_transform(df_train) # %% # Defining first element of SOM shape # Second","= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() #","= sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ##","pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True","# %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train),","dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) -","ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping duplicates #","= df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution","try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som =","df_train = scaler.fit_transform(df_train) # %% # Defining first element of SOM shape #","'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y))","n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'],","df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp =","df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element of SOM","j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9))","audio dataset # %% [markdown] # # Imports # %% import matplotlib.pyplot as","x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top',","cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing","2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis()","lrdisp import numpy as np import pandas as pd import pickle import seaborn","%% import matplotlib.pyplot as plt # import librosa as lr # import librosa.display","import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] #","plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]})","cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show()","fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True)","plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments =","as lrdisp import numpy as np import pandas as pd import pickle import","%% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax","# Defining first element of SOM shape # Second element will be assigned","%% # Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9))","'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering","plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6,","FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) #","%% [markdown] # ## Visualizing distribution of features # %% for column in","del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp","som_y)) # %% # Visualizing distance matrix and activation matrix umatrix = som.distance_matrix()","Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\",","ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown]","abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) #","neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with","'group' df = df[df['grupo'] == 1] # %% # Dropping tail of dataset","j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1","except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,","= som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try:","9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16,","sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)","metadata (day, hour, ...) # Each node is colorized according to its most","orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\",","som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f)","# len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %%","pandas as pd import pickle import seaborn as sns import sklearn.preprocessing from python_som","sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining","activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis()","robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of","by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x,","for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18)","= class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig =","ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation","# # Imports # %% import matplotlib.pyplot as plt # import librosa as","df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments","column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16,","import matplotlib.pyplot as plt # import librosa as lr # import librosa.display as","pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine',","= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig)","= True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for","sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap =","# %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0))","%% # Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1,","cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution","and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16,","on the ratio between the # first two principal components of the train","python-som with audio dataset # %% [markdown] # # Imports # %% import","of features # %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :,","fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios","plt.draw() # %% # Visualizing distance matrix anc activation matrix separately fig =","first two principal components of the train dataset som_x: int = 64 try:","= 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j","plt # import librosa as lr # import librosa.display as lrdisp import numpy","[0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by","Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) =","9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight',","False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown] #","column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i,","%% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing","True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and","using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train =","by metadata (day, hour, ...) # Each node is colorized according to its","df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala, grupo):\",","according to its most frequent label # %% df['days'] = df.index.date df['days'] =","distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in","j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0","df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% #","# %% [markdown] # ## Visualizing distribution of audios by metadata (day, hour,","Second element will be assigned based on the ratio between the # first","df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac']","# %% # Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16,","# %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:,","scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %%","'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in","duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\",","= som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True,","= sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',","column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i,","# import librosa.display as lrdisp import numpy as np import pandas as pd","open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM","df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True #","# %% import matplotlib.pyplot as plt # import librosa as lr # import","df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element","hour, ...) # Each node is colorized according to its most frequent label","transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation matrix separately fig","1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax =","= hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20,","= pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True,","# %% # Defining first element of SOM shape # Second element will","index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True #","dataset # %% [markdown] # # Imports # %% import matplotlib.pyplot as plt","Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap =","# Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index)","hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False","input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train,","= (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour #","umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r',","[markdown] # # Imports # %% import matplotlib.pyplot as plt # import librosa","open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None,","'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in","fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of features #","som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception:","9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis()","index for duplicate analysis df.reset_index(inplace=True) print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates","df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap =","# tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) #","- len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing","9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown]","index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df = df[df['grupo']","# Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group'","df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] =","transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments","# %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train),","in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() #","with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x,","the # first two principal components of the train dataset som_x: int =","= 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j","= SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM","cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T,","Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df","# import librosa as lr # import librosa.display as lrdisp import numpy as","int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except","[markdown] # ## Visualizing distribution of audios by metadata (day, hour, ...) #","# Dropping tail of dataset for class balancing # tail_size = abs( #","bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo'","verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y =","with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape()","print(\"Duplicates by filename:\", df.duplicated(subset=['file_name']).value_counts(), sep='\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime',","ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc","as plt # import librosa as lr # import librosa.display as lrdisp import","plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %%","df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ##","10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt","(day, hour, ...) # Each node is colorized according to its most frequent","= np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0]","# type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping","# %% [markdown] # ## Checking for and dropping duplicates # %% #","principal components of the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj',","df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'],","df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() #","\"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %%","n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x()","Filtering dataset by 'group' df = df[df['grupo'] == 1] # %% # Dropping","hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)","= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1,","%% # Filtering dataset by 'group' df = df[df['grupo'] == 1] # %%","robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %%","import librosa.display as lrdisp import numpy as np import pandas as pd import","# Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for","dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] =","= plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) #","2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count())","hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks':","# %% [markdown] # ## Visualizing distribution of sample dates # %% df_tmp","plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios by metadata (day,","= som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try:","cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing","%% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':,","j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9))","s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %%","1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ##","type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6,","'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking","fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation matrix","# Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap","del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10,","ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png',","plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train =","# %% [markdown] # ## Visualizing distribution of features # %% for column","2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count())","# using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train","ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% #","1] # %% # Dropping tail of dataset for class balancing # tail_size","# %% # Dropping tail of dataset for class balancing # tail_size =","df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour","neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True)","cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution","node is colorized according to its most frequent label # %% df['days'] =","df[df['grupo'] == 1] # %% # Dropping tail of dataset for class balancing","print(\"Duplicates by (datetime, ala, grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True)","Testing python-som with audio dataset # %% [markdown] # # Imports # %%","x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] =","plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count())","be assigned based on the ratio between the # first two principal components","from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading","SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f)","desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine')","fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0,","[0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by","pd import pickle import seaborn as sns import sklearn.preprocessing from python_som import SOM","= sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',","cmap=sns.color_palette(palette=[\"#000000\", \"blue\", \"orange\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show()","python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset","Dropping tail of dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int)","= sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by","ax.patches: ax.annotate(f'\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %%","in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9))","fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26,","= -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150,","df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h')","== 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] #","# Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap","3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',","64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som","= df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element of","# %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% #","dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df =","with audio dataset # %% [markdown] # # Imports # %% import matplotlib.pyplot","df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10))","Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap =","9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% #","grupo):\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index","'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y))","by 'group' df = df[df['grupo'] == 1] # %% # Dropping tail of","x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train),","transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments","by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x,","250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() #","'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset","import pickle import seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX","True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\"whitegrid\", palette=sns.color_palette(\"muted\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac'])","in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap","%% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column]))"],"string":"[\n \"sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac']\",\n \"x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10))\",\n \"cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show()\",\n \"sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10,\",\n \"fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T,\",\n \"matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png',\",\n \"j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9))\",\n \"ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe\",\n \"+ 1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax\",\n \"df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index()\",\n \"shape:', (som_x, som_y)) # %% # Visualizing distance matrix and activation matrix umatrix\",\n \"fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column =\",\n \"sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ##\",\n \"som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian',\",\n \"audios by metadata (day, hour, ...) # Each node is colorized according to\",\n \"1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo'\",\n \"plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p\",\n \"first element of SOM shape # Second element will be assigned based on\",\n \"bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png',\",\n \"'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) #\",\n \"= df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour']\",\n \"as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5,\",\n \"sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% #\",\n \"class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16,\",\n \"as pd import pickle import seaborn as sns import sklearn.preprocessing from python_som import\",\n \"%% [markdown] # # Imports # %% import matplotlib.pyplot as plt # import\",\n \"n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% #\",\n \"y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train)\",\n \"bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios by\",\n \"based on the ratio between the # first two principal components of the\",\n \"sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True)\",\n \"np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] +\",\n \"# first two principal components of the train dataset som_x: int = 64\",\n \"som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG')\",\n \"np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] =\",\n \"sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train)\",\n \"df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df = df[df['grupo'] ==\",\n \"df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping duplicates # %%\",\n \"type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping duplicates\",\n \"bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour'\",\n \"df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax\",\n \"sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches:\",\n \"seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' #\",\n \"# ## Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count']\",\n \"in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap\",\n \"# Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f:\",\n \"sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] =\",\n \"= som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1,\",\n \"Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax =\",\n \"palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}',\",\n \"pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) #\",\n \"assigned based on the ratio between the # first two principal components of\",\n \"%% [markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True)\",\n \"- df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% #\",\n \"librosa as lr # import librosa.display as lrdisp import numpy as np import\",\n \"len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution\",\n \"type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp\",\n \"= pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] =\",\n \"# type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw()\",\n \"class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16,\",\n \"df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of sample dates #\",\n \"import seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1'\",\n \"som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix and activation\",\n \"f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y))\",\n \"data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as\",\n \"tail of dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) ==\",\n \"class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()):\",\n \"sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax =\",\n \"# ## Visualizing distribution of features # %% for column in df.iloc[:, 3:-1].columns:\",\n \"# Filtering dataset by 'group' df = df[df['grupo'] == 1] # %% #\",\n \"cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count())\",\n \"balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0]))\",\n \"class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) ==\",\n \"for and dropping duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True)\",\n \"np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] =\",\n \"%% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x:\",\n \"%% [markdown] # ## Checking for and dropping duplicates # %% # Resetting\",\n \"som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix\",\n \"learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random',\",\n \"df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing\",\n \"Checking for and dropping duplicates # %% # Resetting index for duplicate analysis\",\n \"matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T,\",\n \"## Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] =\",\n \"df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala',\",\n \"[markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index\",\n \"= df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap\",\n \"# ## Checking for and dropping duplicates # %% # Resetting index for\",\n \"np import pandas as pd import pickle import seaborn as sns import sklearn.preprocessing\",\n \"# Testing python-som with audio dataset # %% [markdown] # # Imports #\",\n \"as lr # import librosa.display as lrdisp import numpy as np import pandas\",\n \"hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,\",\n \"[markdown] # # Testing python-som with audio dataset # %% [markdown] # #\",\n \"pickle import seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX =\",\n \"print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for\",\n \"dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som =\",\n \"by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x,\",\n \"## Visualizing distribution of audios by metadata (day, hour, ...) # Each node\",\n \"transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments\",\n \"distribution of audios by metadata (day, hour, ...) # Each node is colorized\",\n \"MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) #\",\n \"= True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'],\",\n \"two principal components of the train dataset som_x: int = 64 try: with\",\n \"plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12)\",\n \"= 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\",\",\n \"as np import pandas as pd import pickle import seaborn as sns import\",\n \"hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\",\",\n \"bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation matrix separately\",\n \"df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n')\",\n \"fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column =\",\n \"pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True\",\n \"inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'],\",\n \"hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig\",\n \"= pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] =\",\n \"between the # first two principal components of the train dataset som_x: int\",\n \"Resetting index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True)\",\n \"of audios by metadata (day, hour, ...) # Each node is colorized according\",\n \"## Checking for and dropping duplicates # %% # Resetting index for duplicate\",\n \"as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x,\",\n \"'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in\",\n \"value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine')\",\n \"= False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\",\",\n \"fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True)\",\n \"[markdown] # ## Visualizing distribution of features # %% for column in df.iloc[:,\",\n \"fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)\",\n \"transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios by metadata\",\n \"Each node is colorized according to its most frequent label # %% df['days']\",\n \"# %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T\",\n \"class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16,\",\n \"-1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis()\",\n \"som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j]\",\n \"plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %%\",\n \"pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) #\",\n \":, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis()\",\n \"df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore\",\n \"# %% # Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig,\",\n \"'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index,\",\n \"of dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1])\",\n \"df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac']\",\n \"%% [markdown] # ## Visualizing distribution of audios by metadata (day, hour, ...)\",\n \"# Resetting index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'],\",\n \"# %% [markdown] # # Testing python-som with audio dataset # %% [markdown]\",\n \"Visualizing distribution of audios by metadata (day, hour, ...) # Each node is\",\n \"= class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig =\",\n \"\\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %%\",\n \"Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i,\",\n \"# Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2)\",\n \"hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,\",\n \"import librosa as lr # import librosa.display as lrdisp import numpy as np\",\n \"hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',\",\n \"ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2,\",\n \"ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9))\",\n \"duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime,\",\n \"%% [markdown] # ## Visualizing distribution of sample dates # %% df_tmp =\",\n \"== 1] # %% # Dropping tail of dataset for class balancing #\",\n \"verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type:\",\n \"df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac']\",\n \"sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing\",\n \"try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T\",\n \"by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala',\",\n \"ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',\",\n \"n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del\",\n \"= df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type:\",\n \"ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() #\",\n \"its most frequent label # %% df['days'] = df.index.date df['days'] = (df['days'] -\",\n \"# %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train),\",\n \"inplace=True) # %% # Filtering dataset by 'group' df = df[df['grupo'] == 1]\",\n \"cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing\",\n \"desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() +\",\n \"df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days)\",\n \"= sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()),\",\n \"# Each node is colorized according to its most frequent label # %%\",\n \"= som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix and\",\n \"matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r',\",\n \"# %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(),\",\n \"[markdown] # ## Checking for and dropping duplicates # %% # Resetting index\",\n \"df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\",\",\n \"= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'),\",\n \"ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2,\",\n \"train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som\",\n \"separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight',\",\n \"lr # import librosa.display as lrdisp import numpy as np import pandas as\",\n \"plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':,\",\n \"= df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'],\",\n \"print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance matrix and activation matrix\",\n \"inplace=True) # %% [markdown] # ## Visualizing distribution of sample dates # %%\",\n \"ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column\",\n \"plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) #\",\n \"%% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy()\",\n \"sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap =\",\n \"df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore\",\n \"sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour'\",\n \"distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del\",\n \"for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig =\",\n \"frequent label # %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days']\",\n \"= hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',\",\n \"and dropping duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates\",\n \"distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1,\",\n \"plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis()\",\n \"for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int)\",\n \"for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by\",\n \"# # Testing python-som with audio dataset # %% [markdown] # # Imports\",\n \"continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\",\",\n \"som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x,\",\n \"robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix\",\n \"bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of features # %%\",\n \"# %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False\",\n \"9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis()\",\n \"som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j]\",\n \"1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days'\",\n \"hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100,\",\n \"%% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':,\",\n \"## Visualizing distribution of features # %% for column in df.iloc[:, 3:-1].columns: hmap\",\n \"try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T\",\n \"sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %%\",\n \"df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac']))\",\n \"= sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax\",\n \"inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by\",\n \"n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% #\",\n \"ratio between the # first two principal components of the train dataset som_x:\",\n \"# df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of sample dates\",\n \"= 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError:\",\n \"%% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column]))\",\n \"FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset # %% df\",\n \"for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j]\",\n \"= False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour']\",\n \"sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap =\",\n \"# %% [markdown] # # Imports # %% import matplotlib.pyplot as plt #\",\n \"by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) #\",\n \"10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count']\",\n \"cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb')\",\n \"except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train)\",\n \"# %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False\",\n \"= sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% #\",\n \"ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] #\",\n \"label # %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] =\",\n \"column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i,\",\n \"= plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig\",\n \"som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) #\",\n \"ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column\",\n \"# %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda\",\n \"som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing distance\",\n \"class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()):\",\n \"# # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index =\",\n \"pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True\",\n \"ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of features\",\n \"in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap\",\n \"df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax =\",\n \"matplotlib.pyplot as plt # import librosa as lr # import librosa.display as lrdisp\",\n \"transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight',\",\n \"print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for\",\n \"sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True)\",\n \"hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white',\",\n \"palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df,\",\n \"10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in\",\n \"print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for\",\n \"as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %%\",\n \"...) # Each node is colorized according to its most frequent label #\",\n \"3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element of SOM shape\",\n \"fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show()\",\n \"som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1\",\n \"palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name']\",\n \"ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column\",\n \"Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap =\",\n \"SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset # %%\",\n \"sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac']\",\n \"robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako',\",\n \"import pandas as pd import pickle import seaborn as sns import sklearn.preprocessing from\",\n \"df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0))\",\n \"element will be assigned based on the ratio between the # first two\",\n \"== 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of\",\n \"0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\",\",\n \"plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments =\",\n \"robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing\",\n \"Visualizing distribution of features # %% for column in df.iloc[:, 3:-1].columns: hmap =\",\n \"= pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True)\",\n \"= False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown]\",\n \"# type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\",\",\n \"activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9))\",\n \"%% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n')\",\n \"sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)\",\n \"df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour\",\n \"bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days'\",\n \"dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] =\",\n \"plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig =\",\n \"anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True)\",\n \"SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear',\",\n \"# Second element will be assigned based on the ratio between the #\",\n \"most frequent label # %% df['days'] = df.index.date df['days'] = (df['days'] - df['days'][0])\",\n \"the ratio between the # first two principal components of the train dataset\",\n \"import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset #\",\n \"f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0,\",\n \"True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\",\",\n \"cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj',\",\n \"will be assigned based on the ratio between the # first two principal\",\n \"len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown]\",\n \"matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2,\",\n \"som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training\",\n \"# Imports # %% import matplotlib.pyplot as plt # import librosa as lr\",\n \"(som_x, som_y)) # %% # Visualizing distance matrix and activation matrix umatrix =\",\n \"data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %%\",\n \"filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(),\",\n \"[markdown] # ## Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count())\",\n \"'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j\",\n \"np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except\",\n \"shape # Second element will be assigned based on the ratio between the\",\n \"sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown]\",\n \"Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac']\",\n \"transparent=True) # %% [markdown] # ## Visualizing distribution of features # %% for\",\n \"colorized according to its most frequent label # %% df['days'] = df.index.date df['days']\",\n \"# %% [markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0,\",\n \"except Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax\",\n \"'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y))\",\n \"3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax =\",\n \"Imports # %% import matplotlib.pyplot as plt # import librosa as lr #\",\n \"False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] =\",\n \"'wb') as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM shape:',\",\n \"ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing\",\n \"size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train\",\n \"p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw()\",\n \"element of SOM shape # Second element will be assigned based on the\",\n \"ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance\",\n \"va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0,\",\n \"= pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] =\",\n \"'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True)\",\n \"= df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw()\",\n \"= True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10))\",\n \"features # %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:,\",\n \"f) # %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %%\",\n \"False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6,\",\n \"%% [markdown] # # Testing python-som with audio dataset # %% [markdown] #\",\n \"figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png',\",\n \"Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax =\",\n \"df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'],\",\n \"df['rac'] = False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %%\",\n \"# %% # Filtering dataset by 'group' df = df[df['grupo'] == 1] #\",\n \"mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y\",\n \"ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of\",\n \"hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"],\",\n \"hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig\",\n \"dataset by 'group' df = df[df['grupo'] == 1] # %% # Dropping tail\",\n \"= pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt,\",\n \"cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw()\",\n \"numpy as np import pandas as pd import pickle import seaborn as sns\",\n \"Visualizing distribution of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name']\",\n \"sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt =\",\n \"Defining first element of SOM shape # Second element will be assigned based\",\n \"\\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() #\",\n \"sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # #\",\n \"(p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using\",\n \"= class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig =\",\n \"desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name']\",\n \"fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis()\",\n \"i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] =\",\n \"(ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako',\",\n \"j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue\",\n \"df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',\",\n \"= abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True)\",\n \"the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f:\",\n \"distribution of features # %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:,\",\n \"librosa.display as lrdisp import numpy as np import pandas as pd import pickle\",\n \"components of the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb')\",\n \"= sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight',\",\n \"-1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250,\",\n \"= -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis')\",\n \"plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments =\",\n \"= 'som64_u_grupo1' # %% [markdown] # # Loading dataset # %% df =\",\n \"plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %%\",\n \"to its most frequent label # %% df['days'] = df.index.date df['days'] = (df['days']\",\n \"df = df[df['grupo'] == 1] # %% # Dropping tail of dataset for\",\n \"ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp =\",\n \"= 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j\",\n \"pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True,\",\n \"%% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig\",\n \"= sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] #\",\n \"hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig\",\n \"= hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3),\",\n \"%% # Dropping tail of dataset for class balancing # tail_size = abs(\",\n \"SOM shape # Second element will be assigned based on the ratio between\",\n \"(df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %%\",\n \"%% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column]))\",\n \"ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True)\",\n \"color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1))\",\n \"tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index,\",\n \"'rb') as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1],\",\n \"= plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True)\",\n \"hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1,\",\n \"is colorized according to its most frequent label # %% df['days'] = df.index.date\",\n \"%% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y))\",\n \"distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T,\",\n \"\\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() #\",\n \"scaler.fit_transform(df_train) # %% # Defining first element of SOM shape # Second element\",\n \"x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments =\",\n \"'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10,\",\n \"analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala,\",\n \"%% # Defining first element of SOM shape # Second element will be\",\n \"value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for\",\n \"of the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as\",\n \"'som64_u_grupo1' # %% [markdown] # # Loading dataset # %% df = pd.read_csv('features_means.csv',\",\n \"# ## Visualizing distribution of audios by metadata (day, hour, ...) # Each\",\n \"som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True)\",\n \"sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center',\",\n \"Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som,\",\n \"ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png',\",\n \"dropping duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by\",\n \"0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing distribution of sample\",\n \"df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime',\",\n \"of sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name']\",\n \"of SOM shape # Second element will be assigned based on the ratio\",\n \"= df[df['grupo'] == 1] # %% # Dropping tail of dataset for class\",\n \"(datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding\",\n \"ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler =\",\n \"0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler\",\n \"import numpy as np import pandas as pd import pickle import seaborn as\",\n \"hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,\",\n \"hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i,\",\n \"for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except\",\n \"p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's MinMaxScaler scaler\",\n \"ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0))\",\n \"try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T\",\n \"sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime',\",\n \"i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] + 1 except Exception:\",\n \"# %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x,\",\n \"fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column =\",\n \"+ 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% # using sklearn's\",\n \"1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first\",\n \"# Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9)) ax\",\n \"# Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap\",\n \"except Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax\",\n \"9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %%\",\n \"df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac'\",\n \"Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax =\",\n \"= scaler.fit_transform(df_train) # %% # Defining first element of SOM shape # Second\",\n \"= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() #\",\n \"= sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ##\",\n \"pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True\",\n \"# %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train),\",\n \"dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int) == 1]) -\",\n \"ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping duplicates #\",\n \"= df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution\",\n \"try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som =\",\n \"df_train = scaler.fit_transform(df_train) # %% # Defining first element of SOM shape #\",\n \"'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y))\",\n \"n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'],\",\n \"df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp =\",\n \"df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element of SOM\",\n \"j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9))\",\n \"audio dataset # %% [markdown] # # Imports # %% import matplotlib.pyplot as\",\n \"x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top',\",\n \"cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing\",\n \"2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis()\",\n \"lrdisp import numpy as np import pandas as pd import pickle import seaborn\",\n \"%% import matplotlib.pyplot as plt # import librosa as lr # import librosa.display\",\n \"import sklearn.preprocessing from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] #\",\n \"plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]})\",\n \"cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show()\",\n \"fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True)\",\n \"plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments =\",\n \"as lrdisp import numpy as np import pandas as pd import pickle import\",\n \"%% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax\",\n \"# Defining first element of SOM shape # Second element will be assigned\",\n \"%% # Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16, 9))\",\n \"'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering\",\n \"plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6,\",\n \"FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) #\",\n \"%% [markdown] # ## Visualizing distribution of features # %% for column in\",\n \"del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp\",\n \"som_y)) # %% # Visualizing distance matrix and activation matrix umatrix = som.distance_matrix()\",\n \"Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\",\",\n \"ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown]\",\n \"abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) #\",\n \"neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with\",\n \"'group' df = df[df['grupo'] == 1] # %% # Dropping tail of dataset\",\n \"j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1\",\n \"except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap,\",\n \"= som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try:\",\n \"9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16,\",\n \"sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)\",\n \"metadata (day, hour, ...) # Each node is colorized according to its most\",\n \"orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\",\",\n \"som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f)\",\n \"# len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %%\",\n \"pandas as pd import pickle import seaborn as sns import sklearn.preprocessing from python_som\",\n \"sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining\",\n \"activation matrix separately fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(umatrix.T, cmap='bone_r', robust=True) ax.invert_yaxis()\",\n \"robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of\",\n \"by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x,\",\n \"for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18)\",\n \"= class_assignments[(i, j)].most_common()[0][0] + 1 except Exception: continue hmap = hmap.T fig =\",\n \"ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation\",\n \"# # Imports # %% import matplotlib.pyplot as plt # import librosa as\",\n \"df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments\",\n \"column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16,\",\n \"import matplotlib.pyplot as plt # import librosa as lr # import librosa.display as\",\n \"pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine',\",\n \"= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig)\",\n \"= True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for\",\n \"sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = -1 hmap =\",\n \"# %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0))\",\n \"%% # Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1,\",\n \"cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution\",\n \"and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16,\",\n \"on the ratio between the # first two principal components of the train\",\n \"python-som with audio dataset # %% [markdown] # # Imports # %% import\",\n \"of features # %% for column in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :,\",\n \"fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios\",\n \"plt.draw() # %% # Visualizing distance matrix anc activation matrix separately fig =\",\n \"first two principal components of the train dataset som_x: int = 64 try:\",\n \"= 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j\",\n \"plt # import librosa as lr # import librosa.display as lrdisp import numpy\",\n \"[0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by\",\n \"Visualizing distance matrix and activation matrix umatrix = som.distance_matrix() fig, (ax1, ax2) =\",\n \"9)) sns.heatmap(umatrix.T, cmap='bone_r', ax=ax1, robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight',\",\n \"False df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown] #\",\n \"column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i,\",\n \"%% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% # Visualizing\",\n \"True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and\",\n \"using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train =\",\n \"by metadata (day, hour, ...) # Each node is colorized according to its\",\n \"df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\",\",\n \"according to its most frequent label # %% df['days'] = df.index.date df['days'] =\",\n \"distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in\",\n \"j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0\",\n \"df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% #\",\n \"# %% [markdown] # ## Visualizing distribution of audios by metadata (day, hour,\",\n \"Second element will be assigned based on the ratio between the # first\",\n \"df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac']\",\n \"# %% # Visualizing distance matrix anc activation matrix separately fig = plt.figure(figsize=(16,\",\n \"# %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:,\",\n \"scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %%\",\n \"'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in\",\n \"duplicates # %% # Resetting index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\",\",\n \"= som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True,\",\n \"= sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',\",\n \"column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i,\",\n \"# import librosa.display as lrdisp import numpy as np import pandas as pd\",\n \"open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape() print('SOM\",\n \"df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True #\",\n \"# %% import matplotlib.pyplot as plt # import librosa as lr # import\",\n \"df_train = df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element\",\n \"hour, ...) # Each node is colorized according to its most frequent label\",\n \"transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation matrix separately fig\",\n \"1 except Exception: continue hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax =\",\n \"= hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20,\",\n \"= pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True,\",\n \"# %% # Defining first element of SOM shape # Second element will\",\n \"index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':, 'rac'] = True #\",\n \"dataset # %% [markdown] # # Imports # %% import matplotlib.pyplot as plt\",\n \"Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap =\",\n \"# Loading dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index)\",\n \"hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] = False\",\n \"input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train,\",\n \"= (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] = df.index.hour #\",\n \"umatrix = som.distance_matrix() fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 9)) sns.heatmap(umatrix.T, cmap='bone_r',\",\n \"[markdown] # # Imports # %% import matplotlib.pyplot as plt # import librosa\",\n \"open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x, y=None,\",\n \"'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j in\",\n \"fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown] # ## Visualizing distribution of features #\",\n \"som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception:\",\n \"9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis()\",\n \"index for duplicate analysis df.reset_index(inplace=True) print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates\",\n \"df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap =\",\n \"# tail_size = abs( # len(df[df['rac'].astype(int) == 1]) - len(df[df['rac'].astype(int) == 0])) #\",\n \"- len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ## Visualizing\",\n \"9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %% [markdown]\",\n \"index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df = df[df['grupo']\",\n \"# Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group'\",\n \"df.index.date df['days'] = (df['days'] - df['days'][0]) df['days'] = df['days'].apply(lambda x: x.days) df['hour'] =\",\n \"transparent=True) plt.show() # %% # Visualizing by 'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments\",\n \"# %% # Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train),\",\n \"in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() #\",\n \"with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som = SOM(x=som_x,\",\n \"the # first two principal components of the train dataset som_x: int =\",\n \"= 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y)) for i, j\",\n \"= SOM(x=som_x, y=None, input_len=df_train.shape[1], learning_rate=0.5, neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM\",\n \"cmap='bone_r', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix.png', bbox_inches='tight', transparent=True) fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T,\",\n \"Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df\",\n \"# import librosa as lr # import librosa.display as lrdisp import numpy as\",\n \"int = 64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except\",\n \"[markdown] # ## Visualizing distribution of audios by metadata (day, hour, ...) #\",\n \"# Dropping tail of dataset for class balancing # tail_size = abs( #\",\n \"bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo'\",\n \"verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y =\",\n \"with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f) # %% som_x, som_y = som.get_shape()\",\n \"print(\\\"Duplicates by filename:\\\", df.duplicated(subset=['file_name']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['file_name'], inplace=True) print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime',\",\n \"ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc\",\n \"as plt # import librosa as lr # import librosa.display as lrdisp import\",\n \"plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) # %%\",\n \"df.loc['2020-09-22':, 'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ##\",\n \"10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt\",\n \"(day, hour, ...) # Each node is colorized according to its most frequent\",\n \"= np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0]\",\n \"# type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking for and dropping\",\n \"# %% [markdown] # ## Checking for and dropping duplicates # %% #\",\n \"principal components of the train dataset som_x: int = 64 try: with open(f'./{FILE_PREFIX}.obj',\",\n \"df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'],\",\n \"df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() #\",\n \"\\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %%\",\n \"n_colors=6, desat=1.0)) ax = sns.countplot(data=df_melt, x='ractopamine', hue='ractopamine') for p in ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x()\",\n \"Filtering dataset by 'group' df = df[df['grupo'] == 1] # %% # Dropping\",\n \"hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True)\",\n \"= plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1,\",\n \"%% # Filtering dataset by 'group' df = df[df['grupo'] == 1] # %%\",\n \"robust=True) sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', ax=ax2, robust=True) ax1.invert_yaxis() ax2.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %%\",\n \"import librosa.display as lrdisp import numpy as np import pandas as pd import\",\n \"# Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for\",\n \"dataset # %% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] =\",\n \"= plt.figure(figsize=(16, 9)) ax = sns.heatmap(som.activation_matrix(data=df_train).T, cmap='mako', robust=True) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_activation_matrix.png', bbox_inches='tight', transparent=True) #\",\n \"2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count())\",\n \"hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks':\",\n \"# %% [markdown] # ## Visualizing distribution of sample dates # %% df_tmp\",\n \"plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution of audios by metadata (day,\",\n \"= som.label_map(np.array(df_train), np.array(df['rac'])) hmap = np.zeros((som_x, som_y)) for i, j in sorted(class_assignments.keys()): try:\",\n \"cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing\",\n \"%% df = pd.read_csv('features_means.csv', index_col=0, verbose=True) df.index = pd.to_datetime(df.index) df['rac'] = False df.loc['2020-09-22':,\",\n \"j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap = hmap.T fig = plt.figure(figsize=(16, 9))\",\n \"s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %%\",\n \"1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] # ##\",\n \"type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6,\",\n \"'rac'] = True # type: ignore df.sort_index(inplace=True) # %% [markdown] # ## Checking\",\n \"fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_umatrix_activation.png', bbox_inches='tight', transparent=True) plt.draw() # %% # Visualizing distance matrix anc activation matrix\",\n \"# Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap\",\n \"del df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10,\",\n \"ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png',\",\n \"plt.draw() # %% # using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train =\",\n \"# %% [markdown] # ## Visualizing distribution of features # %% for column\",\n \"2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by 'days' print(df.groupby('days')['rac'].count())\",\n \"# using sklearn's MinMaxScaler scaler = sklearn.preprocessing.MinMaxScaler(feature_range=(0, 1)) df_train = df.iloc[:, 3:-1].copy() df_train\",\n \"ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %% #\",\n \"1] # %% # Dropping tail of dataset for class balancing # tail_size\",\n \"# %% # Dropping tail of dataset for class balancing # tail_size =\",\n \"df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore df_tmp = df_tmp.reset_index() df_tmp['hour'] = df_tmp['datetime'].dt.hour\",\n \"neighborhood_radius=1.0, neighborhood_function='gaussian', cyclic_x=True, cyclic_y=True, data=df_train) # Training SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True)\",\n \"cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.close(fig=fig) # %% [markdown] # ## Visualizing distribution\",\n \"node is colorized according to its most frequent label # %% df['days'] =\",\n \"df[df['grupo'] == 1] # %% # Dropping tail of dataset for class balancing\",\n \"print(\\\"Duplicates by (datetime, ala, grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True)\",\n \"Testing python-som with audio dataset # %% [markdown] # # Imports # %%\",\n \"x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac'] =\",\n \"plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac']) plt.draw() df_tmp = pd.DataFrame(df['file_name'].resample('1H').count())\",\n \"be assigned based on the ratio between the # first two principal components\",\n \"from python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading\",\n \"SOM som.weight_initialization(mode='linear', data=df_train) som.train(data=df_train, mode='random', verbose=True) with open(f'./{FILE_PREFIX}.obj', 'wb') as f: pickle.dump(som, f)\",\n \"desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine')\",\n \"fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0,\",\n \"[0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by\",\n \"pd import pickle import seaborn as sns import sklearn.preprocessing from python_som import SOM\",\n \"= sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',\",\n \"cmap=sns.color_palette(palette=[\\\"#000000\\\", \\\"blue\\\", \\\"orange\\\"], n_colors=3), cbar_kws={'ticks': [0, 1, 2]}) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_rac.png', bbox_inches='tight', transparent=True) plt.show()\",\n \"python_som import SOM FILE_PREFIX = 'som64_u_grupo1' # %% [markdown] # # Loading dataset\",\n \"Dropping tail of dataset for class balancing # tail_size = abs( # len(df[df['rac'].astype(int)\",\n \"= sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% # Visualizing by\",\n \"ax.patches: ax.annotate(f'\\\\n{p.get_height()}', (p.get_x() + 0.2, p.get_height()), ha='center', va='top', color='white', size=18) plt.draw() # %%\",\n \"in df.iloc[:, 3:-1].columns: hmap = som.get_weights()[:, :, df.iloc[:, 3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9))\",\n \"fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150, 250, s=100, l=20, sep=1, n=26,\",\n \"= -1 hmap = hmap.T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, cmap=sns.diverging_palette(150,\",\n \"df_tmp['hour'] = df_tmp['datetime'].dt.hour plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp['hour'], x=df_tmp['count'], hue=df_tmp['rac'], orient='h')\",\n \"== 1]) - len(df[df['rac'].astype(int) == 0])) # df.drop(df.tail(tail_size).index, inplace=True) # %% [markdown] #\",\n \"# Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap\",\n \"3:-1].columns.get_loc(column)].T fig = plt.figure(figsize=(16, 9)) ax = sns.heatmap(hmap, robust=True, cmap='BrBG') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight',\",\n \"64 try: with open(f'./{FILE_PREFIX}.obj', 'rb') as f: som = pickle.load(f) except FileNotFoundError: som\",\n \"= df.iloc[:, 3:-1].copy() df_train = scaler.fit_transform(df_train) # %% # Defining first element of\",\n \"# %% som_x, som_y = som.get_shape() print('SOM shape:', (som_x, som_y)) # %% #\",\n \"dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset by 'group' df =\",\n \"with audio dataset # %% [markdown] # # Imports # %% import matplotlib.pyplot\",\n \"df_tmp['file_name'] df_tmp['rac'] = False df_tmp.loc['2020-09-22':, 'rac'] = True # type: ignore plt.figure(figsize=(10, 10))\",\n \"Visualizing by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap =\",\n \"9)) ax = sns.heatmap(hmap, cmap='viridis') ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() # %% #\",\n \"grupo):\\\", df.duplicated(subset=['datetime', 'ala', 'grupo']).value_counts(), sep='\\\\n') df.drop_duplicates(subset=['datetime', 'ala', 'grupo'], inplace=True) # Rebuilding dataframe index\",\n \"'hour' print(df.groupby('hour')['rac'].count()) column = 'hour' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x, som_y))\",\n \"by 'group' df = df[df['grupo'] == 1] # %% # Dropping tail of\",\n \"x.days) df['hour'] = df.index.hour # %% # Visualizing 'rac' distribution class_assignments = som.label_map(np.array(df_train),\",\n \"transparent=True) plt.show() # %% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments\",\n \"by 'days' print(df.groupby('days')['rac'].count()) column = 'days' class_assignments = som.label_map(np.array(df_train), np.array(df[column])) hmap = np.zeros((som_x,\",\n \"250, s=100, l=20, sep=1, n=26, center='light'), center=12) ax.invert_yaxis() fig.savefig(f'./output_{FILE_PREFIX}/{FILE_PREFIX}_{column}.png', bbox_inches='tight', transparent=True) plt.show() #\",\n \"'grupo'], inplace=True) # Rebuilding dataframe index df.set_index('datetime', inplace=True) # %% # Filtering dataset\",\n \"import pickle import seaborn as sns import sklearn.preprocessing from python_som import SOM FILE_PREFIX\",\n \"True # type: ignore plt.figure(figsize=(10, 10)) sns.set(style=\\\"whitegrid\\\", palette=sns.color_palette(\\\"muted\\\", n_colors=6, desat=1.0)) sns.barplot(y=df_tmp.index, x=df_tmp['count'], hue=df_tmp['rac'])\",\n \"in sorted(class_assignments.keys()): try: hmap[i][j] = class_assignments[(i, j)].most_common()[0][0] except Exception: hmap[i][j] = 0 hmap\",\n \"%% # Visualizing by 'grupo' print(df.groupby('grupo')['rac'].count()) column = 'grupo' class_assignments = som.label_map(np.array(df_train), np.array(df[column]))\"\n]","isConversational":false}}},{"rowIdx":2694,"cells":{"ngram":{"kind":"list like","value":["code. \"\"\" # return Competition return Competition def update_competition(self, competition): \"\"\" Retrieve data","the most recent fixtures and results for the supplied competition code. \"\"\" return","results. Returns ------- Competition A Competition object with the most recent fixtures and","the supplied competition code. Returns ------- Competition A Competition object with the most","competition code. \"\"\" # return Competition return Competition def update_competition(self, competition): \"\"\" Retrieve","and results for the supplied competition code. \"\"\" # return Competition return Competition","the data model with the most resent fixtures and results.\"\"\" from footy.domain import","Competition A Competition object with the most recent fixtures and results for the","\"\"\" # return Competition return Competition def update_competition(self, competition): \"\"\" Retrieve data and","- Update the data model with the most resent fixtures and results.\"\"\" from","class UpdateEngine: \"\"\"Prediction Engine - Update the data model with the most resent","and results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\" Retrieve","results.\"\"\" from footy.domain import Competition class UpdateEngine: \"\"\"Prediction Engine - Update the data","model with the most resent fixtures and results.\"\"\" from footy.domain import Competition class","fixtures and results for the supplied competition code. \"\"\" # return Competition return","A Competition object with the most recent fixtures and results for the supplied","Returns ------- Competition A Competition object with the most recent fixtures and results","Update the data model with the most resent fixtures and results.\"\"\" from footy.domain","with the most resent fixtures and results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\"","Competition return Competition def update_competition(self, competition): \"\"\" Retrieve data and enrich the supplied","most resent fixtures and results.\"\"\" from footy.domain import Competition class UpdateEngine: \"\"\"Prediction Engine","model with the most resent fixtures and results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine","the most resent fixtures and results.\"\"\" from footy.domain import Competition class UpdateEngine: \"\"\"Prediction","import Competition class UpdateEngine: \"\"\"Prediction Engine - Update the data model with the","\"\"\" Retrieve data for the supplied competition code. Returns ------- Competition A Competition","most recent fixtures and results for the supplied competition code. \"\"\" # return","Competition class UpdateEngine: \"\"\"Prediction Engine - Update the data model with the most","data model with the most resent fixtures and results.\"\"\" from footy.domain import Competition","__init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\" Retrieve data for the","most recent fixtures and results for the supplied competition code. \"\"\" return Competition","return Competition def update_competition(self, competition): \"\"\" Retrieve data and enrich the supplied competition","results for the supplied competition code. \"\"\" # return Competition return Competition def","Retrieve data for the supplied competition code. Returns ------- Competition A Competition object","\"\"\"Prediction Engine - Update the data model with the most resent fixtures and","the most recent fixtures and results for the supplied competition code. \"\"\" #","def get_competition(self, code): \"\"\" Retrieve data for the supplied competition code. Returns -------","return Competition return Competition def update_competition(self, competition): \"\"\" Retrieve data and enrich the","# return Competition return Competition def update_competition(self, competition): \"\"\" Retrieve data and enrich","with the most recent fixtures and results for the supplied competition code. \"\"\"","\"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\" Retrieve data for the supplied","for the supplied competition code. Returns ------- Competition A Competition object with the","recent fixtures and results. Returns ------- Competition A Competition object with the most","supplied competition with the most recent fixtures and results. Returns ------- Competition A","fixtures and results.\"\"\" from footy.domain import Competition class UpdateEngine: \"\"\"Prediction Engine - Update","the supplied competition code. \"\"\" # return Competition return Competition def update_competition(self, competition):","fixtures and results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\"","competition code. Returns ------- Competition A Competition object with the most recent fixtures","for the supplied competition code. \"\"\" # return Competition return Competition def update_competition(self,","with the most recent fixtures and results. Returns ------- Competition A Competition object","code. Returns ------- Competition A Competition object with the most recent fixtures and","object with the most recent fixtures and results for the supplied competition code.","code): \"\"\" Retrieve data for the supplied competition code. Returns ------- Competition A","resent fixtures and results.\"\"\" from footy.domain import Competition class UpdateEngine: \"\"\"Prediction Engine -","def update_competition(self, competition): \"\"\" Retrieve data and enrich the supplied competition with the","resent fixtures and results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self, code):","update_competition(self, competition): \"\"\" Retrieve data and enrich the supplied competition with the most","enrich the supplied competition with the most recent fixtures and results. Returns -------","- Update the data model with the most resent fixtures and results.\"\"\" def","object.\"\"\" def get_competition(self, code): \"\"\" Retrieve data for the supplied competition code. Returns","the most recent fixtures and results. Returns ------- Competition A Competition object with","most recent fixtures and results. Returns ------- Competition A Competition object with the","fixtures and results. Returns ------- Competition A Competition object with the most recent","with the most resent fixtures and results.\"\"\" from footy.domain import Competition class UpdateEngine:","data model with the most resent fixtures and results.\"\"\" def __init__(self): \"\"\"Construct a","supplied competition code. \"\"\" # return Competition return Competition def update_competition(self, competition): \"\"\"","the supplied competition with the most recent fixtures and results. Returns ------- Competition","Update the data model with the most resent fixtures and results.\"\"\" def __init__(self):","a UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\" Retrieve data for the supplied competition","data for the supplied competition code. Returns ------- Competition A Competition object with","and enrich the supplied competition with the most recent fixtures and results. Returns","------- Competition A Competition object with the most recent fixtures and results for","footy.domain import Competition class UpdateEngine: \"\"\"Prediction Engine - Update the data model with","and results.\"\"\" from footy.domain import Competition class UpdateEngine: \"\"\"Prediction Engine - Update the","Competition object with the most recent fixtures and results for the supplied competition","\"\"\" Retrieve data and enrich the supplied competition with the most recent fixtures","recent fixtures and results for the supplied competition code. \"\"\" # return Competition","data and enrich the supplied competition with the most recent fixtures and results.","supplied competition code. Returns ------- Competition A Competition object with the most recent","def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\" Retrieve data for","competition with the most recent fixtures and results. Returns ------- Competition A Competition","UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\" Retrieve data for the supplied competition code.","Retrieve data and enrich the supplied competition with the most recent fixtures and","and results. Returns ------- Competition A Competition object with the most recent fixtures","most resent fixtures and results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self,","competition): \"\"\" Retrieve data and enrich the supplied competition with the most recent","from footy.domain import Competition class UpdateEngine: \"\"\"Prediction Engine - Update the data model","Engine - Update the data model with the most resent fixtures and results.\"\"\"","UpdateEngine: \"\"\"Prediction Engine - Update the data model with the most resent fixtures","get_competition(self, code): \"\"\" Retrieve data for the supplied competition code. Returns ------- Competition","the data model with the most resent fixtures and results.\"\"\" def __init__(self): \"\"\"Construct","results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def get_competition(self, code): \"\"\" Retrieve data","the most resent fixtures and results.\"\"\" def __init__(self): \"\"\"Construct a UpdateEngine object.\"\"\" def","Competition def update_competition(self, competition): \"\"\" Retrieve data and enrich the supplied competition with"],"string":"[\n \"code. \\\"\\\"\\\" # return Competition return Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve data\",\n \"the most recent fixtures and results for the supplied competition code. \\\"\\\"\\\" return\",\n \"results. Returns ------- Competition A Competition object with the most recent fixtures and\",\n \"the supplied competition code. Returns ------- Competition A Competition object with the most\",\n \"competition code. \\\"\\\"\\\" # return Competition return Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve\",\n \"and results for the supplied competition code. \\\"\\\"\\\" # return Competition return Competition\",\n \"the data model with the most resent fixtures and results.\\\"\\\"\\\" from footy.domain import\",\n \"Competition A Competition object with the most recent fixtures and results for the\",\n \"\\\"\\\"\\\" # return Competition return Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve data and\",\n \"- Update the data model with the most resent fixtures and results.\\\"\\\"\\\" from\",\n \"class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the data model with the most resent\",\n \"and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve\",\n \"results.\\\"\\\"\\\" from footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the data\",\n \"model with the most resent fixtures and results.\\\"\\\"\\\" from footy.domain import Competition class\",\n \"fixtures and results for the supplied competition code. \\\"\\\"\\\" # return Competition return\",\n \"A Competition object with the most recent fixtures and results for the supplied\",\n \"Returns ------- Competition A Competition object with the most recent fixtures and results\",\n \"Update the data model with the most resent fixtures and results.\\\"\\\"\\\" from footy.domain\",\n \"with the most resent fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\"\",\n \"Competition return Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve data and enrich the supplied\",\n \"most resent fixtures and results.\\\"\\\"\\\" from footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine\",\n \"model with the most resent fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine\",\n \"the most resent fixtures and results.\\\"\\\"\\\" from footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction\",\n \"import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the data model with the\",\n \"\\\"\\\"\\\" Retrieve data for the supplied competition code. Returns ------- Competition A Competition\",\n \"most recent fixtures and results for the supplied competition code. \\\"\\\"\\\" # return\",\n \"Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the data model with the most\",\n \"data model with the most resent fixtures and results.\\\"\\\"\\\" from footy.domain import Competition\",\n \"__init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve data for the\",\n \"most recent fixtures and results for the supplied competition code. \\\"\\\"\\\" return Competition\",\n \"return Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve data and enrich the supplied competition\",\n \"results for the supplied competition code. \\\"\\\"\\\" # return Competition return Competition def\",\n \"Retrieve data for the supplied competition code. Returns ------- Competition A Competition object\",\n \"\\\"\\\"\\\"Prediction Engine - Update the data model with the most resent fixtures and\",\n \"the most recent fixtures and results for the supplied competition code. \\\"\\\"\\\" #\",\n \"def get_competition(self, code): \\\"\\\"\\\" Retrieve data for the supplied competition code. Returns -------\",\n \"return Competition return Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve data and enrich the\",\n \"# return Competition return Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve data and enrich\",\n \"with the most recent fixtures and results for the supplied competition code. \\\"\\\"\\\"\",\n \"\\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve data for the supplied\",\n \"for the supplied competition code. Returns ------- Competition A Competition object with the\",\n \"recent fixtures and results. Returns ------- Competition A Competition object with the most\",\n \"supplied competition with the most recent fixtures and results. Returns ------- Competition A\",\n \"fixtures and results.\\\"\\\"\\\" from footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update\",\n \"the supplied competition code. \\\"\\\"\\\" # return Competition return Competition def update_competition(self, competition):\",\n \"fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\"\",\n \"competition code. Returns ------- Competition A Competition object with the most recent fixtures\",\n \"for the supplied competition code. \\\"\\\"\\\" # return Competition return Competition def update_competition(self,\",\n \"with the most recent fixtures and results. Returns ------- Competition A Competition object\",\n \"code. Returns ------- Competition A Competition object with the most recent fixtures and\",\n \"object with the most recent fixtures and results for the supplied competition code.\",\n \"code): \\\"\\\"\\\" Retrieve data for the supplied competition code. Returns ------- Competition A\",\n \"resent fixtures and results.\\\"\\\"\\\" from footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine -\",\n \"def update_competition(self, competition): \\\"\\\"\\\" Retrieve data and enrich the supplied competition with the\",\n \"resent fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code):\",\n \"update_competition(self, competition): \\\"\\\"\\\" Retrieve data and enrich the supplied competition with the most\",\n \"enrich the supplied competition with the most recent fixtures and results. Returns -------\",\n \"- Update the data model with the most resent fixtures and results.\\\"\\\"\\\" def\",\n \"object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve data for the supplied competition code. Returns\",\n \"the most recent fixtures and results. Returns ------- Competition A Competition object with\",\n \"most recent fixtures and results. Returns ------- Competition A Competition object with the\",\n \"fixtures and results. Returns ------- Competition A Competition object with the most recent\",\n \"with the most resent fixtures and results.\\\"\\\"\\\" from footy.domain import Competition class UpdateEngine:\",\n \"data model with the most resent fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a\",\n \"supplied competition code. \\\"\\\"\\\" # return Competition return Competition def update_competition(self, competition): \\\"\\\"\\\"\",\n \"the supplied competition with the most recent fixtures and results. Returns ------- Competition\",\n \"Update the data model with the most resent fixtures and results.\\\"\\\"\\\" def __init__(self):\",\n \"a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve data for the supplied competition\",\n \"data for the supplied competition code. Returns ------- Competition A Competition object with\",\n \"and enrich the supplied competition with the most recent fixtures and results. Returns\",\n \"------- Competition A Competition object with the most recent fixtures and results for\",\n \"footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the data model with\",\n \"and results.\\\"\\\"\\\" from footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the\",\n \"Competition object with the most recent fixtures and results for the supplied competition\",\n \"\\\"\\\"\\\" Retrieve data and enrich the supplied competition with the most recent fixtures\",\n \"recent fixtures and results for the supplied competition code. \\\"\\\"\\\" # return Competition\",\n \"data and enrich the supplied competition with the most recent fixtures and results.\",\n \"supplied competition code. Returns ------- Competition A Competition object with the most recent\",\n \"def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve data for\",\n \"competition with the most recent fixtures and results. Returns ------- Competition A Competition\",\n \"UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve data for the supplied competition code.\",\n \"Retrieve data and enrich the supplied competition with the most recent fixtures and\",\n \"and results. Returns ------- Competition A Competition object with the most recent fixtures\",\n \"most resent fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self,\",\n \"competition): \\\"\\\"\\\" Retrieve data and enrich the supplied competition with the most recent\",\n \"from footy.domain import Competition class UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the data model\",\n \"Engine - Update the data model with the most resent fixtures and results.\\\"\\\"\\\"\",\n \"UpdateEngine: \\\"\\\"\\\"Prediction Engine - Update the data model with the most resent fixtures\",\n \"get_competition(self, code): \\\"\\\"\\\" Retrieve data for the supplied competition code. Returns ------- Competition\",\n \"the data model with the most resent fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct\",\n \"results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def get_competition(self, code): \\\"\\\"\\\" Retrieve data\",\n \"the most resent fixtures and results.\\\"\\\"\\\" def __init__(self): \\\"\\\"\\\"Construct a UpdateEngine object.\\\"\\\"\\\" def\",\n \"Competition def update_competition(self, competition): \\\"\\\"\\\" Retrieve data and enrich the supplied competition with\"\n]","isConversational":false}}},{"rowIdx":2695,"cells":{"ngram":{"kind":"list like","value":["Exception(\"No files found to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files)","Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No","you see once you click into a grid cell on the pydeck gallery","DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io, set this variable from","new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck example and","Admin panel raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name =","you click into a grid cell on the pydeck gallery page \"\"\" from","if \"layer\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're","a deck.gl docs link if we're not referencing a layer # Obviously very","not os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io, set this variable from the Admin","the Admin panel raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name","see once you click into a grid cell on the pydeck gallery page","\"layer\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're not","= asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if \"layer\" in deckgl_docs_layer_name: # Don't add","rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f for f","not set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url =","examples into .rst pages with code These populate the files you see once","layer # Obviously very rough, should change this eventually to handle views etc","files found to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if","If running for rtfd.io, set this variable from the Admin panel raise Exception(\"MAPBOX_API_KEY","should change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name","asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if \"layer\" in deckgl_docs_layer_name: # Don't add a","import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If running","= to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if \"layer\" in deckgl_docs_layer_name:","# Don't add a deck.gl docs link if we're not referencing a layer",") rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\") f = open(rst_path, \"w+\") print(\"* Converted","# If running for rtfd.io, set this variable from the Admin panel raise","Obviously very rough, should change this eventually to handle views etc deckgl_doc_url =","= os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck example and move the .html output","create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if \"layer\"","\"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code","import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import","deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're not referencing a","once you click into a grid cell on the pydeck gallery page \"\"\"","add a deck.gl docs link if we're not referencing a layer # Obviously","running for rtfd.io, set this variable from the Admin panel raise Exception(\"MAPBOX_API_KEY not","import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates","\"\"\" from multiprocessing import Pool import os import subprocess import sys from const","# Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck","page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\")","embed pydeck examples into .rst pages with code These populate the files you","html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name,","etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname =","f = open(rst_path, \"w+\") print(\"* Converted %s to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source)","panel raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\",","in deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're not referencing","gallery page \"\"\" from multiprocessing import Pool import os import subprocess import sys","raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\")","= open(rst_path, \"w+\") print(\"* Converted %s to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close()","import os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR,","Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url","f.close() def main(): pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB]","deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\",","open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path","subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__ == \"__main__\": main()","not candidate_files: raise Exception(\"No files found to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR,","set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url = None","), shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code,","a layer # Obviously very rough, should change this eventually to handle views","{fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code =","import Pool import os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB,","pages with code These populate the files you see once you click into","Don't add a deck.gl docs link if we're not referencing a layer #","{html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read() doc_source","= [f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No files found","HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"):","this variable from the Admin panel raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name","(pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f for","we're not referencing a layer # Obviously very rough, should change this eventually","These populate the files you see once you click into a grid cell","change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name #","code These populate the files you see once you click into a grid","+ deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run","fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render(","const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from","examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck example and move the",".rst pages with code These populate the files you see once you click","raise Exception(\"No files found to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst,","not referencing a layer # Obviously very rough, should change this eventually to","\"-\") deckgl_doc_url = None if \"layer\" in deckgl_docs_layer_name: # Don't add a deck.gl","deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the","from multiprocessing import Pool import os import subprocess import sys from const import","rtfd.io, set this variable from the Admin panel raise Exception(\"MAPBOX_API_KEY not set\") def","variable from the Admin panel raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name =","%s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files =","Run the pydeck example and move the .html output subprocess.call( \"{python} {fname}; mv","into a grid cell on the pydeck gallery page \"\"\" from multiprocessing import","os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH","# Obviously very rough, should change this eventually to handle views etc deckgl_doc_url","\".html\") # Run the pydeck example and move the .html output subprocess.call( \"{python}","{html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read()","snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\") f","candidate_files: raise Exception(\"No files found to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True)","pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files:","asset_name + \".rst\") f = open(rst_path, \"w+\") print(\"* Converted %s to %s\" %","Converted %s to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool =","populate the files you see once you click into a grid cell on","if not candidate_files: raise Exception(\"No files found to convert\") subprocess.call(\"mkdir -p %s\" %","the pydeck example and move the .html output subprocess.call( \"{python} {fname}; mv {html_src}","from templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io, set","utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If","os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io, set this variable from the Admin panel","to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html","= os.path.join(GALLERY_DIR, asset_name + \".rst\") f = open(rst_path, \"w+\") print(\"* Converted %s to","the pydeck gallery page \"\"\" from multiprocessing import Pool import os import subprocess","pydeck example and move the .html output subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format(","convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__ == \"__main__\":","= None if \"layer\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link","move the .html output subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname,","import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from","GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if","this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create","= DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\")","% (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f","rough, should change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE +","Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck example","def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if","%s to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4)","handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples","set this variable from the Admin panel raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name):","os.path.join(GALLERY_DIR, asset_name + \".rst\") f = open(rst_path, \"w+\") print(\"* Converted %s to %s\"","multiprocessing import Pool import os import subprocess import sys from const import DECKGL_URL_BASE,","pydeck gallery page \"\"\" from multiprocessing import Pool import os import subprocess import","f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f for f in","sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name,","mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name,","into .rst pages with code These populate the files you see once you","from the Admin panel raise Exception(\"MAPBOX_API_KEY not set\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name)","templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io, set this","files you see once you click into a grid cell on the pydeck","deckgl_doc_url = None if \"layer\" in deckgl_docs_layer_name: # Don't add a deck.gl docs","= Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files: raise","print(\"* Converted %s to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool","to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if \"layer\" in deckgl_docs_layer_name: #","os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck example and move the .html output subprocess.call(","DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name +","subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils","f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No files found to convert\") subprocess.call(\"mkdir","in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No files found to convert\") subprocess.call(\"mkdir -p","to embed pydeck examples into .rst pages with code These populate the files",".html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck example and move","and move the .html output subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name,","html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\") f = open(rst_path, \"w+\")","to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If running for","None if \"layer\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link if","doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR,","eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new","EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE","from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string","subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, )","[f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No files found to","page \"\"\" from multiprocessing import Pool import os import subprocess import sys from","\"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path =","EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No files found to convert\") subprocess.call(\"mkdir -p %s\"","python_code = open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url,","html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") # Run the pydeck example and move the .html","main(): pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not","open(rst_path, \"w+\") print(\"* Converted %s to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def","referencing a layer # Obviously very rough, should change this eventually to handle","HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not","pydeck examples into .rst pages with code These populate the files you see","deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if \"layer\" in deckgl_docs_layer_name: # Don't","python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\") f =",".html output subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ),","DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\".py\", \".html\") #","for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No files found to convert\")","the .html output subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR","\"\"\"Script to embed pydeck examples into .rst pages with code These populate the","import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io, set this variable","deck.gl docs link if we're not referencing a layer # Obviously very rough,","= open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, )","docs link if we're not referencing a layer # Obviously very rough, should","for rtfd.io, set this variable from the Admin panel raise Exception(\"MAPBOX_API_KEY not set\")","from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): #","found to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__","to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io,","candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No files","\"w+\") print(\"* Converted %s to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main():","a grid cell on the pydeck gallery page \"\"\" from multiprocessing import Pool",") python_code = open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname),","hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\") f = open(rst_path,","to %s\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files","grid cell on the pydeck gallery page \"\"\" from multiprocessing import Pool import","cell on the pydeck gallery page \"\"\" from multiprocessing import Pool import os","# Run the pydeck example and move the .html output subprocess.call( \"{python} {fname};","deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\") f = open(rst_path, \"w+\") print(\"*","python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read() doc_source =","rst_path = os.path.join(GALLERY_DIR, asset_name + \".rst\") f = open(rst_path, \"w+\") print(\"* Converted %s","\".rst\") f = open(rst_path, \"w+\") print(\"* Converted %s to %s\" % (pydeck_example_file_name, rst_path))","shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH,","html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \"r\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name),","def main(): pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if","asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\"_\", \"-\") deckgl_doc_url = None if \"layer\" in","click into a grid cell on the pydeck gallery page \"\"\" from multiprocessing","link if we're not referencing a layer # Obviously very rough, should change","Pool import os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR,","views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname","with code These populate the files you see once you click into a","on the pydeck gallery page \"\"\" from multiprocessing import Pool import os import","DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import","if not os.environ.get(\"MAPBOX_API_KEY\"): # If running for rtfd.io, set this variable from the","the files you see once you click into a grid cell on the","output subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True,","= DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name","very rough, should change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE","+ \".rst\") f = open(rst_path, \"w+\") print(\"* Converted %s to %s\" % (pydeck_example_file_name,","to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__ ==","example and move the .html output subprocess.call( \"{python} {fname}; mv {html_src} {html_dest}\".format( python=sys.executable,","if we're not referencing a layer # Obviously very rough, should change this"],"string":"[\n \"Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files)\",\n \"Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No\",\n \"you see once you click into a grid cell on the pydeck gallery\",\n \"DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io, set this variable from\",\n \"new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck example and\",\n \"Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name =\",\n \"you click into a grid cell on the pydeck gallery page \\\"\\\"\\\" from\",\n \"if \\\"layer\\\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're\",\n \"a deck.gl docs link if we're not referencing a layer # Obviously very\",\n \"not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io, set this variable from the Admin\",\n \"the Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name\",\n \"see once you click into a grid cell on the pydeck gallery page\",\n \"\\\"layer\\\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're not\",\n \"= asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if \\\"layer\\\" in deckgl_docs_layer_name: # Don't add\",\n \"rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f for f\",\n \"not set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url =\",\n \"examples into .rst pages with code These populate the files you see once\",\n \"layer # Obviously very rough, should change this eventually to handle views etc\",\n \"files found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if\",\n \"If running for rtfd.io, set this variable from the Admin panel raise Exception(\\\"MAPBOX_API_KEY\",\n \"should change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name\",\n \"asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if \\\"layer\\\" in deckgl_docs_layer_name: # Don't add a\",\n \"import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running\",\n \"= to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if \\\"layer\\\" in deckgl_docs_layer_name:\",\n \"# Don't add a deck.gl docs link if we're not referencing a layer\",\n \") rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"* Converted\",\n \"# If running for rtfd.io, set this variable from the Admin panel raise\",\n \"Obviously very rough, should change this eventually to handle views etc deckgl_doc_url =\",\n \"= os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck example and move the .html output\",\n \"create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if \\\"layer\\\"\",\n \"\\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code\",\n \"import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import\",\n \"deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're not referencing a\",\n \"once you click into a grid cell on the pydeck gallery page \\\"\\\"\\\"\",\n \"add a deck.gl docs link if we're not referencing a layer # Obviously\",\n \"running for rtfd.io, set this variable from the Admin panel raise Exception(\\\"MAPBOX_API_KEY not\",\n \"import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates\",\n \"\\\"\\\"\\\" from multiprocessing import Pool import os import subprocess import sys from const\",\n \"# Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck\",\n \"page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\")\",\n \"embed pydeck examples into .rst pages with code These populate the files you\",\n \"html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name,\",\n \"etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname =\",\n \"f = open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source)\",\n \"panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\",\",\n \"in deckgl_docs_layer_name: # Don't add a deck.gl docs link if we're not referencing\",\n \"gallery page \\\"\\\"\\\" from multiprocessing import Pool import os import subprocess import sys\",\n \"raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\")\",\n \"= open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close()\",\n \"import os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR,\",\n \"Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url\",\n \"f.close() def main(): pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB]\",\n \"deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\",\",\n \"open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path\",\n \"subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__ == \\\"__main__\\\": main()\",\n \"not candidate_files: raise Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR,\",\n \"set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None\",\n \"), shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code,\",\n \"a layer # Obviously very rough, should change this eventually to handle views\",\n \"{fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code =\",\n \"import Pool import os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB,\",\n \"pages with code These populate the files you see once you click into\",\n \"Don't add a deck.gl docs link if we're not referencing a layer #\",\n \"{html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source\",\n \"= [f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No files found\",\n \"HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"):\",\n \"this variable from the Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name\",\n \"(pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f for\",\n \"we're not referencing a layer # Obviously very rough, should change this eventually\",\n \"These populate the files you see once you click into a grid cell\",\n \"change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name #\",\n \"code These populate the files you see once you click into a grid\",\n \"+ deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run\",\n \"fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render(\",\n \"const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from\",\n \"examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck example and move the\",\n \".rst pages with code These populate the files you see once you click\",\n \"raise Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True) pool.map(create_rst,\",\n \"not referencing a layer # Obviously very rough, should change this eventually to\",\n \"\\\"-\\\") deckgl_doc_url = None if \\\"layer\\\" in deckgl_docs_layer_name: # Don't add a deck.gl\",\n \"deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the\",\n \"from multiprocessing import Pool import os import subprocess import sys from const import\",\n \"rtfd.io, set this variable from the Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def\",\n \"variable from the Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name =\",\n \"%s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files =\",\n \"Run the pydeck example and move the .html output subprocess.call( \\\"{python} {fname}; mv\",\n \"into a grid cell on the pydeck gallery page \\\"\\\"\\\" from multiprocessing import\",\n \"os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH\",\n \"# Obviously very rough, should change this eventually to handle views etc deckgl_doc_url\",\n \"\\\".html\\\") # Run the pydeck example and move the .html output subprocess.call( \\\"{python}\",\n \"{html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read()\",\n \"snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f\",\n \"candidate_files: raise Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True)\",\n \"pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files:\",\n \"asset_name + \\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to %s\\\" %\",\n \"Converted %s to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool =\",\n \"populate the files you see once you click into a grid cell on\",\n \"if not candidate_files: raise Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" %\",\n \"the pydeck example and move the .html output subprocess.call( \\\"{python} {fname}; mv {html_src}\",\n \"from templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io, set\",\n \"utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If\",\n \"os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io, set this variable from the Admin panel\",\n \"to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html\",\n \"= os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to\",\n \"the pydeck gallery page \\\"\\\"\\\" from multiprocessing import Pool import os import subprocess\",\n \"pydeck example and move the .html output subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format(\",\n \"convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__ == \\\"__main__\\\":\",\n \"= None if \\\"layer\\\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link\",\n \"move the .html output subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname,\",\n \"import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from\",\n \"GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if\",\n \"this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create\",\n \"= DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\")\",\n \"% (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f\",\n \"rough, should change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE +\",\n \"Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck example\",\n \"def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if\",\n \"%s to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4)\",\n \"handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples\",\n \"set this variable from the Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name):\",\n \"os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to %s\\\"\",\n \"multiprocessing import Pool import os import subprocess import sys from const import DECKGL_URL_BASE,\",\n \"pydeck gallery page \\\"\\\"\\\" from multiprocessing import Pool import os import subprocess import\",\n \"f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files = [f for f in\",\n \"sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name,\",\n \"mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name,\",\n \"into .rst pages with code These populate the files you see once you\",\n \"from the Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\") def create_rst(pydeck_example_file_name): asset_name = to_snake_case_string(file_name=pydeck_example_file_name)\",\n \"templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io, set this\",\n \"files you see once you click into a grid cell on the pydeck\",\n \"deckgl_doc_url = None if \\\"layer\\\" in deckgl_docs_layer_name: # Don't add a deck.gl docs\",\n \"= Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files: raise\",\n \"print(\\\"* Converted %s to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool\",\n \"to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if \\\"layer\\\" in deckgl_docs_layer_name: #\",\n \"os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck example and move the .html output subprocess.call(\",\n \"DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name +\",\n \"subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils\",\n \"f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir\",\n \"in EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir -p\",\n \"to embed pydeck examples into .rst pages with code These populate the files\",\n \".html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck example and move\",\n \"and move the .html output subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name,\",\n \"html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f = open(rst_path, \\\"w+\\\")\",\n \"to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for\",\n \"None if \\\"layer\\\" in deckgl_docs_layer_name: # Don't add a deck.gl docs link if\",\n \"doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR,\",\n \"eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new\",\n \"EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE\",\n \"from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string\",\n \"subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, )\",\n \"[f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No files found to\",\n \"page \\\"\\\"\\\" from multiprocessing import Pool import os import subprocess import sys from\",\n \"\\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path =\",\n \"EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No files found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\"\",\n \"python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url,\",\n \"html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") # Run the pydeck example and move the .html\",\n \"main(): pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not\",\n \"open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def\",\n \"referencing a layer # Obviously very rough, should change this eventually to handle\",\n \"HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not\",\n \"pydeck examples into .rst pages with code These populate the files you see\",\n \"deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if \\\"layer\\\" in deckgl_docs_layer_name: # Don't\",\n \"python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f =\",\n \".html output subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ),\",\n \"DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname = os.path.basename(pydeck_example_file_name).replace(\\\".py\\\", \\\".html\\\") #\",\n \"for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No files found to convert\\\")\",\n \"the .html output subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR\",\n \"\\\"\\\"\\\"Script to embed pydeck examples into .rst pages with code These populate the\",\n \"import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io, set this variable\",\n \"deck.gl docs link if we're not referencing a layer # Obviously very rough,\",\n \"= open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, )\",\n \"docs link if we're not referencing a layer # Obviously very rough, should\",\n \"for rtfd.io, set this variable from the Admin panel raise Exception(\\\"MAPBOX_API_KEY not set\\\")\",\n \"from utils import to_presentation_name, to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): #\",\n \"found to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__\",\n \"to_snake_case_string from templates import DOC_TEMPLATE if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io,\",\n \"candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\\\"No files\",\n \"\\\"w+\\\") print(\\\"* Converted %s to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main():\",\n \"a grid cell on the pydeck gallery page \\\"\\\"\\\" from multiprocessing import Pool\",\n \") python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname),\",\n \"hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f = open(rst_path,\",\n \"to %s\\\" % (pydeck_example_file_name, rst_path)) f.write(doc_source) f.close() def main(): pool = Pool(processes=4) candidate_files\",\n \"grid cell on the pydeck gallery page \\\"\\\"\\\" from multiprocessing import Pool import\",\n \"cell on the pydeck gallery page \\\"\\\"\\\" from multiprocessing import Pool import os\",\n \"# Run the pydeck example and move the .html output subprocess.call( \\\"{python} {fname};\",\n \"deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"*\",\n \"python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source =\",\n \"rst_path = os.path.join(GALLERY_DIR, asset_name + \\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"* Converted %s\",\n \"\\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to %s\\\" % (pydeck_example_file_name, rst_path))\",\n \"shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH,\",\n \"html_src=html_fname, html_dest=HTML_DIR ), shell=True, ) python_code = open(pydeck_example_file_name, \\\"r\\\").read() doc_source = DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name),\",\n \"def main(): pool = Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if\",\n \"asset_name = to_snake_case_string(file_name=pydeck_example_file_name) deckgl_docs_layer_name = asset_name.replace(\\\"_\\\", \\\"-\\\") deckgl_doc_url = None if \\\"layer\\\" in\",\n \"click into a grid cell on the pydeck gallery page \\\"\\\"\\\" from multiprocessing\",\n \"link if we're not referencing a layer # Obviously very rough, should change\",\n \"Pool import os import subprocess import sys from const import DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR,\",\n \"views etc deckgl_doc_url = DECKGL_URL_BASE + deckgl_docs_layer_name # Create new .html examples html_fname\",\n \"with code These populate the files you see once you click into a\",\n \"on the pydeck gallery page \\\"\\\"\\\" from multiprocessing import Pool import os import\",\n \"DECKGL_URL_BASE, EXAMPLE_GLOB, GALLERY_DIR, HTML_DIR, HOSTED_STATIC_PATH from utils import to_presentation_name, to_snake_case_string from templates import\",\n \"if not os.environ.get(\\\"MAPBOX_API_KEY\\\"): # If running for rtfd.io, set this variable from the\",\n \"the files you see once you click into a grid cell on the\",\n \"output subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable, fname=pydeck_example_file_name, html_src=html_fname, html_dest=HTML_DIR ), shell=True,\",\n \"= DOC_TEMPLATE.render( page_title=to_presentation_name(asset_name), snake_name=asset_name, python_code=python_code, hosted_html_path=os.path.join(HOSTED_STATIC_PATH, html_fname), deckgl_doc_url=deckgl_doc_url, ) rst_path = os.path.join(GALLERY_DIR, asset_name\",\n \"very rough, should change this eventually to handle views etc deckgl_doc_url = DECKGL_URL_BASE\",\n \"+ \\\".rst\\\") f = open(rst_path, \\\"w+\\\") print(\\\"* Converted %s to %s\\\" % (pydeck_example_file_name,\",\n \"to convert\\\") subprocess.call(\\\"mkdir -p %s\\\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files) if __name__ ==\",\n \"example and move the .html output subprocess.call( \\\"{python} {fname}; mv {html_src} {html_dest}\\\".format( python=sys.executable,\",\n \"if we're not referencing a layer # Obviously very rough, should change this\"\n]","isConversational":false}}},{"rowIdx":2696,"cells":{"ngram":{"kind":"list like","value":["rotations # inertial frame: located at the origin (waist), but aligned with the","[None] for i in range(0,3): if vector[i] == 0 or vector[i] == 1:","= nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1]","{}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write(' *\\n') out_file.write('","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') # get a string for the enumeration of","# write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj,","2, # right arm 2, 1, 3, 2 # left arm ]) #","+ R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\n')","range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n') for","= nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] =","leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm","end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for i in","j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot contact points absolute position\\n') for","5, # right leg 0, 7, 8, 9, 10, 11, # left leg","17, 18, # right arm 15, 20, 21, 22 # left arm ])","mass) position and velocity\\n') out_file.write(' * feet position, velocity and orientation\\n') out_file.write(' *","vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector =","joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close()","'_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M):","0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'),","nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n') for i in","for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if","= 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')])","# # 17 16 21 # # 18 15 22 # # 19",": derivative of xgi # omi : absolute rotational vector of body i","derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): #","xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1","i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i,","1.0]]) else: return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0],","parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id]","x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] =","left foot contact points absolute position\\n') for i in range(0, nb_contacts): for j","initialization -- //')) != 1: out_file.write(' // -- variables declaration -- //\\n\\n') print_all_declaration(in_temp,","= len(cur_term_split) # no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj,","== 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif","for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if","[rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left foot orientation angle derivatives","+ om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' ,","(flag_jacob)\\n {\\n') for i in range(1, nb_bodies): for j in range(0, 3): out_file.write('","1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count +=","compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon):","// left foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i,","get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1,","= nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] *","def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0","as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left foot orientation","out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' * // //\\n') out_file.write(' * // 17","out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') # get tilde matrix","Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\n') for i in","0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1))","flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg & xgp out_file.write('\\n //","{};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix # save the","sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont","in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in range(0,","and its predecessor # si : sine of the relative angle before body","in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute orientation","return true if it has a shape 'x%a_%b' (indexes %a, %b also returned)","angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso orientation matrix as","or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name =","xgj, xj_print, x_vector, index): # loop on all the joints for i in","i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right foot","derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q))","term_list[0] == '': term_list.pop(0) result = 0 # loop on all terms for","all the joints for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1)","point of body i to its COM (center of mass) G_i,\\n') out_file.write(' *","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute position\\n') for i in range(0,3):","der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {}","trunk 15, 16, 17, 18, # right arm 15, 20, 21, 22 #","// global com absolute velocity\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] =","joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' *","right arm 2, 1, 3, 2 # left arm ]) # parent index","nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0]","R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb","body to the current body i # (previous body is not always body","= atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2],","get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T","Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close() # output file","joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id',","as f: # loop on all the lines for line in f: cut_line","angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the wrists position,","= sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] =","for the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {') count =","flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg & xgp out_file.write('\\n","# # 20 01 24 # # 02 08 # # 03 09","out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out inputs","cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis def","// //\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin: in the waist,","= 3 * [None] for i in range(0,3): if vector[i] == 0 or","Ri : absolute rotational matrix\\n') out_file.write(' * Rti : transpose matrix of Ri\\n')","xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T","rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply","= nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None]","matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix # save the symbolic","new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else:","position vector from the anchor point of the previous body to the current","initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt,","i-1), expressed in the relative # frame of the previous body # DGi","1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write(' // waist orientation","two pitch hip rotations # inertial frame: located at the origin (waist), but","si : sine of the relative angle before body i # ci :","body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1],","in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot absolute orientation jacobian\\n') out_file.write('","i # Omi : rotational vector from the previous body to the current","'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an","'') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n // jacobian","j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] =","returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a,","mass) G_i, # expressed in the relative frame of the current body i","range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot absolute","for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i]","i in range (1, nb_bodies): flag_print = 0 for j in range(0, 3):","right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18]","1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0],","= elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive neg_flag =","-{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in,","R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23]","24 # # 02 08 # # 03 09 # # 04 10","left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg,","= 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n')","if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] =","omi : absolute rotational vector of body i # Ri : absolute rotational","1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n')","i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' //","declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in range(min,","in term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no","xp_r[i])) out_file.write('\\n') out_file.write(' // right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0],","'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n //","= sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')])","c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","= sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0,","if (flag_jacob)\\n {\\n') flag_first = 0 for i in range (1, nb_bodies): flag_print","of 'xgi' # Rji : jacobian of 'Ri' # return true if it","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute position\\n') for i in range(0,3):","*\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' * // //\\n') out_file.write(' * // 17 16","write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\n //","body is not always body i-1), expressed in the relative # frame of","try: int(value) return True except: return False # return true if it has","= int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except: return False,","in_out inputs and outputs class\\n') out_file.write(' *\\n') out_file.write(' * computation of:\\n') out_file.write(' *","get vector axis def get_vector_axis(axis, direct, elem): if direct: if axis == 1:","com absolute velocity\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first","k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k],","x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string,","matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {}","(1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the vector elements","3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative of","in range(1, nb_bodies): flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k]","# left leg 1, 2, 3, # trunk 2, 1, 3, 2, #","velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ')","= {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist absolute orientation jacobian\\n') out_file.write(' if","{};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left wrist absolute velocity\\n') for i in range(0,3):","'./{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) #","= elem_split[1] else: print('Error: {} instead of 1 or 2 in negative detection","08 //\\n') out_file.write(' * // 03 09 //\\n') out_file.write(' * // 04 10","// right wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i,","compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var,","def get_vector_axis(axis, direct, elem): if direct: if axis == 1: return np.array([[elem], [0.0],","positions and velocities\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] *","13 # # # ###################### # # origin: in the waist, middle point","xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis,","absolute rotational matrix # Rti : transpose matrix of Ri # xji :","out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU - angles velocity\\n') for","1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count +=","R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file,","# one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj,","sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) #","= r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] =","# loop on all terms for cur_term in term_list: # detect products cur_term_split","out_file.write(' * Rti : transpose matrix of Ri\\n') out_file.write(' * xji : jacobian","9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in range(0, 9): out_file.write('","jacobian rotation matrices out_file.write('\\n // jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0:","return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if direct: if","the file def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f:","out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on all","sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'),",", 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id'","xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error:","xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1 elif","= sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0,","xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0],","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] =","for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] =","{};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i]))","1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if","j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac !=","out_file.write(' * Di : position vector from the anchor point of the previous","0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None]","# omi : absolute rotational vector of body i # Ri : absolute","and velocity\\n') out_file.write(' * feet position, velocity and orientation\\n') out_file.write(' * waist and","\\\\param[in,out] in_out inputs and outputs class\\n') out_file.write(' *\\n') out_file.write(' * computation of:\\n') out_file.write('","len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count",">= 6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief","out_file.write(' * // //\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin: in","i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all","(info from IMU) # # Di : position vector from the anchor point","absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0,","xj, xgj, xgj_print, x_vector, index): # loop on all the joints for i","body i\\n') out_file.write(' * xgi : absolute position vector of the COM G_i","# trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29'","i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in","== 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1,","in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ')","= 9 * [None] for i in range(0,3): for j in range(0,3): if","nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\n') out_file.write('\\n // IMU -","Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0,","the relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write(' * DGi :","-- //')) != 1: out_file.write(' // -- variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file,","write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n // jacobian com absolute positions\\n')","j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac !=","def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b","R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the","-> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5],","R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1]))","for i in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} =","[0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine,","out_file.write(' // feet absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] =","out_file.write(' * xgji : jacobian of \\'xgi\\'\\n') out_file.write(' * Rji : jacobian of","23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic","flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j],","of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{}","if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print","- angles velocity\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i))","\\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center of mass position","i == 2 and j == 2: out_file.write(';\\n') else: out_file.write(', ') # print","'') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n // jacobian","xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id]","sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) #","else: out_file.write(' + ') out_file.write(' // global com absolute position\\n') for i in","xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\n // rotation matrices\\n') R","cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not","in range (1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac =","{}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out","to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q',''))","sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')])","generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg,","xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if","range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of","|| (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write('","i # (previous body is not always body i-1), expressed in the relative","out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') # from an orientation","in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{}","16 21 //\\n') out_file.write(' * // 18 15 22 //\\n') out_file.write(' * //","# compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x,","* Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing","'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print =","c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot","for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first","R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot","axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis","out_file.write('\\n') else: out_file.write(' }\\n\\n') # from an orientation matrix, compute the roll, pitch,","derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): #","x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max","Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i]","out_file.write(' * origin: in the waist, middle point between the two pitch hip","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n')","'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name,","out_file.write('\\n') out_file.write(' // right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\n') out_file.write('\\n // IMU","nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for","sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right","temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script #","sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')])","body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write('","= int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len","# rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result","if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] =","'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis,","relative angle before body i\\n') out_file.write(' * ci : cosine of the relative","arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name","in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute orientation","out_file.write(' * // 04 10 //\\n') out_file.write(' * // 05 11 //\\n') out_file.write('","vector and replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name):","out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics computation for the COMAN","not always body i-1), expressed in the relative\\n') out_file.write(' * frame of the","{};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left foot absolute velocity\\n') for i in range(0,3):","= r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb)","range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist absolute","right wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","= sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if vector[i] ==","nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1)","Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') ,","1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n // -- Collecting results --","\\'xi\\'\\n') out_file.write(' * xgji : jacobian of \\'xgi\\'\\n') out_file.write(' * Rji : jacobian","= int(value.split('_')[1][1]) return True, a, b, c except: return False, -1, -1, -1","symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first",": absolute position vector of the COM G_i of body i\\n') out_file.write(' *","= x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T","range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for k in","of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string =","'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28',","flag_print: out_file.write('\\n') return new_vector # write symbolic matrix and replace symbolic variable by","sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] =","sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot =","R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l","# right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')])","also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c =",", only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result # write","// left wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i,","sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')])","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,","out_write.write(';\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with","except: return False # return true if it has a shape 'R%a_%b%c' (indexes","point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on all","in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac !=","k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac","if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip())","absolute rotational vector of body i # Ri : absolute rotational matrix #","cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not","Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i]","3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0,","if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(',","body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of","1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0]","else: new_string = cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for","for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac","out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out inputs and outputs class\\n') out_file.write(' *\\n') out_file.write('","= sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist","start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for i","= xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb =","elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] =","xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i],","write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '')","sines\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n //","result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag,","in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left foot absolute velocity\\n') for i","0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] =","om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3,","b = int(value.split('_')[1]) return True, a, b except: return False, -1, -1 #","out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1]))","= 0 with open(in_file,'r') as f: # loop on all the lines for","end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for i in","xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i]","14 23 //\\n') out_file.write(' * // 20 01 24 //\\n') out_file.write(' * //","0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write('","sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')])","# xpi : derivative of xi # xgpi : derivative of xgi #","== 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1,","on all the joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3,","'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' ,","sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name,","the relative frame of the current body i # Omi : rotational vector","if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] #","Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] =","# sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag,","out_file.write(' // right foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] =","{} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix #","range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3,","out_file.write('\\n') out_file.write(' // global com absolute velocity\\n') for i in range(0, 3): out_file.write('","on all the vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj,","xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id =","## COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')])","out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range (1, nb_bodies):","# waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right","to the current body i # (previous body is not always body i-1),","else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n')","1) out_file.write('\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot',","of the anchor point of body i\\n') out_file.write(' * xgi : absolute position","velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' //","out_file.write('\\n') return new_vector # write symbolic matrix and replace symbolic variable by its","j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_',",": derivative of xi\\n') out_file.write(' * xgpi : derivative of xgi\\n') out_file.write(' *","b, c except: return False, -1, -1, -1 # return true if it","= '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j]","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","out_write.write(';\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0,","* COM (center of mass) position and velocity\\n') out_file.write(' * feet position, velocity","in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j,","import re import os ###################### # # # 17 16 21 # #","elif cur_len == 2: # negative neg_flag = 1 pos_str = elem_split[1] else:","= cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot =","out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies):","3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n') for i in","relative velocities\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) #","out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector","= sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i","= R[i].T # jacobian rotation matrices out_file.write('\\n // jacobian rotation matrices\\n') out_file.write(' if","Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_',","the waist, middle point between the two pitch hip rotations # inertial frame:","elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as","// 17 16 21 //\\n') out_file.write(' * // 18 15 22 //\\n') out_file.write('","in the waist, middle point between the two pitch hip rotations # inertial","out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') # get a string","* expressed in the relative frame of the current body i\\n') out_file.write(' *","xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] =","nb_bodies): flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k] != None:","2, 1, 3, 2, 1, 2, # right leg 2, 1, 3, 2,","i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist","in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n') for","of 'Ri' # return true if it is a float def isInt(value): try:","None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} =","= sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j]","xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] =","save_matrix # write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj,","om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1)","inertial frame: located at the origin (waist), but aligned with the ground (info","Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies,","Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point","= x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i]","{\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot =","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] =","open(in_file,'r') as f: # loop on all the lines for line in f:","joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,","out_file.write('\\n\\n // -- symbolic computation -- //\\n') # Rj, xj, xgj and xgj","[elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([])","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n')","Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'),","R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian","write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all the joints","current body i # Omi : rotational vector from the previous body to","matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j],","= -result return result # compute the derivative of an expression (for jacobian)","= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k]))","out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center of mass position and","0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] =","= x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot","1: count = 0 with open(in_file,'r') as f: # loop on all the","range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0:","der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len == 2: result","range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else:","range (1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj,","matrix of Ri\\n') out_file.write(' * xji : jacobian of \\'xi\\'\\n') out_file.write(' * xgji","the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2:","\\\\brief forward kinematics computation for the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n')","3) for i in range(0, 3): for j in range(0, 3): R[0][i,j] =","with open(in_file, 'r') as f: # loop on all the lines for line","relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write(' * Rdi : rotational","{};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right foot absolute velocity\\n') for i in range(0,3):","range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print","flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write('",": sine of the relative angle before body i # ci : cosine","declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy temporary file out_file.write(line) out_file.close()","Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')])","= atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def","'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' ,","(from origin, expressed in the inertial frame) # of the anchor point of","for i in range (1, nb_bodies): flag_print = 0 for j in range(0,9):","# left leg 0, 13, 14, # trunk 15, 16, 17, 18, #","new_matrix = sp.zeros(3, 3) # loop on all the matrix elements for j","{} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic","its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print =","for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1)","== 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else:","Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'),","* // 19 14 23 //\\n') out_file.write(' * // 20 01 24 //\\n')","sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] =","def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for","0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else:","//\\n') out_file.write(' * // 19 14 23 //\\n') out_file.write(' * // 20 01","out_file.write(' * Rdi : rotational matrix between body i and its predecessor\\n') out_file.write('","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0,","masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' ,","Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] =","if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(',","the current body i \\n') out_file.write(' * (previous body is not always body","out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0","// left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop","sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')])","nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj,","//\\n') out_file.write(' * // 04 10 //\\n') out_file.write(' * // 05 11 //\\n')","out_file.write('}\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for","sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12]","waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'),","points absolute position\\n') for i in range(0, nb_contacts): for j in range(0, 3):","Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'),","of the file def write_file_end(out_file): out_file.write('}\\n') # print matrix components declaration def write_matrix_declaration(out_file,","sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] =","sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')])","COM G_i of body i\\n') out_file.write(' * xpi : derivative of xi\\n') out_file.write('","# # 02 08 # # 03 09 # # 04 10 #","out_file.write(' * Ri : absolute rotational matrix\\n') out_file.write(' * Rti : transpose matrix","if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global","R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] =","= 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k],","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')])","body i # ci : cosine of the relative angle before body i","vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name,","== 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else:","leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21',","s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write('","out_file.write(' * Omi : rotational vector from the previous body to the current","cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1):","cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len == 1: result +=","nb_contacts): for j in range (1, nb_bodies): flag_print = 0 for k in","# Rti : transpose matrix of Ri # xji : jacobian of 'xi'","-- Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file,","flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*',''))","if count >= nb_max_char: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n')","range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print","element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count =","2, 1, 3, 2 # left arm ]) # parent index parent_body_index =","'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', #","= 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write(' // waist","range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if","int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len ==","Rji : jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the","joints for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop","[-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis ==","the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] =","outputs out_file.write(' // right foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}]","rotational matrix\\n') out_file.write(' * Rti : transpose matrix of Ri\\n') out_file.write(' * xji","for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i],",", 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', #","xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def","elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b,","sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11]",", sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'),","theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\n')","len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' // -- variables declaration","in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot contact points jacobian\\n')","sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] =","out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0","body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the waist and the","* (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] =","sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] =","17 16 21 # # 18 15 22 # # 19 14 23","count = 0 with open(in_file,'r') as f: # loop on all the lines","# computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T","for i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j]","from the anchor point of body i to its COM (center of mass)","the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write(' *","orientation\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\n') out_file.write(' *\\n') out_file.write('","range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0,","R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045)","out_file.write(' * frame of the previous body\\n') out_file.write(' * Rdi : rotational matrix","= nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i","def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min,","Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0,","in range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count","out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n","return False, -1, -1 # count the number of 'elem' in the file","[-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([])","= cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i in range(1,","j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac","(Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file,","04 10 # # 05 11 # # 06 12 # # 07","loop on all the lines for line in f: # declaration if len(line.split('//","np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2:","aligned with the ground (info from IMU) # # Di : position vector","# print the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if","orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for","hip rotations # inertial frame: located at the origin (waist), but aligned with","+ {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j","= line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2)","// global com absolute position\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] =","j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]]","Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w')","Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'),","f: # loop on all the lines for line in f: cut_line =","end of the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w')","x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of","sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')])","1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] =","save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = '","x & xp out_file.write('\\n // anchor point absolute positions and velocities\\n') x =","Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M","orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left foot","matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix =","= 0 if count != 0: out_write.write(';\\n') # get tilde matrix def get_tilde(v):","21 //\\n') out_file.write(' * // 18 15 22 //\\n') out_file.write(' * // 19","nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i]","epsilon): if epsilon > 0: # epsilon = 1 -> pitch angle in","if not flag_first: flag_first = 1 flag_print = 1 elif not flag_print: flag_print","[vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)]","in range (1, nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac","out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if","range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1)","symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n // -- symbolic computation","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write('","save_matrix = 9 * [None] for i in range(0,3): for j in range(0,3):","= xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print,","in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j","write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i]","'r') as f: # loop on all the lines for line in f:","{};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot contact points absolute","for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string","def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in range(min, max+1):","= write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i]","x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot","Rdi : rotational matrix between body i and its predecessor\\n') out_file.write(' * si","and replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix","an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop","time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{}","15 22 //\\n') out_file.write(' * // 19 14 23 //\\n') out_file.write(' * //","range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left foot jacobian\\n') out_file.write('","R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_',","i-1), expressed in the relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write('","om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations","out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n')","%a, %b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b =","0 pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag = 1","relative angle before body i\\n') out_file.write(' *\\n') out_file.write(' * xi : absolute position","R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1","Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i]","0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12]","nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j]","isInt(value): try: int(value) return True except: return False # return true if it","all the lines for line in f: # declaration if len(line.split('// -- variables","2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis","om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id],","= symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:","[-elem]]) else: return np.array([]) # compute the derivative of an element (for jacobian)","of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center of mass",": jacobian of 'Ri' # return true if it is a float def","predecessor # si : sine of the relative angle before body i #","}\\n\\n') out_file.write(' // right foot contact points absolute position\\n') for i in range(0,","symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic","nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] =","== 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem],","velocities\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0]","'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk","s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0],","def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\n') out_file.write('\\n //","new_vector = sp.zeros(3, 1) # loop on all the vector elements for j","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0]","elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive neg_flag","sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1))","sympy as sp import re import os ###################### # # # 17 16","in the relative # frame of the previous body # DGi : position","(1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2, 1,","nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts *","len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip()))","2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis","for i in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print =","-v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine):","double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\n')","[0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0,","out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i ==","* [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot","None: if not flag_first: flag_first = 1 flag_print = 1 elif not flag_print:","range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts","cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not","cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n')","# waist 0, 1, 2, 3, 4, 5, # right leg 0, 7,","for i in range (1, nb_bodies): flag_print = 0 for j in range(0,","= sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'),","i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative","range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot","anchor point of body i to its COM (center of mass) G_i,\\n') out_file.write('","flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if","i # ci : cosine of the relative angle before body i #","absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n')","s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{}","1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]])","2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]])","i\\n') out_file.write(' * Ri : absolute rotational matrix\\n') out_file.write(' * Rti : transpose","in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] =","symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix","Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names,","axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0],","+ R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing","# 19 14 23 # # 20 01 24 # # 02 08","out_file.write('\\n') out_file.write(' // right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0])","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0])","-sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get","rotational matrix between body i and its predecessor\\n') out_file.write(' * si : sine","right foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i]))","}\\n') # omega out_file.write('\\n // joint absolute velocities\\n') Om = nb_bodies*[None] om =","wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n')","# declaration if len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' //","R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot","print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with","out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n')","*\\n') out_file.write(' * xi : absolute position vector (from origin, expressed in the",", sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'),","previous body to the current body i # (previous body is not always","sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist =","open(in_file, 'r') as f: # loop on all the lines for line in","frame of the previous body # Rdi : rotational matrix between body i","np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return","* [None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i]","Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0,","if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return","omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out,","in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] =","# 17 16 21 # # 18 15 22 # # 19 14","R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\n // jacobian","matrix, compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix,","else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name,","sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'),","* Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T *","vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = '","out_file.write(' * si : sine of the relative angle before body i\\n') out_file.write('","Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' //","out_file.write(' // right foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] =","1 pos_str = elem_split[1] else: print('Error: {} instead of 1 or 2 in","s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot","matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_bodies):","body i\\n') out_file.write(' * Omi : rotational vector from the previous body to","out_file.write(' // right foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso orientation angle derivatives","kinematics computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out inputs and outputs class\\n') out_file.write('","for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for","= atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2],","= 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') # get","the end of the file def write_file_end(out_file): out_file.write('}\\n') # print matrix components declaration","0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name","in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the","# get vector axis def get_vector_axis(axis, direct, elem): if direct: if axis ==","xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot +","xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =","4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id]","out_file.write('\\n // jacobian com absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None]","# left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')])","the previous body # DGi : position vector from the anchor point of","'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj,","= Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj,","Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13]",", 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id'","//\\n') out_file.write(' * // 20 01 24 //\\n') out_file.write(' * // 02 08","out_file.write('\\n') out_file.write(' // global com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for i in","# save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix =","rotation matrices\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] =","np.array([0, # waist 2, 1, 3, 2, 1, 2, # right leg 2,","# left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', #","loop on all the joints for i in range (1, nb_bodies): new_matrix =","(info from IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di : position vector from the","in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left wrist absolute","x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_',","om, 0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis,","new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix # save the symbolic vector","count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f: # loop on","R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file,","sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4]","elem_split[1] else: print('Error: {} instead of 1 or 2 in negative detection !'.format(cur_len))","def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None]","om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for","== 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else:","roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon","velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' //","= cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso =","end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if","j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists absolute orientation\\n') for i in range(0,","= {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector # write","position vector from the anchor point of body i to its COM (center","== 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else:","== 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis ==","torso orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist","xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names,","absolute position vector (from origin, expressed in the inertial frame) # of the","True except: return False # return true if it has a shape 'R%a_%b%c'","-sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine,","expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all","result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj,","'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write('","Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1]","before body i # # xi : absolute position vector (from origin, expressed","prefix): out_file.write(' double ') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix,","np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine],","out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n // -- Collecting results -- //\\n\\n') com_compute(out_file,","R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max):","out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}]","term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product","xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables","R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3)","i+1, j+1)) if i == 2 and j == 2: out_file.write(';\\n') else: out_file.write(',","* frame of the previous body\\n') out_file.write(' * DGi : position vector from","range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n') for i","of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): #","return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis ==","out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com absolute velocity\\n') for i in","cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not","'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg","= symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:","angles and derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' * // //\\n') out_file.write('","i\\n') out_file.write(' * xpi : derivative of xi\\n') out_file.write(' * xgpi : derivative","[0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis == 1: return np.array([[1.0,","x_r[i])) out_file.write('\\n') out_file.write(' // right foot absolute velocity\\n') for i in range(0,3): out_file.write('","matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n')","i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left wrist","if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_bodies): for j","== 0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1","of \\'xgi\\'\\n') out_file.write(' * Rji : jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut","forward kinematics computation for the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum","Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) #","np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if direct: if axis","parent_body_index = np.array([ -1, # waist 0, 1, 2, 3, 4, 5, #","np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id'","range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min","and orientation\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\n') out_file.write(' *\\n')","'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id]","inertial frame) # of the anchor point of body i # xgi :","}\\n\\n') out_file.write(' // left foot contact points absolute position\\n') for i in range(0,","sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) #","yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso orientation matrix as angles [rad]\\n')","of mass) G_i,\\n') out_file.write(' * expressed in the relative frame of the current","= sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'),","out_file.write('\\n') out_file.write(' // right wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}]","out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute","# loop on all the vector elements for j in range(0, 3): new_vector[j]","enumeration\\n') out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)): count +=","# writing outputs out_file.write(' // right foot absolute position\\n') for i in range(0,3):","left leg 0, 13, 14, # trunk 15, 16, 17, 18, # right","'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23',","foot contact points absolute position\\n') for i in range(0, nb_contacts): for j in","Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i in","23 //\\n') out_file.write(' * // 20 01 24 //\\n') out_file.write(' * // 02","xgji : jacobian of \\'xgi\\'\\n') out_file.write(' * Rji : jacobian of \\'Ri\\'\\n') out_file.write('","(1, nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj,","== 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result +=",": absolute position vector (from origin, expressed in the inertial frame)\\n') out_file.write(' *","= 0 # loop on all terms for cur_term in term_list: # detect","= ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count ==","out_file.write(' *\\n') out_file.write(' * Di : position vector from the anchor point of","j == 2: out_file.write(';\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file,","vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print =","* Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i],","# waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0,","sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'),","with open(in_temp, 'r') as f: # loop on all the lines for line","def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f: # loop","* origin: in the waist, middle point between the two pitch hip rotations\\n')","range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0:","###################### # # origin: in the waist, middle point between the two pitch","of body i # Ri : absolute rotational matrix # Rti : transpose","0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif","Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1]","body i\\n') out_file.write(' *\\n') out_file.write(' * xi : absolute position vector (from origin,","out_file.write(' // right wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","elif count >= 6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n')","15 22 # # 19 14 23 # # 20 01 24 #","of xgi\\n') out_file.write(' * omi : absolute rotational vector of body i\\n') out_file.write('","om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3,","+ R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r =","Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'),","0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0])","09 //\\n') out_file.write(' * // 04 10 //\\n') out_file.write(' * // 05 11","xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T","der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted ,","matrices out_file.write('\\n // jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","= write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies):","jacobian x out_file.write('\\n // jacobian anchor point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","= -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{},","between the two pitch hip rotations # inertial frame: located at the origin","= 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else:","for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if","(!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot","16 21 # # 18 15 22 # # 19 14 23 #","in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont =","1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len","* inertial frame: located at the origin (waist), but aligned with the ground","') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\n')","sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left","/ c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation","in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right foot jacobian\\n')","vector from the anchor point of the previous body to the current body","[rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot =","Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts","= nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot =","write_file_end(out_file): out_file.write('}\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ')","leg 0, 13, 14, # trunk 15, 16, 17, 18, # right arm","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results","i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] =","//\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin: in the waist, middle","om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the wrists position, velocity and orientation def","out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write('","float def isInt(value): try: int(value) return True except: return False # return true","if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1","// //\\n') out_file.write(' * // 17 16 21 //\\n') out_file.write(' * // 18","for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >=","= np.array([0, # waist 2, 1, 3, 2, 1, 2, # right leg","derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) #","y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for","in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right foot absolute","range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0])","'') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\n // jacobian rotation matrices\\n')","Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1)","Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb","[sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine,","1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right foot","in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j]","k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n')","pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag = 1 pos_str","end_name) save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix,","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","# derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len","Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont =","0 for k in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first:","trunk 2, 1, 3, 2, # right arm 2, 1, 3, 2 #","# list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '':","orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id,","= symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) #","np.array([ -1, # waist 0, 1, 2, 3, 4, 5, # right leg","axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0],","3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot contact","2, 3, 4, 5, # right leg 0, 7, 8, 9, 10, 11,","Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb","for line in f: cut_line = line.split(elem) if len(cut_line) == 2: count +=","line in f: cut_line = line.split(elem) if len(cut_line) == 2: count += 1","cosine of the relative angle before body i\\n') out_file.write(' *\\n') out_file.write(' * xi","points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0,","[0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3:","cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n') for i in range(1, nb_bodies): out_file.write(' s{}","+= 1 if count >= nb_max_line: out_write.write(';\\n') count = 0 if count !=","out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j","* Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\n') for i","# # 04 10 # # 05 11 # # 06 12 #","point between the two pitch hip rotations # inertial frame: located at the","xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i]","if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com absolute velocity\\n') for","f: # loop on all the lines for line in f: # declaration","position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot","sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')])","xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else:","arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint names joint_id_names","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0,","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,","# save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector =","= int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return False, -1,","1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n')","velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' //","in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute position\\n')","xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None]","new_vector # write symbolic matrix and replace symbolic variable by its name def","sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] =","= sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'),","l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb =",": jacobian of 'xgi' # Rji : jacobian of 'Ri' # return true","3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if","main kinematics computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out inputs and outputs class\\n')","[rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso orientation matrix as angles","%b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0])","range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix,","= print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index,","pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8]))","0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) #","from IMU) # # Di : position vector from the anchor point of","atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0],",", # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id'","om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute","R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute","position\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' //","0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name","{};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix","// joint cosines\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i]))","of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element","lines for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ')","for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write('","result # write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n')","matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name,","j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return","// 05 11 //\\n') out_file.write(' * // 06 12 //\\n') out_file.write(' * //","# loop on all the lines for line in f: # declaration if","out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write(' // global","right foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' //","1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result +=","* (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] =","R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2]","# symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end","save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3,","current body i # (previous body is not always body i-1), expressed in","Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag:",", 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id'",": absolute rotational matrix # Rti : transpose matrix of Ri # xji","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name,","1 !'.format(cur_len-1)) exit() return result # write the beginning of the file def","def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n')","'M_28', 'M_29' # left arm ]) # joint names joint_id_names = np.array(['0', #","joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225)","[rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and orientation","enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string","file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script # rotation","-0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23,","or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name =","xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file,","// right foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write('","parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i]","orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso orientation","* Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj,","out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\n') count = 0","elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return","= nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3,","of body i # xpi : derivative of xi # xgpi : derivative","xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj,","der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only implemented for 0","return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis ==","left arm ]) # parent index parent_body_index = np.array([ -1, # waist 0,","1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis","'omega_torso', 'torso') # compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names,","= -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb)","save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix","flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k] != None: if","left leg 1, 2, 3, # trunk 2, 1, 3, 2, # right","// jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25',","# Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None]","previous body # DGi : position vector from the anchor point of body","if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine],","with open(in_file,'r') as f: # loop on all the lines for line in","in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n","cur_jac != 0: if not flag_first: flag_first = 1 flag_print = 1 elif","nb_max_char): count = 0 with open(in_file,'r') as f: # loop on all the","symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write","nb_bodies): flag_print = 0 for k in range(0, 3): if xj_print[i][j][k] != None:","output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary","right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if","= {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1)","'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17',","velocities\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3,","for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1)))","return result # compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj,","line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1)","return new_vector, save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name,","else: return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0,","in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for k","out_file.write('\\n\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) #","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k,","0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return","angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso","= 1.0 / c_y_torso;\\n\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id],","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,","9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index),","'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id',","x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file,","in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0],","// wrists absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i,","i)) out_file.write('\\n // joint cosines\\n') for i in range(1, nb_bodies): out_file.write(' c{} =","== 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count","out_file.write(' if (flag_jacob)\\n {\\n') for i in range(1, nb_bodies): for j in range(0,","i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right foot","cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not","ground (info from IMU) # # Di : position vector from the anchor","arm 15, 20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index) ##","int(value.split('_')[1][1]) return True, a, b, c except: return False, -1, -1, -1 #","in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in","position\\n') for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}]","waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id],","= r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] =","out_file.write('\\n // joint absolute velocities\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde =","results out_file.write('\\n // -- Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg,","end of the file def write_file_end(out_file): out_file.write('}\\n') # print matrix components declaration def","# main script # rotation axis for each joint before body i (1:x,","r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb","ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center of mass position and velocity def com_compute(out_file,","# right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0,","1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max","pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8]))","the current body i # Omi : rotational vector from the previous body","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists absolute orientation\\n') for i","theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and orientation def","om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist","}\\n') # results out_file.write('\\n // -- Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names,","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] =","# 05 11 # # 06 12 # # 07 13 # #","yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n')","(R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for","(s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) #","return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis ==","out_file.write(' * // 05 11 //\\n') out_file.write(' * // 06 12 //\\n') out_file.write('","sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'),","// right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i]","j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write('","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}]","# trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id'","write symbolic vector and replace symbolic variable by its name def write_symb_vector(out_file, vector,","((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n')","i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2,","x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies,","if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if","return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis","sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'),","xgpi : derivative of xgi\\n') out_file.write(' * omi : absolute rotational vector of","nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n') else: out_file.write('","if i == max: out_file.write(';\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file,","orientaion angles and derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' * // //\\n')","a, b, c except: return False, -1, -1, -1 # return true if","velocity and orientation\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\n') out_file.write('","out_file.write(' }\\n\\n') out_file.write(' // right foot contact points absolute position\\n') for i in","0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if","symb_var, der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0]","= 0 for i in range(1, len(joint_id_names)): count += 1 if i ==","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id'","x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}]","# angles (position and derivative) of the waist and the torso def torso_waist_angles(out_file,","open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp,","out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') # print all declarations","left arm ]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id',","nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot","# # 19 14 23 # # 20 01 24 # # 02","def write_file_end(out_file): out_file.write('}\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double","matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left foot","omi : absolute rotational vector of body i\\n') out_file.write(' * Ri : absolute","Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'),","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine],","om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T *","left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9]","of body i to its COM (center of mass) G_i, # expressed in","== 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count ==","writing outputs out_file.write(' // right foot absolute position\\n') for i in range(0,3): out_file.write('","sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) ||","if count != 0: out_write.write(';\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char):","neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead of 1 or","symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first","out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line:","Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0]","= 0 for i in range(0,3): for j in range(0,3): if matrix[i,j] ==","as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write('","out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist","\\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics computation for","3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if","left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22]","v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct:","the joints for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) #","derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left foot orientation angle","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists absolute orientation\\n') for i in range(0, 9):","= om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot =","out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute","for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3","count += 1 if count >= nb_max_line: out_write.write(';\\n') count = 0 if count","'Rfoot') out_file.write('\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot',","inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1],","return true if it has a shape 'R%a_%b%c' (indexes %a, %b, %c also","out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write(' // global com absolute position\\n') for i","sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19]","len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\n') count = 0 if count","(string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","its predecessor # si : sine of the relative angle before body i","R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon","= isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag: result","body i and its predecessor # si : sine of the relative angle","in the relative frame of the current body i\\n') out_file.write(' * Omi :","x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot *","= in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable by its","v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis,","# trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15]","absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write('","ci : cosine of the relative angle before body i\\n') out_file.write(' *\\n') out_file.write('","= sp.zeros(3, 1) save_vector = 3 * [None] for i in range(0,3): if","out_file.write(' // left wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] =","Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)),","sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses","inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write(' //","{} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector #","vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for","theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part,","np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis","= x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] =","Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all the joints for","xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right","sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right","# temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of","(1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj,","for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\n') else:","// -- symbolic computation -- //\\n') # Rj, xj, xgj and xgj (jacobian)","right leg 2, 1, 3, 2, 1, 2, # left leg 1, 2,","0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'),","07 13 # # # ###################### # # origin: in the waist, middle","range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0:","out_file.write(' // -- variables initialization -- //\\n') out_file.write('\\n // IMU - rotation matrices\\n')","0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0,","* // 07 13 //\\n') out_file.write(' * // //\\n') out_file.write(' * ////////////////////////\\n') out_file.write('","# return true if it is a float def isInt(value): try: int(value) return","tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0],","axis for each joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0,","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot contact points absolute position\\n')","file def write_file_end(out_file): out_file.write('}\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write('","vector from the previous body to the current body i \\n') out_file.write(' *",">= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string,","compute the center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M,","== 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] =","the previous body to the current body i \\n') out_file.write(' * (previous body","Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm","len(cur_term_split) # no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj,","0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name","Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for","lines for line in f: # declaration if len(line.split('// -- variables initialization --","the current body i\\n') out_file.write(' * Omi : rotational vector from the previous","+ {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1],","'{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the file def write_file_end(out_file):","[rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\n')","}\\n\\n') out_file.write(' // left foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}]","neg_flag: result = -result return result # compute the derivative of an expression","other else: print('Error: {} * counted , only implemented for 0 or 1","range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i]","sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'),","in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first:","get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj,","str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first:","= nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print =","nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0,","2, # left leg 1, 2, 3, # trunk 2, 1, 3, 2,","symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write","nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for","detection !'.format(cur_len)) exit() # compute derivative result = 0 # cosine if pos_str","out_file.write(' }\\n\\n') out_file.write(' // left foot absolute position\\n') for i in range(0,3): out_file.write('","return true if it is a float def isInt(value): try: int(value) return True","elem_split[0] elif cur_len == 2: # negative neg_flag = 1 pos_str = elem_split[1]","19 14 23 //\\n') out_file.write(' * // 20 01 24 //\\n') out_file.write(' *","symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix,","leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] =","body i # xgi : absolute position vector of the COM G_i of","= 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): #","= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1))","{}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\n') count = 0 if","sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 ,","= cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot =","out_file.write('\\n') out_file.write(' // right foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}]","torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1)","and velocities\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0]","cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis == 1:","* Rdi : rotational matrix between body i and its predecessor\\n') out_file.write(' *","for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' //","sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a rotation matrix def","cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1],","else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write(' *\\n') out_file.write(' *","out_file.write(' }\\n\\n') # get a string for the enumeration of joints def get_string_enum(cur_string):","1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot =","# return true if it has a shape 'x%a_%b' (indexes %a, %b also","of the waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write('","of the previous body\\n') out_file.write(' * Rdi : rotational matrix between body i","= nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j","only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result # write the",", # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm","1 or 2 in negative detection !'.format(cur_len)) exit() # compute derivative result =","sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in","else: return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif","Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3","* Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj,","isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True,","right foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write('","'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies,","j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first =","!= nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') # from an orientation matrix, compute the","nb_max_char: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') # get tilde","a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix","direct, elem): if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif","s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write('","'waist') out_file.write('\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso')","if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one","out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k ==","leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' ,","xgj, der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative","= sp.zeros(3, 3) # loop on all the matrix elements for j in","new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name)","Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3,","always body i-1), expressed in the relative\\n') out_file.write(' * frame of the previous","out_file.write('\\n // joint relative velocities\\n') for i in range(1, nb_bodies): out_file.write(' Om{} =","names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', #","+= sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d,","between the two pitch hip rotations\\n') out_file.write(' * inertial frame: located at the","= 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right","= {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right foot absolute velocity\\n') for i in","left wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","[None] # rotation matrices out_file.write('\\n // rotation matrices\\n') R = nb_bodies*[None] Rt =","out_file.write(' // wrists absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] =","i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i]","# inertial frame: located at the origin (waist), but aligned with the ground","result # compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj,","elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')","matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i in","for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj,","symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first","elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return","= xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' //","of:\\n') out_file.write(' * COM (center of mass) position and velocity\\n') out_file.write(' * feet","joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0]","& xp out_file.write('\\n // anchor point absolute positions and velocities\\n') x = nb_bodies*[None]","1, 2, # left leg 1, 2, 3, # trunk 2, 1, 3,","\\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics computation for the COMAN model\\n') out_file.write('","frame of the current body i # Omi : rotational vector from the","count # print the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line):","sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6',","Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max","= parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] *","3, 4, 5, # right leg 0, 7, 8, 9, 10, 11, #","(!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso","else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first:","axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]])","== 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count","'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\n')","-1, -1, -1 # return true if it has a shape 'x%a_%b' (indexes","current body i\\n') out_file.write(' * Omi : rotational vector from the previous body","G_i of body i # xpi : derivative of xi # xgpi :","and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id,","velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' //","0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j])))","max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\n') else: out_file.write(', ') # variables","sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8]","]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id',","nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i]","in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first =","'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\n // anchor point","return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis ==","range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j]","out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n')","body i\\n') out_file.write(' * ci : cosine of the relative angle before body","out_file.write(' *\\n') out_file.write(' * computation of:\\n') out_file.write(' * COM (center of mass) position","range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace","out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable","elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i]","in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1,","== nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write(' // global com absolute position\\n')","parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close() # output","body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3,","Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r =","feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08,","i # Ri : absolute rotational matrix # Rti : transpose matrix of","= sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist)","range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right foot absolute velocity\\n')","com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on","j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j]","* // 02 08 //\\n') out_file.write(' * // 03 09 //\\n') out_file.write(' *","class\\n') out_file.write(' *\\n') out_file.write(' * computation of:\\n') out_file.write(' * COM (center of mass)","previous body to the current body i \\n') out_file.write(' * (previous body is","}\\n') # xg & xgp out_file.write('\\n // com absolute positions and velocities\\n') xg","}\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n')","sp.zeros(3, 1) save_vector = 3 * [None] for i in range(0,3): if vector[i]","j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac !=","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n')","Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj,","1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x","jacobian com absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","//\\n') out_file.write(' * // 02 08 //\\n') out_file.write(' * // 03 09 //\\n')","# results out_file.write('\\n // -- Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M,","out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com absolute velocity\\n') for i in range(0, 3):","xji : jacobian of 'xi' # xgji : jacobian of 'xgi' # Rji","0: out_write.write(';\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2],","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute position\\n') for i","y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot","if neg_flag: result = -result return result # compute the derivative of an","= symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:","= np.array([ -1, # waist 0, 1, 2, 3, 4, 5, # right","the vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj,","also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True,","= 0 for i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print","to its COM (center of mass) G_i, # expressed in the relative frame","out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_bodies): for","'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg,","matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print =","|| (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write('","a string for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if","cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n')","rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on","in range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i]","0, 13, 14, # trunk 15, 16, 17, 18, # right arm 15,","* xgpi : derivative of xgi\\n') out_file.write(' * omi : absolute rotational vector","and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print","range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j))","symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of","isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag: result +=","# compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write('","'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print =","= sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'),","# 20 01 24 # # 02 08 # # 03 09 #","== 2 and j == 2: out_file.write(';\\n') else: out_file.write(', ') # print variables","in the relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write(' * Rdi","* // 05 11 //\\n') out_file.write(' * // 06 12 //\\n') out_file.write(' *","= 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({},","sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] =","}\\n\\n') out_file.write(' // wrists absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}]","IMU - rotation matrices\\n') for i in range(0, 3): for j in range(0,","angles (position and derivative) of the waist and the torso def torso_waist_angles(out_file, R,","& xgp out_file.write('\\n // com absolute positions and velocities\\n') xg = nb_bodies*[None] xgp","xgi : absolute position vector of the COM G_i of body i #","= 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg & xgp out_file.write('\\n // com","on all the lines for line in f: cut_line = line.split(' = ')","joint absolute velocities\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0]","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","the center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg,","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n","copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script","rotation matrices out_file.write('\\n // jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count == 0:","sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'),",": cosine of the relative angle before body i\\n') out_file.write(' *\\n') out_file.write(' *","+ om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1)","detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len","'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i]","[0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis ==","out_file.write(' * waist and torso orientaion angles and derivatives\\n') out_file.write(' *\\n') out_file.write(' *","in range (1, nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac","def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out,","out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i ==","1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0],","out_file.write(' * xgpi : derivative of xgi\\n') out_file.write(' * omi : absolute rotational","inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] =","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot contact points absolute position\\n') for i","out_file.write(' * // //\\n') out_file.write(' * // 17 16 21 //\\n') out_file.write(' *","nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n // -- symbolic computation --","out_file.write('\\n') out_file.write(' // right wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","true if it is a float def isInt(value): try: int(value) return True except:","pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result","i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' +","out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot orientation","result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return result","position vector of the COM G_i of body i # xpi : derivative","cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) #","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') # get a string for the enumeration","leg 0, 7, 8, 9, 10, 11, # left leg 0, 13, 14,","xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file,","sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4]","i \\n') out_file.write(' * (previous body is not always body i-1), expressed in","der_var) # one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj,","sine of the relative angle before body i\\n') out_file.write(' * ci : cosine","in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j])))","= y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i","= symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:",": cosine of the relative angle before body i # # xi :","file os.remove(in_temp) # main script # rotation axis for each joint before body",", sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6]","out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} +","'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id'","// left foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n')","range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if","mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write('","R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives","write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in","= om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i]","// right foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_contacts): for j","len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6:","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'),","{}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\n') count =","// 07 13 //\\n') out_file.write(' * // //\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n')","j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print,","* Di : position vector from the anchor point of the previous body","line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('['))","the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if","pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c]","//\\n') out_file.write(' * // //\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin:","in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] =","relative angle before body i # # xi : absolute position vector (from","point of the previous body to the current body i \\n') out_file.write(' *","all the lines for line in f: cut_line = line.split(elem) if len(cut_line) ==","(previous body is not always body i-1), expressed in the relative\\n') out_file.write(' *","in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j))","False, -1, -1 # count the number of 'elem' in the file def","to its COM (center of mass) G_i,\\n') out_file.write(' * expressed in the relative","xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id]","symbolic computation -- //\\n') # Rj, xj, xgj and xgj (jacobian) Rj =","= vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {}","= None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} =","Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm","# (previous body is not always body i-1), expressed in the relative #","# 18 15 22 # # 19 14 23 # # 20 01","of mass) G_i, # expressed in the relative frame of the current body","== 2: count += 1 return count # print the declaration of an","x_l[i])) out_file.write('\\n') out_file.write(' // left wrist absolute velocity\\n') for i in range(0,3): out_file.write('","23 # # 20 01 24 # # 02 08 # # 03","in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot","* Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i","all the lines for line in f: cut_line = line.split(' = ') if","9, 10, 11, # left leg 0, 13, 14, # trunk 15, 16,","elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0,","point of body i\\n') out_file.write(' * xgi : absolute position vector of the","j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1:","= 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n')","R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","count != 0: out_write.write(';\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2],","0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if","= x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id]","flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j],","range(0,3): for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1:",", 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics'","= get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb +","{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0],","# # Di : position vector from the anchor point of the previous","cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n')","i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name)","* xpi : derivative of xi\\n') out_file.write(' * xgpi : derivative of xgi\\n')","# parent index parent_body_index = np.array([ -1, # waist 0, 1, 2, 3,","= atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] =","body i\\n') out_file.write(' * xpi : derivative of xi\\n') out_file.write(' * xgpi :","i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i]","sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) #","expressed in the relative # frame of the previous body # Rdi :","x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id]","count != 0: out_write.write(';\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count","9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in range(0, 9): out_file.write('","-- variables initialization -- //\\n') out_file.write('\\n // IMU - rotation matrices\\n') for i","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet absolute orientation\\n') for i in range(0, 9):","i != nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') # from an orientation matrix, compute","return np.array([]) # compute the derivative of an element (for jacobian) def der_elem(elem_str,","out_file.write(' // waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write('","{};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative)","all terms for cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len","Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj,","cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product","out_file.write(' // global com absolute velocity\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}]","atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0],","inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file,","sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9')","9): if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print =","point absolute positions and velocities\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] =","# epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0]","return False # return true if it has a shape 'R%a_%b%c' (indexes %a,","c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation angle","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n')","r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y","= R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1)))","# Ri : absolute rotational matrix # Rti : transpose matrix of Ri","0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0,","for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if","') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count == 0:","20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor point","the relative # frame of the previous body # DGi : position vector","= '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] =",": derivative of xgi\\n') out_file.write(' * omi : absolute rotational vector of body","open(in_temp, 'r') as f: # loop on all the lines for line in","derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon =","def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for","x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l","= ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix #","xpi : derivative of xi # xgpi : derivative of xgi # omi","leg 2, 1, 3, 2, 1, 2, # left leg 1, 2, 3,","= print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def","range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the vector","def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1: return np.array([[1.0,","[xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names,","of the current body i # Omi : rotational vector from the previous","'xi' # xgji : jacobian of 'xgi' # Rji : jacobian of 'Ri'","vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result =","joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) #","= sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts","i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0]","out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\n') else: out_file.write(', ') # variables initialization","and derivative) of the waist and the torso def torso_waist_angles(out_file, R, om, waist_id,","# copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main","of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): #","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') #","# # # ###################### # # origin: in the waist, middle point between","xgi # omi : absolute rotational vector of body i # Ri :","expressed in the inertial frame) # of the anchor point of body i","but aligned with the ground (info from IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di","out_file.write(' // waist orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n')","orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the wrists","r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id]","range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print","* (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist)","om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0,","omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part,","in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] =","the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp,","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] =","G_i, # expressed in the relative frame of the current body i #","sine): if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine,","[0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0,","str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i !=","1) flag_print = 0 for i in range(0,3): if vector[i] == 0 or","= {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix # save","write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i]","of the relative angle before body i\\n') out_file.write(' * ci : cosine of","declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id]","get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T","if term_list[0] == '': term_list.pop(0) result = 0 # loop on all terms","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6,","nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0])","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,","of the COM G_i of body i # xpi : derivative of xi","right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') ,","sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] =","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file,","axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split)","sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] =","1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist =","xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first =","jacobian of 'xgi' # Rji : jacobian of 'Ri' # return true if","False # return true if it has a shape 'R%a_%b%c' (indexes %a, %b,","Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for","in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right","in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist jacobian\\n')","out_file.write(' }\\n\\n') out_file.write(' // left foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","new_string = cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for i","# jacobian xg out_file.write('\\n // jacobian com absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n')","sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) #","computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the","{};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot contact points jacobian\\n') out_file.write(' if","= elem_split[0] elif cur_len == 2: # negative neg_flag = 1 pos_str =","// left wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i]","xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1","= 0 for k in range(0, 3): if xgj_print[i][j][k] != None: if not","9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist absolute orientation","= cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist =","xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1 elif","body i # Omi : rotational vector from the previous body to the","out_file.write('\\n // jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist jacobian\\n')","from the anchor point of the previous body to the current body i","out_file.write(' * \\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics","== nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com absolute","body # DGi : position vector from the anchor point of body i","= nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1)","sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')])","{\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso =","R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3,","write the end of the file def write_file_end(out_file): out_file.write('}\\n') # print matrix components","return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative of an","= r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb","cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and","3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis == 1:","= 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","IMU - angles velocity\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1,","nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in","'.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' +","= sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a rotation matrix","+ om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1)","point of body i to its COM (center of mass) G_i, # expressed","jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range (1,","sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] =","trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] =","computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T *","-sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else:","[Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor","rotational vector of body i # Ri : absolute rotational matrix # Rti","05 11 //\\n') out_file.write(' * // 06 12 //\\n') out_file.write(' * // 07","of the relative angle before body i\\n') out_file.write(' *\\n') out_file.write(' * xi :","orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist =","c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso","position vector (from origin, expressed in the inertial frame) # of the anchor","product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) #","an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to","has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try:","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute orientation jacobian\\n')","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n //","r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb","xp_l[i])) out_file.write('\\n') out_file.write(' // left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3,","sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')])","j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' //","= filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string","if i != nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') # from an orientation matrix,","[-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] =","main script # rotation axis for each joint before body i (1:x, 2:y,","') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n') else:","out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write('","0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine,","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,","elem): count = 0; with open(in_file, 'r') as f: # loop on all","+ {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1],","x_l[i])) out_file.write('\\n') out_file.write(' // left foot absolute velocity\\n') for i in range(0,3): out_file.write('","0, 7, 8, 9, 10, 11, # left leg 0, 13, 14, #","body i # Ri : absolute rotational matrix # Rti : transpose matrix","cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n')","out_file.write(' * DGi : position vector from the anchor point of body i","# compute derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q): result","sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0]","out_file.write(' // torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute","# write the end of the file def write_file_end(out_file): out_file.write('}\\n') # print matrix","xi # xgpi : derivative of xgi # omi : absolute rotational vector","for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i","= sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0,","== 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj,","write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' *","other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str)","parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id]","sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j]","'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id'","= 1 pos_str = elem_split[1] else: print('Error: {} instead of 1 or 2","+= der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip())","Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print,","}\\n\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id],","//\\n') out_file.write(' * // 17 16 21 //\\n') out_file.write(' * // 18 15","\\n') out_file.write(' * (previous body is not always body i-1), expressed in the","Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0,","nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com absolute velocity\\n')","= sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29')","out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left foot jacobian\\n') out_file.write(' if","in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write('","') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization","out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin: in the waist, middle point","range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write('","2 # left arm ]) # parent index parent_body_index = np.array([ -1, #","symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first","* (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] =","+= 1 return count # print the declaration of an element def print_declaration_elem(in_file,","len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the","print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt,","xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb +",", 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16',","i in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print = 0","!= 0: out_write.write(';\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count =","= ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n')","if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first =","= nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0,","out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write('","3) flag_print = 0 for i in range(0,3): for j in range(0,3): if","1) save_vector = 3 * [None] for i in range(0,3): if vector[i] ==","xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None] # rotation matrices","3) R[0] = sp.zeros(3, 3) for i in range(0, 3): for j in","return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([])","'{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if","= matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j]","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg & xgp",": derivative of xi # xgpi : derivative of xgi # omi :","Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i]","R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices","= nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0]","flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com jacobian\\n') out_file.write(' if (flag_jacob)\\n","rotational vector from the previous body to the current body i # (previous","om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '')","# cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str","xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') #","s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n')","element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q =","inertial frame)\\n') out_file.write(' * of the anchor point of body i\\n') out_file.write(' *","write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i","6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj,","out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot contact","x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] *","= x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot =","the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem))","out_file.write(' }\\n\\n') out_file.write(' // feet absolute orientation\\n') for i in range(0, 9): out_file.write('","om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id =","xji : jacobian of \\'xi\\'\\n') out_file.write(' * xgji : jacobian of \\'xgi\\'\\n') out_file.write('","= {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left wrist absolute velocity\\n') for i in","= nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1):","out_file.write(' * expressed in the relative frame of the current body i\\n') out_file.write('","','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-')","out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double","one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip())","for j in range(1, nb_bodies): flag_print = 0 for k in range(0, 9):","= 0 if count != 0: out_write.write(';\\n') # print all declarations def print_all_declaration(in_file,","[-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine,","nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n')","Rj, xj, xgj, xgj_print, x_vector, index): # loop on all the joints for","com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om,","0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return","xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot","for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i","sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'),","IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di : position vector from the anchor point","Rj_print, R_matrix, index): # loop on all the joints for i in range","of mass) position and velocity\\n') out_file.write(' * feet position, velocity and orientation\\n') out_file.write('","{};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i]))","cur_len == 1: # positive neg_flag = 0 pos_str = elem_split[0] elif cur_len","xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot","for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on","out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center of mass position and velocity def","Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\n') for i in","joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min,","') # print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ')","the anchor point of body i # xgi : absolute position vector of","of Ri # xji : jacobian of 'xi' # xgji : jacobian of","result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} *","sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] =","declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id]","flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n // --","1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3","for i in range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i])))","write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for j in range(0,3):","out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] =","waist and torso orientaion angles and derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write('","nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first","'{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] =","* Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot +","write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg","positions and velocities\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0,","exit() return result # write the beginning of the file def write_file_beginning(out_file, joint_id_names):","[pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] =","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] =","feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max,","3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU - angles velocity\\n')","-sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0],","if epsilon > 0: # epsilon = 1 -> pitch angle in [-pi/2","print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point def write_symb_xgj(nb_bodies,","write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print,","for j in range (1, nb_bodies): flag_print = 0 for k in range(0,","'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg","+ R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot)","cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i in","# symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id]","sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a,","c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position","sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None]","out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right foot absolute velocity\\n') for","'_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj,","= elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split","absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write('","3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n","om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i","= sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14]","out_file.write(' }\\n') # results out_file.write('\\n // -- Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies,","matrix\\n') out_file.write(' * Rti : transpose matrix of Ri\\n') out_file.write(' * xji :","position\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' //","nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] +","# write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write('","{') count = 0 for i in range(1, len(joint_id_names)): count += 1 if","}\\n\\n') # from an orientation matrix, compute the roll, pitch, yaw angles (and","out_file.write(';\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' //","x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max","= sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') ,","= R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16')","= x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] =","Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot orientation matrix as","if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis ==",": absolute position vector (from origin, expressed in the inertial frame) # of","of the COM G_i of body i\\n') out_file.write(' * xpi : derivative of","the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix",", 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies,","torso orientaion angles and derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' * //","return new_vector # write symbolic matrix and replace symbolic variable by its name","its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print =","= x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot","point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0,","# write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj,","computation of:\\n') out_file.write(' * COM (center of mass) position and velocity\\n') out_file.write(' *","sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')])","if cur_len == 1: # positive neg_flag = 0 pos_str = elem_split[0] elif","+ om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T","save_vector = 3 * [None] for i in range(0,3): if vector[i] == 0","Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0, 3):","om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file,","= '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string","print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 *","rotational vector from the previous body to the current body i \\n') out_file.write('","cut_line = line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1:","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega","as f: # loop on all the lines for line in f: #","xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian","'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id'","// left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation","R, om, 0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies,","right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0,","out_file.write(' * COM (center of mass) position and velocity\\n') out_file.write(' * feet position,","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","* si : sine of the relative angle before body i\\n') out_file.write(' *","*/\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center of mass position and velocity","in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2]","M, xg, xgp, xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies):","out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists absolute","sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10]","nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3,","IMU) # # Di : position vector from the anchor point of the","{};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0, 3): for j","pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{}","COM (center of mass) G_i,\\n') out_file.write(' * expressed in the relative frame of","(previous body is not always body i-1), expressed in the relative # frame","# output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: #","xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =","= '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j]","in range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first = 1","in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print = 0 for","on all the lines for line in f: # declaration if len(line.split('// --","at the origin (waist), but aligned with the ground (info from IMU) #","= om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist","'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies,","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] =","left foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\n') else: out_file.write(', ') #","# right arm 2, 1, 3, 2 # left arm ]) # parent","1, 2, 3, # trunk 2, 1, 3, 2, # right arm 2,","range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in range(0, 9):","out_file.write(' * computation of:\\n') out_file.write(' * COM (center of mass) position and velocity\\n')","out_file.write(' // -- variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy","# get a string for the enumeration of joints def get_string_enum(cur_string): cur_split =","flag_print: out_file.write('\\n') return new_matrix # save the symbolic vector for print def print_save_symb_vector(vector,","i # xgi : absolute position vector of the COM G_i of body","for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0","out_file.write(' * // 20 01 24 //\\n') out_file.write(' * // 02 08 //\\n')","on all the matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj,","\\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward","frame)\\n') out_file.write(' * of the anchor point of body i\\n') out_file.write(' * xgi","start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i in range(0,3):","xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] =","anchor point of body i # xgi : absolute position vector of the","trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] =","on all the joints for i in range (1, nb_bodies): new_vector = sp.zeros(3,","[0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0,","= 0 for i in range (1, nb_bodies): flag_print = 0 for j","pos_str = elem_split[1] else: print('Error: {} instead of 1 or 2 in negative","R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso',","sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] =","= symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first:","Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x,","M) # end of the file write_file_end(file_temp) file_temp.close() # output file out_file =","torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet","Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj,","nb_bodies): flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj,","of the relative angle before body i # # xi : absolute position","out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write('","+ R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r =","1, 3, 2, 1, 2, # left leg 1, 2, 3, # trunk",": absolute rotational vector of body i\\n') out_file.write(' * Ri : absolute rotational","name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0","the origin (waist), but aligned with the ground (info from IMU)\\n') out_file.write(' *\\n')","10, 11, # left leg 0, 13, 14, # trunk 15, 16, 17,",", 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18',","= write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i","nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj,","frame of the previous body # DGi : position vector from the anchor","x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb","for i in range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1,","sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] =","') out_file.write(' // global com absolute position\\n') for i in range(0, 3): out_file.write('","# rotation matrices out_file.write('\\n // rotation matrices\\n') R = nb_bodies*[None] Rt = nb_bodies*[None]","-1 # return true if it has a shape 'x%a_%b' (indexes %a, %b","new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for i in range(0,3):","// IMU - angles velocity\\n') for i in range(0, 3): out_file.write(' omega_{} =",">= nb_max_line: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') # print","out_file.write(' // right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","out_file.write(' *\\n') out_file.write(' * origin: in the waist, middle point between the two","om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot","Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T #","Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg","vector axis def get_vector_axis(axis, direct, elem): if direct: if axis == 1: return","else: return np.array([]) # compute the derivative of an element (for jacobian) def","computation for the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {') count","flag_first = 0 for i in range(0, nb_bodies): for j in range(1, nb_bodies):","get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet absolute orientation\\n') for i in","variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1)","cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str ==","Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x &","out_file.write('\\n // -- Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp,","{}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0]))","}\\n\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n')","leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' ,","return True except: return False # return true if it has a shape","k in range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first =","in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont =","= sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M =","filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string =","axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the","R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1,","x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute","except: return False, -1, -1 # count the number of 'elem' in the","// 03 09 //\\n') out_file.write(' * // 04 10 //\\n') out_file.write(' * //","for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint","xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write","the lines for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' =","return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine,","# loop on all the lines for line in f: cut_line = line.split(elem)","the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3)","def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all the","joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for i in range(0,","// right foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n')","point of body i # xgi : absolute position vector of the COM","out_file.write(' * \\\\brief forward kinematics computation for the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('//","out_file.write(' // global com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for i in range(1,","i\\n') out_file.write(' * xgi : absolute position vector of the COM G_i of","om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T *","xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) #","and j == 2: out_file.write(';\\n') else: out_file.write(', ') # print variables declaration def","end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return","'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13',","absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write('","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i],","positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive","= nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i]","{}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0]))","elem): if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis","* Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x","sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) #","body i # xpi : derivative of xi # xgpi : derivative of","arm ]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id',","is a float def isInt(value): try: int(value) return True except: return False #","right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id'","sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix,","(and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon","R_matrix[0])) else: # epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2]","0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1))","i in range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] =","= sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] =","'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24',","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write('","initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\n') out_file.write('\\n","result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) #","new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string))","cur_len = len(cur_term_split) # no product if cur_len == 1: result += der_elem(cur_term_split[0],","index parent_body_index = np.array([ -1, # waist 0, 1, 2, 3, 4, 5,","= nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1)","i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split","np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) #","for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left","arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M)","= 3 * [None] # rotation matrices out_file.write('\\n // rotation matrices\\n') R =","]) # parent index parent_body_index = np.array([ -1, # waist 0, 1, 2,","= om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist =","M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,","1, 2, # right leg 2, 1, 3, 2, 1, 2, # left","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute position\\n') for","Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg","# left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0])","it has a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try:","joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2,","xpi : derivative of xi\\n') out_file.write(' * xgpi : derivative of xgi\\n') out_file.write('","but aligned with the ground (info from IMU) # # Di : position","= sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'),","body i\\n') out_file.write(' * Ri : absolute rotational matrix\\n') out_file.write(' * Rti :","-1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name,","= inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1]","int(value.split('_')[1]) return True, a, b except: return False, -1, -1 # count the","orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot =","Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') #","# vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result","sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11]","range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right foot jacobian\\n') out_file.write('","out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r')","out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write('","if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 /","double ') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1))","absolute positions and velocities\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0]","new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector # write symbolic matrix and","lines for line in f: cut_line = line.split(' = ') if len(cut_line) ==","line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) ==","# detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1:","out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right foot jacobian\\n') out_file.write(' if","out_file.write(' // left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","at the origin (waist), but aligned with the ground (info from IMU)\\n') out_file.write('","flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k] != None: if","body # Rdi : rotational matrix between body i and its predecessor #","= sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21]","out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet","of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj):","in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i]","'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\n // jacobian rotation","3): if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print =","isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag:","in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0:","mass) G_i,\\n') out_file.write(' * expressed in the relative frame of the current body","0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if","it has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value):","in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n') for i in range(1, nb_bodies): out_file.write(' c{}","[v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct,","for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1)","sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n","Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'),","# si : sine of the relative angle before body i # ci","= 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot contact points","sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n')","//\\n') out_file.write(' * // 05 11 //\\n') out_file.write(' * // 06 12 //\\n')","vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i in","shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try: a =","1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] =","-- symbolic computation -- //\\n') # Rj, xj, xgj and xgj (jacobian) Rj","np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2:","terms for cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len =","rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0,","sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 ,","ci : cosine of the relative angle before body i # # xi","1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists","'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm","xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i]","out_file.write(' // left foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1)","xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] =","= 0 for k in range(0, 3): if xj_print[i][j][k] != None: if not","sp.zeros(3, 3) # loop on all the matrix elements for j in range(0,","# left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt,","foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the","count = 0 for i in range(1, len(joint_id_names)): count += 1 if i","Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0, 3):","###################### # # # 17 16 21 # # 18 15 22 #","body i to its COM (center of mass) G_i,\\n') out_file.write(' * expressed in","the inertial frame)\\n') out_file.write(' * of the anchor point of body i\\n') out_file.write('","om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3,","position\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0","x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] *","# xgi : absolute position vector of the COM G_i of body i","= x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb =","}\\n\\n') # get a string for the enumeration of joints def get_string_enum(cur_string): cur_split","x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min","jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index):","om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] =","file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f:","xj, xgj, xj_print, x_vector, index): # loop on all the joints for i","in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left foot absolute","2, 1, 3, 2, # right arm 2, 1, 3, 2 # left","and outputs class\\n') out_file.write(' *\\n') out_file.write(' * computation of:\\n') out_file.write(' * COM (center","1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if","xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id]","(R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\n') for","// com absolute positions and velocities\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for","e] = isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector","count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if","= cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len == 1: result","out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist jacobian\\n') out_file.write(' if","* ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin: in the waist, middle point between","nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] =","# symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id]","'Lfoot') out_file.write('\\n') # compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names,","= 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot =","# loop on all the lines for line in f: cut_line = line.split('","0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] =","i in range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1))","out_file.write(' * // 19 14 23 //\\n') out_file.write(' * // 20 01 24","def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon = 1","str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 # loop on","if flag_print: out_file.write('\\n') return new_vector # write symbolic matrix and replace symbolic variable","a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return False,","0 # loop on all terms for cur_term in term_list: # detect products","or vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name =","out_file.write('\\n // jacobian anchor point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","max: out_file.write(';\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write('","j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac !=","* // 18 15 22 //\\n') out_file.write(' * // 19 14 23 //\\n')","18 15 22 # # 19 14 23 # # 20 01 24","ground (info from IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di : position vector from","out_file.write(' * omi : absolute rotational vector of body i\\n') out_file.write(' * Ri","angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write('","x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot =","(1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the matrix elements","-v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] =","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}]","rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n // -- symbolic computation -- //\\n') #","out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot","# write symbolic vector and replace symbolic variable by its name def write_symb_vector(out_file,","-- //\\n') # Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj","i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left foot","\\'xgi\\'\\n') out_file.write(' * Rji : jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n')","03 09 //\\n') out_file.write(' * // 04 10 //\\n') out_file.write(' * // 05","string for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split)","!= None: if not flag_first: flag_first = 1 flag_print = 1 elif not","i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] =","1, 3, 2, 1, 2, # right leg 2, 1, 3, 2, 1,","2, 1, 2, # left leg 1, 2, 3, # trunk 2, 1,","int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except: return False, -1,","= x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] =","vector (from origin, expressed in the inertial frame) # of the anchor point","foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n')","out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n // joint absolute velocities\\n') Om = nb_bodies*[None]","for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None]","= R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x","x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id]","range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0:","origin, expressed in the inertial frame)\\n') out_file.write(' * of the anchor point of","= str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 # loop","out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part,","COM G_i of body i # xpi : derivative of xi # xgpi","jacobian of \\'xgi\\'\\n') out_file.write(' * Rji : jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void","relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write(' * DGi : position","vector (from origin, expressed in the inertial frame)\\n') out_file.write(' * of the anchor","[sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis","= sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for","range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist jacobian\\n') out_file.write('","i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\n') else: out_file.write(',","// joint absolute velocities\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None]","4, 5, # right leg 0, 7, 8, 9, 10, 11, # left","0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0,","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}]","R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): #","in the file def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as","= symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) #","range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id]","- rotation matrices\\n') for i in range(0, 3): for j in range(0, 3):","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","term_list.pop(0) result = 0 # loop on all terms for cur_term in term_list:","velocity\\n') out_file.write(' * feet position, velocity and orientation\\n') out_file.write(' * waist and torso","xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n // jacobian anchor","100) out_file.write('\\n\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp)","i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] *","nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies,","{\\n') for i in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}]","os.remove(in_temp) # main script # rotation axis for each joint before body i","trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' #","j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont","# variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis,","declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >=","// left foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') #","1: # positive neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2:","sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n","# write symbolic matrix and replace symbolic variable by its name def write_symb_matrix(out_file,","== 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif","out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} +","in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1","0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'),","right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint names","+ Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i])","+ s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out,","get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get","xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om,","for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left","* computation of:\\n') out_file.write(' * COM (center of mass) position and velocity\\n') out_file.write('","COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) #","'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ])","sp import re import os ###################### # # # 17 16 21 #","file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on","j+1)) if i == 2 and j == 2: out_file.write(';\\n') else: out_file.write(', ')","of xi\\n') out_file.write(' * xgpi : derivative of xgi\\n') out_file.write(' * omi :","/ c_y_torso;\\n\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist')","out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot","the lines for line in f: # declaration if len(line.split('// -- variables initialization","3 * [None] # rotation matrices out_file.write('\\n // rotation matrices\\n') R = nb_bodies*[None]","variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in","write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on all the joints","orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for","j in range (1, nb_bodies): flag_print = 0 for k in range(0, 3):","(R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) #","1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name)","in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp,","as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write('","= cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for i in","[xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com","else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\n') count =","absolute position\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first =","the origin (waist), but aligned with the ground (info from IMU) # #","# loop on all the lines for line in f: cut_line_1 = line.split(elem)","2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else:","out_file.write(' * // 02 08 //\\n') out_file.write(' * // 03 09 //\\n') out_file.write('","the ground (info from IMU) # # Di : position vector from the","omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0]))","in range(0,3): for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] ==","absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write('","-1 # count the number of 'elem' in the file def count_elem(in_file, elem):","# no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj,","joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names,","= sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for print def","waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg","1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name,","= xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot =","1 if count >= nb_max_line: out_write.write(';\\n') count = 0 if count != 0:","sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine,",", 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , #","for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac","7, 8, 9, 10, 11, # left leg 0, 13, 14, # trunk","len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double","c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot","3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies):","R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i]","14, # trunk 15, 16, 17, 18, # right arm 15, 20, 21,","waist, middle point between the two pitch hip rotations # inertial frame: located","def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp","// left foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","+ {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j ==","* xi : absolute position vector (from origin, expressed in the inertial frame)\\n')","22 //\\n') out_file.write(' * // 19 14 23 //\\n') out_file.write(' * // 20","// right wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","out_file.write(' * inertial frame: located at the origin (waist), but aligned with the","the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M):","3, 2, 1, 2, # left leg 1, 2, 3, # trunk 2,","before body i\\n') out_file.write(' *\\n') out_file.write(' * xi : absolute position vector (from",", sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'),","its predecessor\\n') out_file.write(' * si : sine of the relative angle before body","!= 0: out_write.write(';\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]],","declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for j","i-1), expressed in the relative # frame of the previous body # Rdi","*\\n') out_file.write(' * computation of:\\n') out_file.write(' * COM (center of mass) position and","3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{}","om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist',","= open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on all the","range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector,","Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14]","out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17]","s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso","0: if not flag_first: flag_first = 1 flag_print = 1 elif not flag_print:","in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac !=","end_name) save_vector[i] = ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector,","{};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right wrist absolute velocity\\n') for i in range(0,3):","in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch","-0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output","of body i to its COM (center of mass) G_i,\\n') out_file.write(' * expressed","center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp,","(center of mass) G_i, # expressed in the relative frame of the current","out_file.write(' // right foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n')","return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation","out_file.write(' * feet position, velocity and orientation\\n') out_file.write(' * waist and torso orientaion","loop on all terms for cur_term in term_list: # detect products cur_term_split =","for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i]","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot orientation matrix as angles","absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write('","sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19]","np import sympy as sp import re import os ###################### # # #","len(cut_line) == 2: count += 1 return count # print the declaration of","(waist), but aligned with the ground (info from IMU)\\n') out_file.write(' *\\n') out_file.write(' *","1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] =","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg &","the COM G_i of body i # xpi : derivative of xi #","compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0]","derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len =","in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split =","3*i+j)) out_file.write('\\n // IMU - angles velocity\\n') for i in range(0, 3): out_file.write('","// jacobian anchor point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","com absolute positions and velocities\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i","0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] =","count = 0 if count != 0: out_write.write(';\\n') # print all declarations def","line in f: cut_line = line.split(' = ') if len(cut_line) == 2: if","13, 14, # trunk 15, 16, 17, 18, # right arm 15, 20,","out_file.write(' * of the anchor point of body i\\n') out_file.write(' * xgi :","out_file.write('\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id],","to the current body i \\n') out_file.write(' * (previous body is not always","= sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0,","range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None,","= sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0]","# get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis","out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n') for i in range(1,","xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic",">= 1: count = 0 with open(in_file,'r') as f: # loop on all","file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp,","'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15',","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","+= 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count =","Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2]","//\\n') out_file.write(' * // 18 15 22 //\\n') out_file.write(' * // 19 14","16, 17, 18, # right arm 15, 20, 21, 22 # left arm","pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon >","* Ri : absolute rotational matrix\\n') out_file.write(' * Rti : transpose matrix of","s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part,","= sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0,","sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9","sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0,","i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '')","a, b except: return False, -1, -1 # count the number of 'elem'","Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i","sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] =","= Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None] #","the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var):","# left arm ]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id',","body i # # xi : absolute position vector (from origin, expressed in","i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector","# frame of the previous body # DGi : position vector from the","0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'),","# compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj,","om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] =","out_file.write(' // left wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] =","-0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,","xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06,","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n')","j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]]","R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i],","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","= cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n') for i in range(1, nb_bodies): out_file.write('","xgj, der_var) # one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj,","Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions","left wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i]))","= symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first:","write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print,","flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n')","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac))","out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write('","# negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead of","out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {') count = 0 for i in","','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 # loop on all","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute position\\n') for","xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1","cosines\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n //","name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0","(c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} -","+ Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i])","sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8' ,","transpose matrix of Ri # xji : jacobian of 'xi' # xgji :","+ Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i])","/ c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write(' // waist orientation angle","b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except: return","= sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'),","0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0,","out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write('","the previous body\\n') out_file.write(' * DGi : position vector from the anchor point","script # rotation axis for each joint before body i (1:x, 2:y, 3:z)","Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation","left foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute","'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' ,","i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j","anchor point absolute positions and velocities\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0]","leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] =","def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms term_list =","out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of","write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print,","# right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19',","positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_bodies):","or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name =","// feet absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i,","j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j ==","flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j],","angle before body i # # xi : absolute position vector (from origin,","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] =","0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'),","12 //\\n') out_file.write(' * // 07 13 //\\n') out_file.write(' * // //\\n') out_file.write('","end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for","== '': term_list.pop(0) result = 0 # loop on all terms for cur_term","out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write('","= sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0,","absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write('","int(value) return True except: return False # return true if it has a","the COM G_i of body i\\n') out_file.write(' * xpi : derivative of xi\\n')","os ###################### # # # 17 16 21 # # 18 15 22","index): # loop on all the joints for i in range (1, nb_bodies):","15, 16, 17, 18, # right arm 15, 20, 21, 22 # left","relative # frame of the previous body # DGi : position vector from","out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left wrist absolute velocity\\n') for","waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id'","[0.0], [-elem]]) else: return np.array([]) # compute the derivative of an element (for","compute the derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj,","in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1","<NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics computation for the","out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n')","of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic","matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] =","return True, a, b, c except: return False, -1, -1, -1 # return","out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute","atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle in","// 06 12 //\\n') out_file.write(' * // 07 13 //\\n') out_file.write(' * //","out_file.write(' *\\n') out_file.write(' * xi : absolute position vector (from origin, expressed in","sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] =","24 //\\n') out_file.write(' * // 02 08 //\\n') out_file.write(' * // 03 09","jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_contacts):","else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com absolute velocity\\n') for i in range(0,","Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')])","out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left foot absolute velocity\\n') for","nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first =","out_file.write(' // right foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None","cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len == 1:","0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine,","out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i","l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot","if it has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def","0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) #","== 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem],","0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name","in f: # declaration if len(line.split('// -- variables initialization -- //')) != 1:","# positive neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2: #","2: # negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead","* Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0],","absolute position vector (from origin, expressed in the inertial frame)\\n') out_file.write(' * of","= xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts):","re import os ###################### # # # 17 16 21 # # 18","// right wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i,",": sine of the relative angle before body i\\n') out_file.write(' * ci :","start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i in range(0,3):","'') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\n // anchor point absolute","Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1),","results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R,","R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of",": jacobian of \\'xgi\\'\\n') out_file.write(' * Rji : jacobian of \\'Ri\\'\\n') out_file.write(' */\\n')","x_vector, index): # loop on all the joints for i in range (1,","20 01 24 # # 02 08 # # 03 09 # #","R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i","and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id,","Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')])","om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist =","# 02 08 # # 03 09 # # 04 10 # #","1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right","compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp,","3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj,","out_file.write('\\n // anchor point absolute positions and velocities\\n') x = nb_bodies*[None] xp =","np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis == 1: return np.array([[-elem],","elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0,","05 11 # # 06 12 # # 07 13 # # #","true if it has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned)","on all terms for cur_term in term_list: # detect products cur_term_split = cur_term.split('*')","symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot","of the current body i\\n') out_file.write(' * Omi : rotational vector from the","Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15)","range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] +","= atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle","angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity","out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_contacts): for","xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k] !=","2 and j == 2: out_file.write(';\\n') else: out_file.write(', ') # print variables declaration","12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,","// torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the","out_file.write(' // left foot contact points absolute position\\n') for i in range(0, nb_contacts):","Dpt, Dg, M): out_file.write('\\n\\n // -- symbolic computation -- //\\n') # Rj, xj,","sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] =","of xi # xgpi : derivative of xgi # omi : absolute rotational","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] =",", 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14',","* (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\n')","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] =","= 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i]))","= sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16]","# return true if it has a shape 'R%a_%b%c' (indexes %a, %b, %c","elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative","{};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n","prefix, min, max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i))","0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine,","0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left","out_file.close() # remove temporary file os.remove(in_temp) # main script # rotation axis for","out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation angle derivatives","np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative of an element","no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)","sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] =","0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1))","= nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print =","leg 1, 2, 3, # trunk 2, 1, 3, 2, # right arm","xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id]","direct, cosine, sine): if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0],","2: count += 1 return count # print the declaration of an element","= {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot contact points jacobian\\n') out_file.write('","out_file.write(' * frame of the previous body\\n') out_file.write(' * DGi : position vector","for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id]","x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb","= cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot =","= sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0,","# origin: in the waist, middle point between the two pitch hip rotations","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,",", 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' ,","the waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' //","for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac","xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n // jacobian com absolute positions\\n') out_file.write('","out_file.write(' * // 07 13 //\\n') out_file.write(' * // //\\n') out_file.write(' * ////////////////////////\\n')","theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the wrists position, velocity and orientation","in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end","om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7]","if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1","Rti : transpose matrix of Ri\\n') out_file.write(' * xji : jacobian of \\'xi\\'\\n')","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i,","file def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f: #","range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU - angles","matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso orientation","symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first","of 'xi' # xgji : jacobian of 'xgi' # Rji : jacobian of","18 15 22 //\\n') out_file.write(' * // 19 14 23 //\\n') out_file.write(' *","the previous body to the current body i # (previous body is not","= x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot =","Rj, xj, xgj, xj_print, x_vector, index): # loop on all the joints for","global com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for i in range(1, nb_bodies): for","nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for k in range(0,","out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left","out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute","list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0)","out_file, 100) out_file.write('\\n\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary file","* (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2]))","loop on all the lines for line in f: cut_line = line.split(' =","xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i],","1, 3, 2 # left arm ]) # parent index parent_body_index = np.array([","in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist absolute orientation jacobian\\n') out_file.write('","j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\n')","out_file.write(' }\\n\\n') out_file.write(' // left foot contact points absolute position\\n') for i in","print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i","= matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write('","xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for","i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i])))","outputs class\\n') out_file.write(' *\\n') out_file.write(' * computation of:\\n') out_file.write(' * COM (center of","== nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com jacobian\\n')","def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for j in","= {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i,","as f: # loop on all the lines for line in f: cut_line_1","or 1 !'.format(cur_len-1)) exit() return result # write the beginning of the file","nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic","# jacobian rotation matrices out_file.write('\\n // jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n')","waist 2, 1, 3, 2, 1, 2, # right leg 2, 1, 3,","0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'),","right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ])","new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i]","Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj,","right foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","True, a, b, c except: return False, -1, -1, -1 # return true","in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet absolute orientation\\n')","Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable by","cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper()","def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for","jacobian xg out_file.write('\\n // jacobian com absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","// -- Collecting results -- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj)","symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the file","0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return","out_file.write(' double ') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1,","of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write('","variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb =","and replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector","kinematics computation for the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {')","waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist","0 for i in range (1, nb_bodies): flag_print = 0 for j in","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}]","nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n')","new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write('","is not always body i-1), expressed in the relative\\n') out_file.write(' * frame of","'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso',","1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","a float def isInt(value): try: int(value) return True except: return False # return","range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n') for i","'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies):","body i-1), expressed in the relative\\n') out_file.write(' * frame of the previous body\\n')","R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle in [pi/2 ;","out_file.write('\\n') out_file.write(' // left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1]","out_file.write('\\n') out_file.write(' // left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","{};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector # write symbolic","= 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else:","== 1: # positive neg_flag = 0 pos_str = elem_split[0] elif cur_len ==","# frame of the previous body # Rdi : rotational matrix between body","; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{},","# loop on all the joints for i in range (1, nb_bodies): new_vector","= sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] =","torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as","jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index):","22 # # 19 14 23 # # 20 01 24 # #","absolute velocities\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] =","= sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg","the lines for line in f: cut_line = line.split(' = ') if len(cut_line)","= get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj,","# end of the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name),","= sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')])","in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot","+= -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other","out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for i","11, # left leg 0, 13, 14, # trunk 15, 16, 17, 18,","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0] =","yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n')","xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian","0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index,","[0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine,","save_vector[i] = ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector","// right foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","body\\n') out_file.write(' * DGi : position vector from the anchor point of body","wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range","= sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write('","out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist absolute orientation jacobian\\n')","located at the origin (waist), but aligned with the ground (info from IMU)","* [None] for i in range(0,3): if vector[i] == 0 or vector[i] ==","elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split =","in the waist, middle point between the two pitch hip rotations\\n') out_file.write(' *","out_file.write('\\n // rotation matrices\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None]","for j in range(1, nb_bodies): flag_print = 0 for k in range(0, 3):","its COM (center of mass) G_i,\\n') out_file.write(' * expressed in the relative frame","j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n')","cur_split[i].upper()) return new_string # write the end of the file def write_file_end(out_file): out_file.write('}\\n')","= sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot =","nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None]","3, 2, # right arm 2, 1, 3, 2 # left arm ])","and its predecessor\\n') out_file.write(' * si : sine of the relative angle before","-{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0],","i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n')","range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first","return count # print the declaration of an element def print_declaration_elem(in_file, out_write, elem,","out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\n') count","rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp","write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on all the joints","jacobian anchor point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet absolute orientation\\n') for i in range(0,","* (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot)","out_file.write(' }\\n') # omega out_file.write('\\n // joint absolute velocities\\n') Om = nb_bodies*[None] om","nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp)","elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute","the joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) #","* Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i],","rot_axis = np.array([0, # waist 2, 1, 3, 2, 1, 2, # right","1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1,","= sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts *","range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] =","an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count","out_file.write('\\n') out_file.write(' // left foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}]","if len(cut_line) == 2: count += 1 return count # print the declaration","xp out_file.write('\\n // anchor point absolute positions and velocities\\n') x = nb_bodies*[None] xp","'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n // jacobian com absolute positions\\n') out_file.write(' if","= 0 pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag =","elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1))","s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if","'_d{}'.format(i+1)) # write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj,","out_file.write(' // right foot contact points absolute position\\n') for i in range(0, nb_contacts):","i in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1,","sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')])","out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write('","for i in range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i]","derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} *","temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the","%a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1])","for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write","absolute velocity\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first =","xgj_print, x_vector, index): # loop on all the joints for i in range","= './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names)","0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis","j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute orientation jacobian\\n') out_file.write(' if",": position vector from the anchor point of body i to its COM","om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\n // anchor point absolute positions","0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine,","write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i","flag_print = 0 for i in range(0,3): for j in range(0,3): if matrix[i,j]","1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] =","save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {}","') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1:","// left wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i,","range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the matrix","position and velocity\\n') out_file.write(' * feet position, velocity and orientation\\n') out_file.write(' * waist","the two pitch hip rotations # inertial frame: located at the origin (waist),","06 12 # # 07 13 # # # ###################### # # origin:","8, 9, 10, 11, # left leg 0, 13, 14, # trunk 15,","negative detection !'.format(cur_len)) exit() # compute derivative result = 0 # cosine if","r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb =","np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12',","with the ground (info from IMU) # # Di : position vector from","end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return","out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg & xgp out_file.write('\\n // com absolute positions","its COM (center of mass) G_i, # expressed in the relative frame of","c_y_torso;\\n\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n')","cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string =","right leg 0, 7, 8, 9, 10, 11, # left leg 0, 13,","in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k,","parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp =","foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if","'Ri' # return true if it is a float def isInt(value): try: int(value)","cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine,","orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso","matrix # Rti : transpose matrix of Ri # xji : jacobian of","foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n')","for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left","of 'elem' in the file def count_elem(in_file, elem): count = 0; with open(in_file,","axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]])","Rj, xj, xgj, der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip()","out_file.write(' * // 06 12 //\\n') out_file.write(' * // 07 13 //\\n') out_file.write('","3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if","xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j]","sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'),","1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i],","= sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13]","the relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write(' * Rdi :","cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n')","for line in f: # declaration if len(line.split('// -- variables initialization -- //'))","* // 04 10 //\\n') out_file.write(' * // 05 11 //\\n') out_file.write(' *","out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n')","the relative # frame of the previous body # Rdi : rotational matrix","{}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if","cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string #","Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z #","10 # # 05 11 # # 06 12 # # 07 13","symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector","# waist 2, 1, 3, 2, 1, 2, # right leg 2, 1,","r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb +","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}]","if xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1","compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp,","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg =","s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if","len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\",","# 03 09 # # 04 10 # # 05 11 # #","+ {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and","Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other","out_file.write(' // left foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] =","x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r =","* Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l =","nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0]","range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in","xj, xgj, Rj_print, R_matrix, index): # loop on all the joints for i","cosine of the relative angle before body i # # xi : absolute","= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1))","symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb","com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for i in range(1, nb_bodies): for j","out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if","inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' //","= len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] =","sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) ||","parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n // -- symbolic computation -- //\\n') # Rj,","* ////////////////////////\\n') out_file.write(' * // //\\n') out_file.write(' * // 17 16 21 //\\n')","Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm","'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right","joints enumeration\\n') out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)): count","sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] =","as sp import re import os ###################### # # # 17 16 21","'{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f: # loop on","loop on all the lines for line in f: cut_line = line.split(elem) if","02 08 //\\n') out_file.write(' * // 03 09 //\\n') out_file.write(' * // 04","the anchor point of body i to its COM (center of mass) G_i,\\n')","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute orientation jacobian\\n') out_file.write('","in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1","return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif","R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\n') for","in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n","xj_print, x_vector, index): # loop on all the joints for i in range","# # 18 15 22 # # 19 14 23 # # 20","xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] =","'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) #","cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0,","out_file.write(' * xi : absolute position vector (from origin, expressed in the inertial","1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis","+ ') out_file.write(' // global com absolute position\\n') for i in range(0, 3):","(flag_jacob)\\n {\\n') flag_first = 0 for i in range (1, nb_bodies): flag_print =","cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0,","* // //\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin: in the","write symbolic matrix and replace symbolic variable by its name def write_symb_matrix(out_file, matrix,","= nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0]","wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n')","'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names,","xgpi : derivative of xgi # omi : absolute rotational vector of body","# detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if","3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if","of xgi # omi : absolute rotational vector of body i # Ri","relative frame of the current body i\\n') out_file.write(' * Omi : rotational vector","f: cut_line = line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) ==","= sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7'","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg","* (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot +","Rti : transpose matrix of Ri # xji : jacobian of 'xi' #","xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12,","x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id'","sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2]","nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] =","right foot contact points absolute position\\n') for i in range(0, nb_contacts): for j","components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i,","right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3]","xg, xgp, xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i]))","term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 #","Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None]","c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso","if cur_jac != 0: if not flag_first: flag_first = 1 flag_print = 1","variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3)","!= 0: if not flag_first: flag_first = 1 flag_print = 1 elif not","not always body i-1), expressed in the relative # frame of the previous","== 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2","i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right wrist","xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj,","sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8'","# right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint","count = 0 if count != 0: out_write.write(';\\n') # get tilde matrix def","for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] =","//\\n') out_file.write(' * // 07 13 //\\n') out_file.write(' * // //\\n') out_file.write(' *","else: print('Error: {} instead of 1 or 2 in negative detection !'.format(cur_len)) exit()","= nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] =","x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T *","# from an orientation matrix, compute the roll, pitch, yaw angles (and derivative)","else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')","middle point between the two pitch hip rotations # inertial frame: located at","# # 03 09 # # 04 10 # # 05 11 #","theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file,","-result return result # compute the derivative of an expression (for jacobian) def","Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'),","if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print","# of the anchor point of body i # xgi : absolute position","* ci : cosine of the relative angle before body i\\n') out_file.write(' *\\n')","# generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt,","x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r","print('Error: {} * counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit()","return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 /","sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0],","; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{},","orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left","* Rji : jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute","# Di : position vector from the anchor point of the previous body",": absolute position vector of the COM G_i of body i # xpi","as np import sympy as sp import re import os ###################### # #","'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q))","i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the","== 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0],","replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector =","matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n')","rotation matrices out_file.write('\\n // rotation matrices\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd","jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","* [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] =","matrix between body i and its predecessor\\n') out_file.write(' * si : sine of","nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in","rotations\\n') out_file.write(' * inertial frame: located at the origin (waist), but aligned with","out_file.write(' // global com absolute position\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}]","body i-1), expressed in the relative # frame of the previous body #","* xgji : jacobian of \\'xgi\\'\\n') out_file.write(' * Rji : jacobian of \\'Ri\\'\\n')","new_vector, save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name):","in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right foot absolute velocity\\n') for i","line in f: # declaration if len(line.split('// -- variables initialization -- //')) !=","= sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write('","out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n","the relative angle before body i\\n') out_file.write(' *\\n') out_file.write(' * xi : absolute","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg","out_file.write('\\n // IMU - rotation matrices\\n') for i in range(0, 3): for j","torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name,","for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic","xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] *","R[i].T # jacobian rotation matrices out_file.write('\\n // jacobian rotation matrices\\n') out_file.write(' if (flag_jacob)\\n","a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write('","result = -result return result # compute the derivative of an expression (for","out_file.write(' * ci : cosine of the relative angle before body i\\n') out_file.write('","j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU","-> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5],","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n //","b except: return False, -1, -1 # count the number of 'elem' in","= sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None]","in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left wrist absolute velocity\\n') for i","in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0:","i in range (1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac","wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y,","sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i],","nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None]","flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n')","out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot","i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0]","left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right","range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot","Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist","# left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')])","xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0],","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n // -- symbolic computation -- //\\n')","out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\n')","torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\n')","out_file.write(' * xgi : absolute position vector of the COM G_i of body","out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","the previous body # Rdi : rotational matrix between body i and its","range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else:","COM (center of mass) position and velocity\\n') out_file.write(' * feet position, velocity and","waist, middle point between the two pitch hip rotations\\n') out_file.write(' * inertial frame:","x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T *","variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot =","17 16 21 //\\n') out_file.write(' * // 18 15 22 //\\n') out_file.write(' *","foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","= nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0]","computation -- //\\n') # Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None]","= sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0]","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","= ') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) ==","= om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot","0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0],","origin (waist), but aligned with the ground (info from IMU) # # Di","0: out_write.write(';\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0","Omi : rotational vector from the previous body to the current body i","= xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb","xj, xgj, der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip() #","def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for","== 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else:","if it is a float def isInt(value): try: int(value) return True except: return","* Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write('","R_r_foot[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n')","'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1])","positive neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2: # negative","]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist",", 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index,","sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9')","out_file.write(' // left foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n')","for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n //","declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f: #","i in range(0,3): for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j]","new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1))","0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >=","axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]])","//\\n') out_file.write('\\n // IMU - rotation matrices\\n') for i in range(0, 3): for","elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result","for k in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first","negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead of 1","sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1),","# beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names)","get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot contact points absolute position\\n')","[rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the wrists position, velocity and","in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] =","Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb","[0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else:","x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R,","for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for","right wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i]))","xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =","j)) flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0:","out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet absolute","!'.format(cur_len-1)) exit() return result # write the beginning of the file def write_file_beginning(out_file,","angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot","velocities\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] =","symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms term_list = str(symb_var).replace('-","* \\\\param[in,out] in_out inputs and outputs class\\n') out_file.write(' *\\n') out_file.write(' * computation of:\\n')","out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n') for i in range(1,","nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i]","try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a,","of the previous body to the current body i \\n') out_file.write(' * (previous","feet absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i]))","omega out_file.write('\\n // joint absolute velocities\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde","1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left","0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if","Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0,","{};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of","0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] =","frame of the previous body\\n') out_file.write(' * DGi : position vector from the","Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1]","= y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16')","range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n') for i in range(0, 9):","xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return result # compute","return True, a, b except: return False, -1, -1 # count the number","position\\n') for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}]","out_file.write('\\n') # compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R,","m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1:","{} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the","sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'),","cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)):","out_file.write('\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot')","= 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result","c except: return False, -1, -1, -1 # return true if it has","yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left foot orientation matrix as angles","out_file.write('\\n // joint sines\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1,","* [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot","# right leg 0, 7, 8, 9, 10, 11, # left leg 0,","# right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0])","get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file,","or 2 in negative detection !'.format(cur_len)) exit() # compute derivative result = 0","sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7' ,","sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')])","in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in range(0,","sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0],","return False, -1, -1, -1 # return true if it has a shape","compute derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q): result +=","Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist","i+1, end_name) save_vector[i] = ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return","'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the wrists position, velocity and orientation def wrists_compute(out_file,","def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None]","left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j],","Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0])","= open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization","jacobian of \\'xi\\'\\n') out_file.write(' * xgji : jacobian of \\'xgi\\'\\n') out_file.write(' * Rji","+= der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","the anchor point of the previous body to the current body i \\n')","j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n","gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp =","vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return","# right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' ,","sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')])","a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b,","2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj,","# right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0])","om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and orientation def feet_compute(out_file,","0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1))","= {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left foot absolute velocity\\n') for i in","get a string for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_')","axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0],","02 08 # # 03 09 # # 04 10 # # 05","the anchor point of body i\\n') out_file.write(' * xgi : absolute position vector","point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) #","except: return False, -1, -1, -1 # return true if it has a","0.0, 1.0]]) else: return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0,","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","= '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if","nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the matrix elements for","21 # # 18 15 22 # # 19 14 23 # #","in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part,","3, 2, 1, 2, # right leg 2, 1, 3, 2, 1, 2,","flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') #","'{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print:","* \\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics computation","out_file.write(' * // 18 15 22 //\\n') out_file.write(' * // 19 14 23","i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j","= sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j]","anchor point of body i\\n') out_file.write(' * xgi : absolute position vector of","x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] #","= sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2]","i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all","0 or 1 !'.format(cur_len-1)) exit() return result # write the beginning of the","= 0 for i in range(0,3): if vector[i] == 0 or vector[i] ==","range(1, nb_bodies): flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k] !=","0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) #","def isInt(value): try: int(value) return True except: return False # return true if","matrices\\n') for i in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{}","* // 03 09 //\\n') out_file.write(' * // 04 10 //\\n') out_file.write(' *","exit() # compute derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q):","sp.zeros(3, 3) for i in range(0, 3): for j in range(0, 3): R[0][i,j]","range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]]","on all the lines for line in f: cut_line = line.split(elem) if len(cut_line)","in f: cut_line = line.split(elem) if len(cut_line) == 2: count += 1 return","vector of body i # Ri : absolute rotational matrix # Rti :","of 1 or 2 in negative detection !'.format(cur_len)) exit() # compute derivative result","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic","* Omi : rotational vector from the previous body to the current body","nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1,","angles velocity\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n","for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on","= nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1]","'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', #","* // //\\n') out_file.write(' * // 17 16 21 //\\n') out_file.write(' * //","Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0,","0; with open(in_file, 'r') as f: # loop on all the lines for","joint cosines\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n","(1, nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj,","1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1,","before body i # ci : cosine of the relative angle before body","com absolute position\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first","22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,","left foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","out_file.write('\\n') return new_matrix # save the symbolic vector for print def print_save_symb_vector(vector, start_name,","Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file,","if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q):","&in_out)\\n{\\n') # compute the center of mass position and velocity def com_compute(out_file, nb_bodies,","result = 0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) #","= sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n') for i in range(1, nb_bodies):","out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write('","else: # epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write('","[0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3:","shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1])","replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix =","j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for","else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2:","nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] =","previous body\\n') out_file.write(' * Rdi : rotational matrix between body i and its","= 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","xi : absolute position vector (from origin, expressed in the inertial frame) #","Rj, xj, xgj, der_var) # one product elif cur_len == 2: result +=","not flag_first: flag_first = 1 flag_print = 1 elif not flag_print: flag_print =","= 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1,","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double","def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) #","3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index),","velocities\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i]","wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n')","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","# right leg 2, 1, 3, 2, 1, 2, # left leg 1,","sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ##","rotation matrices\\n') for i in range(0, 3): for j in range(0, 3): out_file.write('","axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]])","R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon =","-0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output file","* \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics computation for the COMAN model\\n')","xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb","= write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies):","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') #","01 24 # # 02 08 # # 03 09 # # 04","from IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di : position vector from the anchor","epsilon = 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] =","in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n')","* // 20 01 24 //\\n') out_file.write(' * // 02 08 //\\n') out_file.write('","nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot)","R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist","frame of the previous body\\n') out_file.write(' * Rdi : rotational matrix between body","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n",", sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] =","[None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3,","* Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l =","sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10]","derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' * // //\\n') out_file.write(' * //","x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb","get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1],","09 # # 04 10 # # 05 11 # # 06 12","x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot","= {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot contact points jacobian\\n') out_file.write('","point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on all","'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , #","elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive neg_flag = 0","in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1]","in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies): if","flag_first = 0 for i in range (1, nb_bodies): flag_print = 0 for","= symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:","d, e] = isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] #","rotational matrix # Rti : transpose matrix of Ri # xji : jacobian","'s{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str)","Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3,","// right foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i,","the lines for line in f: cut_line = line.split(elem) if len(cut_line) == 2:","xgp, xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if","and velocities\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies):","print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f: # loop on","# # ###################### # # origin: in the waist, middle point between the","in the inertial frame) # of the anchor point of body i #","else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) #","flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j],","sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if","Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in","for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left","body i to its COM (center of mass) G_i, # expressed in the","# symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n","= int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c","expressed in the relative frame of the current body i # Omi :","x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] *","all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result =","i\\n') out_file.write(' * Omi : rotational vector from the previous body to the","-- variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy temporary file","if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1,","get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l =","%c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c","if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i != nb_bodies-1:","product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result","* waist and torso orientaion angles and derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n')","flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1))","out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write('","1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in","of the previous body to the current body i # (previous body is","temporary file os.remove(in_temp) # main script # rotation axis for each joint before","get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis ==","2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2, 1, 2,","arm ]) # parent index parent_body_index = np.array([ -1, # waist 0, 1,","foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n')","= line.split(elem) if len(cut_line) == 2: count += 1 return count # print","all the matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj,","01 24 //\\n') out_file.write(' * // 02 08 //\\n') out_file.write(' * // 03","range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0,","'torso') # compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R,","1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont","i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist","sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2]","i to its COM (center of mass) G_i, # expressed in the relative","# omega out_file.write('\\n // joint absolute velocities\\n') Om = nb_bodies*[None] om = nb_bodies*[None]","= y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in","the inertial frame) # of the anchor point of body i # xgi","= {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian","Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies,","contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] =","out_file.write(' // right wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] =","print('Error: {} instead of 1 or 2 in negative detection !'.format(cur_len)) exit() #","x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') #","+= len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\n') count = 0 if","* [None] for i in range(0,3): for j in range(0,3): if matrix[i,j] ==","x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot)","(from origin, expressed in the inertial frame)\\n') out_file.write(' * of the anchor point","in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the","anchor point of body i to its COM (center of mass) G_i, #","new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for print","# remove temporary file os.remove(in_temp) # main script # rotation axis for each","out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write('","for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right","instead of 1 or 2 in negative detection !'.format(cur_len)) exit() # compute derivative","in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in","= Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] =","xgp out_file.write('\\n // com absolute positions and velocities\\n') xg = nb_bodies*[None] xgp =","the relative frame of the current body i\\n') out_file.write(' * Omi : rotational",": rotational vector from the previous body to the current body i #","xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot *","in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute position\\n')","in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1]","k in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first =","'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' ,","= 0; with open(in_file, 'r') as f: # loop on all the lines","0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write('","the ground (info from IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di : position vector","R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time",": jacobian of 'xi' # xgji : jacobian of 'xgi' # Rji :","c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist","matrices\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3,","* \\\\brief forward kinematics computation for the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints","== 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1,","absolute position vector of the COM G_i of body i # xpi :","= om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1)","0 if count != 0: out_write.write(';\\n') # get tilde matrix def get_tilde(v): return","c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint sines\\n') for i in range(1, nb_bodies):","R[0] = sp.zeros(3, 3) for i in range(0, 3): for j in range(0,","* // 06 12 //\\n') out_file.write(' * // 07 13 //\\n') out_file.write(' *","# compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x,","i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n')","= {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet absolute orientation\\n') for","{};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // feet absolute orientation\\n') for i","in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot contact points jacobian\\n')","write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg,","x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T *","jacobian of 'Ri' # return true if it is a float def isInt(value):","// right foot contact points absolute position\\n') for i in range(0, nb_contacts): for","der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] ==","x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","== 2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip()))","!= 1: out_file.write(' // -- variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n')","*/\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {') count = 0 for i in range(1,","sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n') for i in range(1, nb_bodies): out_file.write('","-sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name,","= x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist","if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine,","number of 'elem' in the file def count_elem(in_file, elem): count = 0; with","nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T","xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb =","# waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' ,","of the relative angle before body i # ci : cosine of the","12 # # 07 13 # # # ###################### # # origin: in","(indexes %a, %b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b","in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first:","body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) +","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') #","sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector # write symbolic matrix and replace symbolic","'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n // jacobian anchor point positions\\n') out_file.write(' if","returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1])","sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'),","1 return count # print the declaration of an element def print_declaration_elem(in_file, out_write,","np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3:","i+1, j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print:","in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2]","joint relative velocities\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i]))","right arm 15, 20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index)","flag_first: flag_first = 1 flag_print = 1 elif not flag_print: flag_print = 1","03 09 # # 04 10 # # 05 11 # # 06","if count >= nb_max_line: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n')","i in range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif","i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for","// left foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i,","= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1))","in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i]","range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1","flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n // joint absolute","'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22',","R_r_elb[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n')","origin: in the waist, middle point between the two pitch hip rotations #","before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1,","range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if","pitch hip rotations # inertial frame: located at the origin (waist), but aligned","direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0,","if ((!c_y_waist) || (!c_y_torso))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0 /","}\\n\\n') out_file.write(' // left wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') # get a","1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont =","j, cur_jac)) out_file.write(' }\\n\\n') # get a string for the enumeration of joints","// 18 15 22 //\\n') out_file.write(' * // 19 14 23 //\\n') out_file.write('","if len(cur_split) >= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string","# # origin: in the waist, middle point between the two pitch hip","0 for i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print =","COM (center of mass) G_i, # expressed in the relative frame of the","sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0,","count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f: # loop","the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation","symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3,",": position vector from the anchor point of the previous body to the","out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script # rotation axis","symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3,","result = 0 # loop on all terms for cur_term in term_list: #","global com absolute position\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i))","20 01 24 //\\n') out_file.write(' * // 02 08 //\\n') out_file.write(' * //","write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\n //","= 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","07 13 //\\n') out_file.write(' * // //\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write('","//\\n') # Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj =","2 if count >= nb_max_char: out_write.write(';\\n') count = 0 if count != 0:","by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print","for k in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first","each joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist","body is not always body i-1), expressed in the relative\\n') out_file.write(' * frame","in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists absolute orientation\\n') for","xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] =","> 0: # epsilon = 1 -> pitch angle in [-pi/2 ; pi/2]","in f: cut_line = line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('['))","vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} =","'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27',","(for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string)","xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i],","'elem' in the file def count_elem(in_file, elem): count = 0; with open(in_file, 'r')","= symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:","as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso orientation matrix","flag_print = 0 for i in range(0,3): if vector[i] == 0 or vector[i]","for i in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] =","write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '')","range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j]))","14 23 # # 20 01 24 # # 02 08 # #","i to its COM (center of mass) G_i,\\n') out_file.write(' * expressed in the","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right wrist absolute","expressed in the inertial frame)\\n') out_file.write(' * of the anchor point of body","r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z","# left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt =","position vector of the COM G_i of body i\\n') out_file.write(' * xpi :","xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] =","xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only implemented","'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} +","xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] =","variables initialization -- //\\n') out_file.write('\\n // IMU - rotation matrices\\n') for i in","arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] #","beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n')","range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j]))","save_matrix[3*i+j] = ' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix","0 for k in range(0, 3): if xj_print[i][j][k] != None: if not flag_first:","* counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result","epsilon > 0: # epsilon = 1 -> pitch angle in [-pi/2 ;","flag_print = 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj,","write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write('","flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n // -- Collecting","vector from the anchor point of body i to its COM (center of","out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute position\\n') for i in range(0,3): out_file.write('","expressed in the relative # frame of the previous body # DGi :","j+1, 3*i+j)) out_file.write('\\n // IMU - angles velocity\\n') for i in range(0, 3):","in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable by its name","om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute orientation jacobian\\n')","joint_id_names[i])) out_file.write('\\n // joint sines\\n') for i in range(1, nb_bodies): out_file.write(' s{} =","xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id]","x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1,","xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om,","# # # 17 16 21 # # 18 15 22 # #","for line in f: cut_line = line.split(' = ') if len(cut_line) == 2:","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac))","print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in","right wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i]))","nb_bodies): flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k] != None:","# trunk 2, 1, 3, 2, # right arm 2, 1, 3, 2","forward_kinematics.cc\\n') out_file.write(' * \\\\brief forward kinematics computation for the COMAN model\\n') out_file.write(' */\\n\\n')","out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i])))","position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,","if count != 0: out_write.write(';\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0,","= nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3)","matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a","M): out_file.write('\\n\\n // -- symbolic computation -- //\\n') # Rj, xj, xgj and","*\\n') out_file.write(' * origin: in the waist, middle point between the two pitch","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot","symbolic matrix and replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name,","return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif","= 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j],","x_r[i])) out_file.write('\\n') out_file.write(' // right wrist absolute velocity\\n') for i in range(0,3): out_file.write('","= sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix # save the symbolic vector for","R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3,","# loop on all the joints for i in range (1, nb_bodies): new_matrix","vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None","= sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0,","# element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q","position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,","xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =","* Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb +","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","for i in range(0,3): for j in range(0,3): if matrix[i,j] == 0 or","absolute rotational matrix\\n') out_file.write(' * Rti : transpose matrix of Ri\\n') out_file.write(' *","= sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17]","= xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print,","body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the waist","+ c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles","out_file.write(' in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist jacobian\\n') out_file.write(' if","np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else:","for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right","range (1, nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac =","axis def get_vector_axis(axis, direct, elem): if direct: if axis == 1: return np.array([[elem],","write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id","count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics","= cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso =","-1, # waist 0, 1, 2, 3, 4, 5, # right leg 0,","der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) #","1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0","x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T *","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') #","xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1","range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector,","start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for i","Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] =",": rotational matrix between body i and its predecessor # si : sine","sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')])","-sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0,","om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot","in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] =","symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first","joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z):","sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')])","= write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\n","rotation axis for each joint before body i (1:x, 2:y, 3:z) rot_axis =","= c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} *","position\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' //","symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first","R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] =","Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i]","nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] *","com absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right wrist","2: out_file.write(';\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min,","in_out.rp_Lwrist[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n","the two pitch hip rotations\\n') out_file.write(' * inertial frame: located at the origin","xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =","sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk","Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0]","# right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8')","int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return False, -1, -1","# other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] =","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot contact points absolute position\\n') for i","== 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif","joint_id_names): out_file.write(' // -- variables initialization -- //\\n') out_file.write('\\n // IMU - rotation","len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip()))","range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left wrist absolute velocity\\n')","nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j],","Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5]","yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0:","if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write('","flag_print = 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj,","9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot absolute orientation","range (1, nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac =","3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return","range(0, nb_contacts): for j in range (1, nb_bodies): flag_print = 0 for k","i)) if i == max: out_file.write(';\\n') else: out_file.write(', ') # variables initialization def","sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] =","k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac","a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a =","c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write(' // waist orientation angle derivatives","# epsilon = 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0]","import sympy as sp import re import os ###################### # # # 17","cos(in_out.theta_waist[2]);\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n')","1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left","in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac !=","flag_first = 0 for i in range(0, nb_contacts): for j in range (1,","{};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine,","out_file.write(' // left foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","xgj, xgj_print, x_vector, index): # loop on all the joints for i in","symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n //","# x & xp out_file.write('\\n // anchor point absolute positions and velocities\\n') x","in range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xj_print[i][j][k]","Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0,","'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk","def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on all the","-- //\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x,","// 20 01 24 //\\n') out_file.write(' * // 02 08 //\\n') out_file.write(' *","3 * [None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\n //","# trunk 15, 16, 17, 18, # right arm 15, 20, 21, 22","{}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j ==","i # xpi : derivative of xi # xgpi : derivative of xgi","# compute the derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj,","joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop","// anchor point absolute positions and velocities\\n') x = nb_bodies*[None] xp = nb_bodies*[None]","+ 2 if count >= nb_max_char: out_write.write(';\\n') count = 0 if count !=","21, 22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions","* Rti : transpose matrix of Ri\\n') out_file.write(' * xji : jacobian of","relative angle before body i # ci : cosine of the relative angle","and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot =","((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n')","'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left","True, a, b except: return False, -1, -1 # count the number of","in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) ==","out_file.write(' * Rji : jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') #","derivative of xi\\n') out_file.write(' * xgpi : derivative of xgi\\n') out_file.write(' * omi","def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on all the","right foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i]))","= sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0","always body i-1), expressed in the relative # frame of the previous body","vector from the previous body to the current body i # (previous body","flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name,","of the anchor point of body i # xgi : absolute position vector","'{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name)","R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else:","3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0]","write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '')","if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine,","* [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None] #","atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({},","sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] =","= 0 for k in range(0, 9): if Rj_print[i][j][k] != None: if not","parent index parent_body_index = np.array([ -1, # waist 0, 1, 2, 3, 4,","feet position, velocity and orientation\\n') out_file.write(' * waist and torso orientaion angles and","vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector # write symbolic matrix","= sp.zeros(3, 1) # loop on all the vector elements for j in","c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag:","matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1]","xgj, Rj_print, R_matrix, index): # loop on all the joints for i in","np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis","0 for i in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print","omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{})","= {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot contact points","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}]","foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' //","out_file.write(' + ') out_file.write(' // global com absolute position\\n') for i in range(0,","// left foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0","Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7')","Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj","= nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in","positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right","body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part,","1, 3, 2, # right arm 2, 1, 3, 2 # left arm","# waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7')","Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk","= get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l","for k in range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first","(R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T","out_file.write(' }\\n') # xg & xgp out_file.write('\\n // com absolute positions and velocities\\n')","// IMU - rotation matrices\\n') for i in range(0, 3): for j in","= sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot)","j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]]","1) # loop on all the vector elements for j in range(0, 3):","for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] =","out_file.write(' // right wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] =","xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0,","jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index):","write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix,","i == max: out_file.write(';\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies,","atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{},","velocity\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0","= nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r","== 2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip()))","= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('","nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write(' // global com absolute position\\n') for","# write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj,","has a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a","np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix","{};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j]","xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n","0, 1, 2, 3, 4, 5, # right leg 0, 7, 8, 9,","//\\n') out_file.write(' * // 06 12 //\\n') out_file.write(' * // 07 13 //\\n')","out_file.write('\\n') out_file.write(' // right foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'),","= 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file,","= {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","(position and derivative) of the waist and the torso def torso_waist_angles(out_file, R, om,","cosine, sine): if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0,","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}]","1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n // joint absolute velocities\\n') Om","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}]","def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x,","IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU - angles velocity\\n') for i","nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] +","a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try: a","else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0,","foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n')","get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i","9 * [None] for i in range(0,3): for j in range(0,3): if matrix[i,j]","of body i # xgi : absolute position vector of the COM G_i","true if it has a shape 'x%a_%b' (indexes %a, %b also returned) def","elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if","0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine,","anchor point of the previous body to the current body i \\n') out_file.write('","cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n')","= R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1)","i in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]),","for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts):","in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j == 2:","om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot =","# rotation axis for each joint before body i (1:x, 2:y, 3:z) rot_axis","19 14 23 # # 20 01 24 # # 02 08 #","left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","-0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output file def","= sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file,","= sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] =","!'.format(cur_len)) exit() # compute derivative result = 0 # cosine if pos_str ==","return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis == 1: return","om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i]","out_file.write('\\n') out_file.write(' // left wrist absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}]","0 with open(in_file,'r') as f: # loop on all the lines for line","-- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy temporary file out_file.write(line) out_file.close() #","x out_file.write('\\n // jacobian anchor point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","cur_len == 2: # negative neg_flag = 1 pos_str = elem_split[1] else: print('Error:","rotational vector of body i\\n') out_file.write(' * Ri : absolute rotational matrix\\n') out_file.write('","matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name,","sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] =","'w') with open(in_temp, 'r') as f: # loop on all the lines for","else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\n')","xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic","left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9]","* frame of the previous body\\n') out_file.write(' * Rdi : rotational matrix between","origin (waist), but aligned with the ground (info from IMU)\\n') out_file.write(' *\\n') out_file.write('","= 0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine","detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: #","# # 07 13 # # # ###################### # # origin: in the","out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)): count += 1","nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f:","and torso orientaion angles and derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *","j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first","the COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {') count = 0","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}]","new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1))","nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M)","= cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist =","body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the","contact points absolute position\\n') for i in range(0, nb_contacts): for j in range(0,","out_file.write(' // torso orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n')","joint_id_names[i])) # write symbolic vector and replace symbolic variable by its name def","y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2]","for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i],","cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine],","vector of body i\\n') out_file.write(' * Ri : absolute rotational matrix\\n') out_file.write(' *","xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print =","'M_29' # left arm ]) # joint names joint_id_names = np.array(['0', # waist","(jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None]","* Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj,","output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop","om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1,","//\\n') out_file.write(' * // 03 09 //\\n') out_file.write(' * // 04 10 //\\n')","+ om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] =","out_file.write(' }\\n\\n') out_file.write(' // wrists absolute orientation\\n') for i in range(0, 9): out_file.write('","in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n') for","try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return","isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except:","elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1))","write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '')","3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2, 1, 2, #","'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left","der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis","located at the origin (waist), but aligned with the ground (info from IMU)\\n')","beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) #",": absolute rotational vector of body i # Ri : absolute rotational matrix","in negative detection !'.format(cur_len)) exit() # compute derivative result = 0 # cosine","position vector (from origin, expressed in the inertial frame)\\n') out_file.write(' * of the","= nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] =","[rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation","18, # right arm 15, 20, 21, 22 # left arm ]) nb_bodies","right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18]","i # # xi : absolute position vector (from origin, expressed in the","matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3):","i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i,","= cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string",": jacobian of \\'xi\\'\\n') out_file.write(' * xgji : jacobian of \\'xgi\\'\\n') out_file.write(' *","out_file.write('\\n // IMU - angles velocity\\n') for i in range(0, 3): out_file.write(' omega_{}","'') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] +","# Rji : jacobian of 'Ri' # return true if it is a","si : sine of the relative angle before body i\\n') out_file.write(' * ci","angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left foot orientation","= nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None]","print the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file,","'') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i]","products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len ==","1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg & xgp out_file.write('\\n // com absolute","= sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20]","matrix of Ri # xji : jacobian of 'xi' # xgji : jacobian","of Ri\\n') out_file.write(' * xji : jacobian of \\'xi\\'\\n') out_file.write(' * xgji :","matrix between body i and its predecessor # si : sine of the","else: return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if direct:","omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the waist and","point between the two pitch hip rotations\\n') out_file.write(' * inertial frame: located at","velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,","= 0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first:","R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1),","numpy as np import sympy as sp import re import os ###################### #","in the relative frame of the current body i # Omi : rotational","Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only","i and its predecessor\\n') out_file.write(' * si : sine of the relative angle","jacobian of 'xi' # xgji : jacobian of 'xgi' # Rji : jacobian","= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xj_print[i][j][k]))","sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'),","range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >=","om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist","j in range(1, nb_bodies): flag_print = 0 for k in range(0, 9): if","sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'),","# right arm 15, 20, 21, 22 # left arm ]) nb_bodies =","3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index),","Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb =","* [None] # rotation matrices out_file.write('\\n // rotation matrices\\n') R = nb_bodies*[None] Rt","cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not","3, # trunk 2, 1, 3, 2, # right arm 2, 1, 3,","np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return","nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0,","2, # right leg 2, 1, 3, 2, 1, 2, # left leg","Ri : absolute rotational matrix # Rti : transpose matrix of Ri #","xi\\n') out_file.write(' * xgpi : derivative of xgi\\n') out_file.write(' * omi : absolute","positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right","return new_matrix, save_matrix # write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies,","if flag_print: out_file.write('\\n') return new_matrix # save the symbolic vector for print def","M = np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10',","xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb","0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM","body i \\n') out_file.write(' * (previous body is not always body i-1), expressed","body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} +","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","initialization -- //\\n') out_file.write('\\n // IMU - rotation matrices\\n') for i in range(0,","}\\n\\n') out_file.write(' // left wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}]","# xgpi : derivative of xgi # omi : absolute rotational vector of","Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0,","the relative angle before body i\\n') out_file.write(' * ci : cosine of the","range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and","out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n')","out_file.write('\\n') out_file.write(' // left foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive","cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n","angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon = 1 -> pitch","def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f: # loop","3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies):","jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string) elem_str","hip rotations\\n') out_file.write(' * inertial frame: located at the origin (waist), but aligned","file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write(' * \\\\file","Dg, M): out_file.write('\\n\\n // -- symbolic computation -- //\\n') # Rj, xj, xgj","absolute positions and velocities\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in","for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' //","# apply negative if neg_flag: result = -result return result # compute the","sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] =","nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] =","= sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0,","om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1)","1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id],","right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20',","Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm","== 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1,","None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\n'.format(elem_name,","k in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first =","the number of 'elem' in the file def count_elem(in_file, elem): count = 0;","= sp.zeros(3, 3) save_matrix = 9 * [None] for i in range(0,3): for","der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if","the current body i # (previous body is not always body i-1), expressed","y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot =","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0,","the relative angle before body i # # xi : absolute position vector","atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file,","out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write('","absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot","= int(value.split('_')[1]) return True, a, b except: return False, -1, -1 # count","range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j],","xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation","range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot +","{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle in [pi/2","else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name,","model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {') count = 0 for i","i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '')","{} * counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit() return","= {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot absolute orientation jacobian\\n') out_file.write(' if","outputs out_file.write(' // right wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}]","cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not","1.0 / c_y_torso;\\n\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist',","x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb *","= xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb =","of the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with","nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name)","flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com absolute velocity\\n') for i","in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\n') else: out_file.write(', ')","int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except:","angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write('","from the previous body to the current body i # (previous body is","derivative) of the waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id):","joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045,","joint sines\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n","(waist), but aligned with the ground (info from IMU) # # Di :","get_vector_axis(axis, direct, elem): if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]])","epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] =","of the previous body # DGi : position vector from the anchor point","// -- variables initialization -- //\\n') out_file.write('\\n // IMU - rotation matrices\\n') for","[rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left foot orientation matrix as","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","-- variables initialization -- //')) != 1: out_file.write(' // -- variables declaration --","(flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_contacts): for j in","// 04 10 //\\n') out_file.write(' * // 05 11 //\\n') out_file.write(' * //","= sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')])","elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name)","1) out_file.write('\\n') out_file.write(' // torso orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id],","computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n // --","x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file,","# count the number of 'elem' in the file def count_elem(in_file, elem): count","absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write('","if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return","loop on all the vector elements for j in range(0, 3): new_vector[j] =","elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0,","0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02,","////////////////////////\\n') out_file.write(' *\\n') out_file.write(' * origin: in the waist, middle point between the","== 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1","sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5]","= sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0","xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1","omega_in[0], omega_in[2])) # angles (position and derivative) of the waist and the torso","# left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' ,","r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb","!= 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k],","0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if","[0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0,","# ###################### # # origin: in the waist, middle point between the two","// right foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\n'.format(i,","direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2:","%b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return","xg & xgp out_file.write('\\n // com absolute positions and velocities\\n') xg = nb_bodies*[None]","print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as","// left foot contact points absolute position\\n') for i in range(0, nb_contacts): for","from the previous body to the current body i \\n') out_file.write(' * (previous","Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0,","2, 3, # trunk 2, 1, 3, 2, # right arm 2, 1,","(flag_jacob)\\n {\\n') flag_first = 0 for i in range(0, nb_bodies): for j in","jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms term_list",": rotational vector from the previous body to the current body i \\n')","for cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split)","axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]])","sine of the relative angle before body i # ci : cosine of","2 in negative detection !'.format(cur_len)) exit() # compute derivative result = 0 #","xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only implemented for","R_matrix, epsilon): if epsilon > 0: # epsilon = 1 -> pitch angle","+ om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist","def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string) elem_str =","expressed in the relative frame of the current body i\\n') out_file.write(' * Omi","in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot orientation matrix","range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] else:","loop on all the lines for line in f: cut_line_1 = line.split(elem) cut_line_2","[0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis ==","* xgi : absolute position vector of the COM G_i of body i\\n')","= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k]))","= sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'),","out_file.write(' // right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","}\\n\\n') out_file.write(' // left foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists absolute orientation\\n') for i in","Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","0 for i in range(1, len(joint_id_names)): count += 1 if i == 1:","= None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {}","* R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i],","matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]])","False, -1, -1, -1 # return true if it has a shape 'x%a_%b'","Rji : jacobian of 'Ri' # return true if it is a float","pitch hip rotations\\n') out_file.write(' * inertial frame: located at the origin (waist), but","19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the","= nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] *","and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation","Dg, M) # end of the file write_file_end(file_temp) file_temp.close() # output file out_file","= 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","3 * [None] for i in range(0,3): if vector[i] == 0 or vector[i]","3): if xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print =","/ c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot',","of body i\\n') out_file.write(' * Ri : absolute rotational matrix\\n') out_file.write(' * Rti","[0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) #","global com absolute velocity\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i))","// -- variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy temporary","symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max:","joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n') for i in range(1, nb_bodies): out_file.write(' Om{}","sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] =","x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id]","joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right","out_file.write(' * xpi : derivative of xi\\n') out_file.write(' * xgpi : derivative of",", 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' #","out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write('","np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis ==","= get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot +","if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com jacobian\\n') out_file.write(' if","// waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' //","for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else:","jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center of","new_matrix = sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for j","1: new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name)","0 for k in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first:","in the relative # frame of the previous body # Rdi : rotational","if (flag_jacob)\\n {\\n') for i in range(1, nb_bodies): for j in range(0, 3):","return new_matrix # save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name):","f: # loop on all the lines for line in f: cut_line_1 =","an orientation matrix, compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file,","# joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id',","xgi\\n') out_file.write(' * omi : absolute rotational vector of body i\\n') out_file.write(' *","out_file.write(' // left wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies,","# print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for","max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i","origin: in the waist, middle point between the two pitch hip rotations\\n') out_file.write('","'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id],","i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont","wrists absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i]))","np.array([]) # compute the derivative of an element (for jacobian) def der_elem(elem_str, Rj,","= write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian","(Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file,","= isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif","implemented for 0 or 1 !'.format(cur_len-1)) exit() return result # write the beginning","'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' ,","symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector,","the relative angle before body i # ci : cosine of the relative","in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n","between body i and its predecessor\\n') out_file.write(' * si : sine of the","in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] =","sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left","sp.zeros(3, 1) # loop on all the vector elements for j in range(0,","angle before body i # ci : cosine of the relative angle before","in range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies,","sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([","write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as","count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) +","(center of mass) position and velocity\\n') out_file.write(' * feet position, velocity and orientation\\n')","range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist jacobian\\n') out_file.write('","* of the anchor point of body i\\n') out_file.write(' * xgi : absolute","1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left","out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // wrists absolute orientation\\n')","line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count","left foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write('","in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2","sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n')","else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write('","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1","omega_in[2])) # angles (position and derivative) of the waist and the torso def","i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i ==","= parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] +","'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', #","the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\n') out_file.write(' * \\\\author","= x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T","the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om,","* DGi : position vector from the anchor point of body i to","aligned with the ground (info from IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di :","body to the current body i \\n') out_file.write(' * (previous body is not","3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 *","= '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write('","= sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min","absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\n'.format(i, R_r_elb[i])) out_file.write('\\n')","sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'),",", 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left","}\\n\\n') out_file.write(' // feet absolute orientation\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}]","in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and","* (previous body is not always body i-1), expressed in the relative\\n') out_file.write('","waist 0, 1, 2, 3, 4, 5, # right leg 0, 7, 8,","Di : position vector from the anchor point of the previous body to","out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i","out_file.write(' // right foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1)","# masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9'","= 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 *","orientation matrix, compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name,","anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0])","out_file.write('/*! \\n') out_file.write(' * \\\\author <NAME>\\n') out_file.write(' * \\\\file forward_kinematics.cc\\n') out_file.write(' * \\\\brief","[None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None] # rotation",", 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' ,","xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point","'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj,","print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for","15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names,","body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part,","cur_len = len(elem_split) if cur_len == 1: # positive neg_flag = 0 pos_str","in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] =","in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] =","= 1 out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n // -- Collecting results","i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in","c_z_torso = cos(in_out.theta_torso[2]);\\n\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist","absolute rotational vector of body i\\n') out_file.write(' * Ri : absolute rotational matrix\\n')","for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right","left foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i]))","M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning","\\\\brief main kinematics computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out inputs and outputs","cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string = cur_split[0].upper() for i in range(1,","sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23]","if len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' // -- variables","axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0],",", sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] =","0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1))","3, 2 # left arm ]) # parent index parent_body_index = np.array([ -1,","= {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i,","out_write, nb_max_char): count = 0 with open(in_file,'r') as f: # loop on all","out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix","of body i\\n') out_file.write(' * xgi : absolute position vector of the COM","1: out_file.write(' // -- variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') #","xgj, symb_var, der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if","else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // --","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] =","xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n // jacobian com","0.0, 1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem):","predecessor\\n') out_file.write(' * si : sine of the relative angle before body i\\n')","f: # declaration if len(line.split('// -- variables initialization -- //')) != 1: out_file.write('","c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot')","0 for i in range(0,3): if vector[i] == 0 or vector[i] == 1:","if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n')","om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0]","frame) # of the anchor point of body i # xgi : absolute","axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]])",">= nb_max_char: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') # get","= sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_vector # write symbolic matrix and replace","nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None]","+= der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len == 2:","2, 1, 2, # right leg 2, 1, 3, 2, 1, 2, #","result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len ==","out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write('","flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n')","frame of the current body i\\n') out_file.write(' * Omi : rotational vector from","13 //\\n') out_file.write(' * // //\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' *\\n') out_file.write(' *","0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0],","R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1],","np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3:","the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om,","== 2: # negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {}","two pitch hip rotations\\n') out_file.write(' * inertial frame: located at the origin (waist),","out_file.write(' }\\n\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot',","sp.zeros(3, 3) save_matrix = 9 * [None] for i in range(0,3): for j","from an orientation matrix, compute the roll, pitch, yaw angles (and derivative) def","(center of mass) G_i,\\n') out_file.write(' * expressed in the relative frame of the","+ {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1:","j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac","flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n // joint",": transpose matrix of Ri # xji : jacobian of 'xi' # xgji","= 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n // joint absolute velocities\\n')","parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id]","out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n') for i in","j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute orientation jacobian\\n') out_file.write(' if","[None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] =","-- //\\n') out_file.write('\\n // IMU - rotation matrices\\n') for i in range(0, 3):","x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id]","in the inertial frame)\\n') out_file.write(' * of the anchor point of body i\\n')","get_tilde(om[i]) # x & xp out_file.write('\\n // anchor point absolute positions and velocities\\n')","i\\n') out_file.write(' *\\n') out_file.write(' * xi : absolute position vector (from origin, expressed","len(cur_split) >= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string =","'') # jacobian xg out_file.write('\\n // jacobian com absolute positions\\n') out_file.write(' if (flag_jacob)\\n","body i # (previous body is not always body i-1), expressed in the","# compute the center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names,","R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) #","counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result #","out_file.write(' * // 03 09 //\\n') out_file.write(' * // 04 10 //\\n') out_file.write('","= sin(in_out.theta_waist[1]);\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso =","x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot","in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left foot jacobian\\n')","xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n // jacobian anchor point positions\\n') out_file.write('","between body i and its predecessor # si : sine of the relative","0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return","print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary","len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0,","// jacobian com absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","out_file.write('\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\n') # results out_file.write('\\n // -- Collecting results -- //\\n\\n')","variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index,","sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'),","jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for i in range(1, nb_bodies): for j in","out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 ->","arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] =","get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1: return np.array([[1.0, 0.0,","= x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i]","range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left foot absolute velocity\\n')","waist orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' //","for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right","count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\n') count = 0","* omi : absolute rotational vector of body i\\n') out_file.write(' * Ri :","= {};\\n'.format(i, xp_l[i])) out_file.write('\\n') out_file.write(' // left foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon = 1 ->","= atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] =","previous body # Rdi : rotational matrix between body i and its predecessor","xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =","= {};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n')","xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file,","= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1))","of the previous body # Rdi : rotational matrix between body i and","range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts","= {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot contact points absolute",": jacobian of \\'Ri\\'\\n') out_file.write(' */\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\n{\\n') # compute the center","nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj,","in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac !=","0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0,","left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22]","j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com","derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and","+ R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist)","Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0,","// rotation matrices\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0]","= cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string","return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0 /","== 2: out_file.write(';\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix,","1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print,","0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis,","orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position,","out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables","= nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in","line.split(elem) if len(cut_line) == 2: count += 1 return count # print the","all the joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3)","out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write('","file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file","3) save_matrix = 9 * [None] for i in range(0,3): for j in","0: # epsilon = 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write('","absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write('","3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot contact","* Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\n') for i","1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right","by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print","# ci : cosine of the relative angle before body i # #","= vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] =","R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out,","# # 06 12 # # 07 13 # # # ###################### #","j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left wrist absolute position\\n') for i in","= 3 * [None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\n","xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate","sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i","* (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{}","nb_max_line: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') # print all","j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' //","# writing outputs out_file.write(' // right wrist absolute position\\n') for i in range(0,3):","Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb =","the matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj,","f: cut_line = line.split(elem) if len(cut_line) == 2: count += 1 return count","out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right wrist absolute velocity\\n') for","= print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point def","s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n') for i in range(1,","return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else:","j in range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if","y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in","*\\n') out_file.write(' * \\\\param[in,out] in_out inputs and outputs class\\n') out_file.write(' *\\n') out_file.write(' *","vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name,","in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i]","i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if","* [None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\n // rotation","om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R,","xgji : jacobian of 'xgi' # Rji : jacobian of 'Ri' # return","file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r')","+= Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if","' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save","else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} =","Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')])","and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write('","out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0","derivatives [rad/s]\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\n') # compute the wrists position, velocity","for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j,","0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6]","Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1),","out_file.write(' * (previous body is not always body i-1), expressed in the relative\\n')","if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if","sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] #","(Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file,","out_file.write(' * xji : jacobian of \\'xi\\'\\n') out_file.write(' * xgji : jacobian of","out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') # from","xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first =","i and its predecessor # si : sine of the relative angle before","sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')])","count the number of 'elem' in the file def count_elem(in_file, elem): count =","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] =","{}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1:","flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else:","loop on all the joints for i in range (1, nb_bodies): new_vector =","//\\n\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp,","flag_print = 0 for k in range(0, 3): if xj_print[i][j][k] != None: if","sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'),","3) # loop on all the matrix elements for j in range(0, 9):","velocity\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n //","[None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot +","sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] =","= np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11',","= sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0, 3): for","new_string # write the end of the file def write_file_end(out_file): out_file.write('}\\n') # print","R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] =","sp.Symbol(elem_name) if flag_print: out_file.write('\\n') return new_matrix # save the symbolic vector for print","04 10 //\\n') out_file.write(' * // 05 11 //\\n') out_file.write(' * // 06","out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write(' // right foot contact points","[0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis, direct,","f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute position\\n') for i","xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write('","new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name)","# other else: print('Error: {} * counted , only implemented for 0 or","in_out.{}[2] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles'","r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4","in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] *","i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write(' // global com absolute","= x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] =","9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None]","in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot contact","new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a rotation","sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'),","R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part):","[None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot =","sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')])","the previous body\\n') out_file.write(' * Rdi : rotational matrix between body i and","out_file.write(' }\\n\\n') # from an orientation matrix, compute the roll, pitch, yaw angles","sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if vector[i] == 0","out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j == 2: out_file.write(';\\n') else:","count >= nb_max_line: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') #","an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of","print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for","nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i]","rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1:","terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0","left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None]","count = 0; with open(in_file, 'r') as f: # loop on all the","feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj,","# xi : absolute position vector (from origin, expressed in the inertial frame)","if it has a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value):","cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) == 2:","previous body\\n') out_file.write(' * DGi : position vector from the anchor point of","== 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif","COMAN model\\n') out_file.write(' */\\n\\n') out_file.write('// joints enumeration\\n') out_file.write('enum {') count = 0 for","if i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write(' // global com","= 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j],","= {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic","for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '')","nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts):","Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] =","= inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part,","c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot","{};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute orientation jacobian\\n') out_file.write('","1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j]","om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i]","Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0,","'xgi' # Rji : jacobian of 'Ri' # return true if it is","# jacobian x out_file.write('\\n // jacobian anchor point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n')","computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T *","x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id]","xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 *","elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))","1, 2, 3, 4, 5, # right leg 0, 7, 8, 9, 10,","out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot","i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' // left foot","for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\n') out_file.write('","the anchor point of the previous body to the current body i #","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]),","waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') ,","sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3]","flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if","out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}]","'{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\"([A-Z][^A-Z]*)\", new_string)) new_string =","for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' +","writing outputs out_file.write(' // right wrist absolute position\\n') for i in range(0,3): out_file.write('","{\\n') flag_first = 0 for i in range(0, nb_contacts): for j in range","the derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var):","out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write('","wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n')","nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3)","for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j","in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU -","= get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i],","xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0]","for each joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, #","point of the previous body to the current body i # (previous body","# trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15]","'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name =","= 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' //","remove temporary file os.remove(in_temp) # main script # rotation axis for each joint","//\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy temporary file out_file.write(line) out_file.close() # remove","c = int(value.split('_')[1][1]) return True, a, b, c except: return False, -1, -1,","the file def write_file_end(out_file): out_file.write('}\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix):","sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for j in range(0,3):","sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] =","b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if","of body i\\n') out_file.write(' * xpi : derivative of xi\\n') out_file.write(' * xgpi","in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // right foot contact points","file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file","result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative","# Omi : rotational vector from the previous body to the current body","com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for i","R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2]","= symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) #","variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\n')","sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')])","{}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char:","Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'),","= write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian","xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb *","'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b","Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0,","min, max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if","frame: located at the origin (waist), but aligned with the ground (info from","x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T *","lines for line in f: cut_line = line.split(elem) if len(cut_line) == 2: count","== 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0],","it is a float def isInt(value): try: int(value) return True except: return False","# DGi : position vector from the anchor point of body i to","position, velocity and orientation\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\n')","the waist, middle point between the two pitch hip rotations\\n') out_file.write(' * inertial","before body i\\n') out_file.write(' * ci : cosine of the relative angle before","= cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot =","# loop on all the matrix elements for j in range(0, 9): new_matrix[j]","print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies,","symbolic vector and replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name,","= sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for j in","for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\n'.format(i, R_r_foot[i])) out_file.write('\\n') for i","# get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]],","# left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0])","R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs","k, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot contact points absolute position\\n') for","10 //\\n') out_file.write(' * // 05 11 //\\n') out_file.write(' * // 06 12","anchor point of the previous body to the current body i # (previous","wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj,","xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] =","import numpy as np import sympy as sp import re import os ######################","# 04 10 # # 05 11 # # 06 12 # #","return result # write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*!","the anchor point of body i to its COM (center of mass) G_i,","joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp,","Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0,","xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1","of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list","range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in","relative # frame of the previous body # Rdi : rotational matrix between","in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left","omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\n'.format(omega_out, body_part, body_part, omega_in[0],","= 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k],","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21]","out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out inputs and","{}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\n') count","else: out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') # from an","else: out_file.write(' }\\n\\n') # from an orientation matrix, compute the roll, pitch, yaw","left foot absolute velocity\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\n'.format(i, xp_l[i]))","range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j == 2: out_file.write(';\\n')","i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\n'.format(i, R_l_elb[i])) out_file.write('\\n') out_file.write(' // right","inputs and outputs class\\n') out_file.write(' *\\n') out_file.write(' * computation of:\\n') out_file.write(' * COM","flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j],","matrix and replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name):","out_file.write('\\n // com absolute positions and velocities\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None]","sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')])","'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint names joint_id_names =","# # 05 11 # # 06 12 # # 07 13 #","count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count","// waist orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write('","nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i]","angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\n') out_file.write(' // left foot orientation matrix","file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables","wrist absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp,","xgi : absolute position vector of the COM G_i of body i\\n') out_file.write('","with the ground (info from IMU)\\n') out_file.write(' *\\n') out_file.write(' * Di : position",": transpose matrix of Ri\\n') out_file.write(' * xji : jacobian of \\'xi\\'\\n') out_file.write('","len(elem_split) if cur_len == 1: # positive neg_flag = 0 pos_str = elem_split[0]","else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first:","out_file.write(' // left foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","{\\n') flag_first = 0 for i in range(0, nb_bodies): for j in range(1,","0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if","= '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the file def","count >= 6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*!","[-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if","# left arm ]) # parent index parent_body_index = np.array([ -1, # waist","{};\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot contact points jacobian\\n') out_file.write(' if","for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint","3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return","i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n')","Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')])","on all the lines for line in f: cut_line_1 = line.split(elem) cut_line_2 =","'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a","all the lines for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split('","Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0,","sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7]","angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: #","Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'),","arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] =","[0.0], [elem]]) else: return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0],","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot absolute position\\n') for i in","sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts *","R_matrix, index): # loop on all the joints for i in range (1,","// joint sines\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i]))","cos(in_out.theta_Rfoot[2]);\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\n')","= len(elem_split) if cur_len == 1: # positive neg_flag = 0 pos_str =","nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com jacobian\\n') out_file.write('","new_matrix # save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector","= line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if","range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj,","angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso orientation angle","Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1 elif","'M_27', 'M_28', 'M_29' # left arm ]) # joint names joint_id_names = np.array(['0',","omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2]","sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')])","= 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*',''))","= sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'),","Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot","11 //\\n') out_file.write(' * // 06 12 //\\n') out_file.write(' * // 07 13","R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0]","[None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\n // rotation matrices\\n')","new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1))","'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26',","loop on all the matrix elements for j in range(0, 9): new_matrix[j] =","'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left","1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '')","# xji : jacobian of 'xi' # xgji : jacobian of 'xgi' #","# # xi : absolute position vector (from origin, expressed in the inertial",", sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'),","i\\n') out_file.write(' * ci : cosine of the relative angle before body i\\n')","symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector,","for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] =","the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1)","out_file.write(' // left wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","== nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\n') else: out_file.write('0.0;\\n') if i != nb_bodies-1: out_file.write('\\n') else:","out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write(' *\\n')","1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left","anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on","for i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0","2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]])","0 for i in range(0,3): for j in range(0,3): if matrix[i,j] == 0","= sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file,","Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i]","xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic","expressed in the relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write(' *","# print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r')","inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\n\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\n')","nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\n'.format(i+1, joint_id_names[i])) out_file.write('\\n // joint relative velocities\\n') for i","arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] =","'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id'","# 06 12 # # 07 13 # # # ###################### # #","*\\n') out_file.write(' * Di : position vector from the anchor point of the","= sin(in_out.theta_Lfoot[2]);\\n\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\n {\\n') out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot","left wrist absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\n'.format(i, x_l[i]))","'') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1,","body\\n') out_file.write(' * Rdi : rotational matrix between body i and its predecessor\\n')","-1, -1 # return true if it has a shape 'x%a_%b' (indexes %a,","0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19,","= np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg","flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write('","'': term_list.pop(0) result = 0 # loop on all terms for cur_term in","x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1),","11 # # 06 12 # # 07 13 # # # ######################","of the previous body\\n') out_file.write(' * DGi : position vector from the anchor","for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9","Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\n // jacobian rotation matrices\\n') out_file.write('","all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f:","Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') ,","# Rdi : rotational matrix between body i and its predecessor # si","new_vector = sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if vector[i]","* feet position, velocity and orientation\\n') out_file.write(' * waist and torso orientaion angles",", sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') ,","count >= nb_max_char: out_write.write(';\\n') count = 0 if count != 0: out_write.write(';\\n') #","R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: #","s_z_waist = sin(in_out.theta_waist[2]);\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\n\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\n {\\n')","xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1),","R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'),","3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts *","out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\n') # omega out_file.write('\\n // joint absolute velocities\\n') Om =","else: out_file.write('0.0;\\n') out_file.write('\\n') out_file.write(' // global com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for","om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\n // anchor","foot orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\n') out_file.write(' // left","x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min","xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1","om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] =","Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None]","derivative of xgi\\n') out_file.write(' * omi : absolute rotational vector of body i\\n')","= y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts):","== 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count","= sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y","xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts =","in range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k]","neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag","Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0])","1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write('","of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1:","orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id,","wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005,","declaration if len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' // --","new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for i in range(0,3):","# 07 13 # # # ###################### # # origin: in the waist,","# xgji : jacobian of 'xgi' # Rji : jacobian of 'Ri' #","nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for i in","in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU - angles velocity\\n') for i in range(0,","def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return","get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l =","def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\n\\n // -- symbolic","count += 1 return count # print the declaration of an element def","// global com jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') for i in range(1, nb_bodies):","write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x","in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right wrist absolute velocity\\n') for i",": rotational matrix between body i and its predecessor\\n') out_file.write(' * si :","new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper())","velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,","angle before body i\\n') out_file.write(' * ci : cosine of the relative angle","out_file.write(' // left foot absolute position\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] =","out_file.write('\\n') out_file.write(' // right foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else:","xj, xgj, der_var) # one product elif cur_len == 2: result += der_elem(cur_term_split[0],","(indexes %a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b =","- s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} *","06 12 //\\n') out_file.write(' * // 07 13 //\\n') out_file.write(' * // //\\n')","= {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right wrist absolute velocity\\n') for i in","Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations","Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0,","angle before body i\\n') out_file.write(' *\\n') out_file.write(' * xi : absolute position vector","cut_line = line.split(elem) if len(cut_line) == 2: count += 1 return count #","= get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] =","absolute position vector of the COM G_i of body i\\n') out_file.write(' * xpi","R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print","{} instead of 1 or 2 in negative detection !'.format(cur_len)) exit() # compute","new_matrix, save_matrix # write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj,","leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8')","if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1))","computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out] in_out inputs and outputs class\\n') out_file.write(' *\\n')","c_z_{}*{} - s_z_{}*{};\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{}","y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0,","j, x_l_cont[i][j])) out_file.write('\\n') out_file.write(' // left foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n","matrices out_file.write('\\n // rotation matrices\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd =","= ' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector #","range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xj_print[i][j][k] !=","sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')])","## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0,","R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1]))","derivative of xi # xgpi : derivative of xgi # omi : absolute","= sp.zeros(3, 3) for i in range(0, 3): for j in range(0, 3):","= in_out.IMU_Orientation[{}];\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\n // IMU - angles velocity\\n') for i in","def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles","write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n // jacobian anchor point positions\\n')","variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\n\\n') # copy temporary file out_file.write(line)","arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] =","waist orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\n') out_file.write(' // torso","# xg & xgp out_file.write('\\n // com absolute positions and velocities\\n') xg =","'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' ,","om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables","transpose matrix of Ri\\n') out_file.write(' * xji : jacobian of \\'xi\\'\\n') out_file.write(' *","= write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n // jacobian anchor point","== max: out_file.write(';\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names):","// 02 08 //\\n') out_file.write(' * // 03 09 //\\n') out_file.write(' * //","in the relative\\n') out_file.write(' * frame of the previous body\\n') out_file.write(' * DGi","om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb","'') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id","Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'),","x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): #","variables initialization -- //')) != 1: out_file.write(' // -- variables declaration -- //\\n\\n')","'theta_waist', R[waist_id], 1) out_file.write('\\n') out_file.write(' // torso orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file,","'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id'","nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None]","= write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n // jacobian com absolute","relative frame of the current body i # Omi : rotational vector from","apply negative if neg_flag: result = -result return result # compute the derivative","* s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part,","# computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T","double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >=","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] =","range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i]","= nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies):","<filename>symbolicR/python/forward_kin.py import numpy as np import sympy as sp import re import os","rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result +=","symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): #","position\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\n'.format(i, x_l[i])) out_file.write('\\n') out_file.write(' //","anchor point positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in","= write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\n","R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file,","# print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i","out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\n\\n') out_file.write('","R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i]","sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')])","for 0 or 1 !'.format(cur_len-1)) exit() return result # write the beginning of","sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', #","open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on all the lines","08 # # 03 09 # # 04 10 # # 05 11","is not always body i-1), expressed in the relative # frame of the","range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\n'.format(i, x_r[i])) out_file.write('\\n') out_file.write(' // right wrist absolute velocity\\n')","nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3):","out_file.write('\\n') out_file.write(' // torso orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1)","in_out.{}[0] = atan2(-{}, -{});\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\n'.format(angle_name,","Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg","nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write(' + ') out_file.write(' //","0 if count != 0: out_write.write(';\\n') # print all declarations def print_all_declaration(in_file, out_write,","sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e]","= 0 for i in range(0, nb_contacts): for j in range (1, nb_bodies):","i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right","middle point between the two pitch hip rotations\\n') out_file.write(' * inertial frame: located","derivative of xgi # omi : absolute rotational vector of body i #","rot_axis, parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close() #",": absolute rotational matrix\\n') out_file.write(' * Rti : transpose matrix of Ri\\n') out_file.write('","Ri # xji : jacobian of 'xi' # xgji : jacobian of 'xgi'","if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1","rotational matrix between body i and its predecessor # si : sine of","out_file.write(' * // 17 16 21 //\\n') out_file.write(' * // 18 15 22","2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i])","om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot","not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('","6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main","= get_tilde(om[i]) # x & xp out_file.write('\\n // anchor point absolute positions and","xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts","== 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative","if i == 2 and j == 2: out_file.write(';\\n') else: out_file.write(', ') #","(for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms","+= xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return result #","import os ###################### # # # 17 16 21 # # 18 15","* // 17 16 21 //\\n') out_file.write(' * // 18 15 22 //\\n')","= 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i]","print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 *","DGi : position vector from the anchor point of body i to its","xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj =","sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'),","body i and its predecessor\\n') out_file.write(' * si : sine of the relative","'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\n') out_file.write('","{\\n') flag_first = 0 for i in range (1, nb_bodies): flag_print = 0","in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') # get a string for the","right foot contact points jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for","[None] for i in range(0,3): for j in range(0,3): if matrix[i,j] == 0","{};\\n'.format(i, xp_r[i])) out_file.write('\\n') out_file.write(' // right wrist jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first","xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print","for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\n\\n') else: out_file.write('","'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\n //","R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1]","= sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16]","Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i","absolute position\\n') for i in range(0, nb_contacts): for j in range(0, 3): out_file.write('","xi : absolute position vector (from origin, expressed in the inertial frame)\\n') out_file.write('","out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\n') # xg & xgp out_file.write('\\n // com absolute positions and","[elem]]) else: return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]])","origin, expressed in the inertial frame) # of the anchor point of body","foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range","write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '')","2, 1, 3, 2, 1, 2, # left leg 1, 2, 3, #",", 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' ,","out_file.write(' * ////////////////////////\\n') out_file.write(' * // //\\n') out_file.write(' * // 17 16 21","= sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8]","1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]])","= xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot","1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\n'.format(i, get_string_enum(joint_id_names[j]),","Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0,","= '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name)","= xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot =","current body i \\n') out_file.write(' * (previous body is not always body i-1),","new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the file","// joint relative velocities\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\n'.format(i+1,","all the vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj,","vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j],","cur_jac)) out_file.write(' }\\n\\n') # get a string for the enumeration of joints def","= line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if","out_file.write(' in_out.Lfoot_or[{}] = {};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot absolute orientation jacobian\\n')","out_file.write(' return;\\n }\\n\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\n') out_file.write(' inv_c_y_Lfoot = 1.0","negative if neg_flag: result = -result return result # compute the derivative of","0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis == 1: return","0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector","-r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r","xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for","sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')])","new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return","orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i in range","= x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb =","Rdi : rotational matrix between body i and its predecessor # si :","= in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n') for i in range(1, nb_bodies): out_file.write('","xj, xgj, symb_var, der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+')","and derivatives\\n') out_file.write(' *\\n') out_file.write(' * ////////////////////////\\n') out_file.write(' * // //\\n') out_file.write(' *","[rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\n') out_file.write(' c_y_torso =","out_file.write('\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') #","xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i],","'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id'","= R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1)","-1, -1 # count the number of 'elem' in the file def count_elem(in_file,","out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count","if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\n","15, 20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor","out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') # get a string for","flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] =","Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1]","in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\n'.format(i, j,","return new_string # write the end of the file def write_file_end(out_file): out_file.write('}\\n') #","Ri\\n') out_file.write(' * xji : jacobian of \\'xi\\'\\n') out_file.write(' * xgji : jacobian","cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not","out_file.write('\\n // joint cosines\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\n'.format(i+1,","else: print('Error: {} * counted , only implemented for 0 or 1 !'.format(cur_len-1))","// right foot jacobian\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first = 0 for i","* xji : jacobian of \\'xi\\'\\n') out_file.write(' * xgji : jacobian of \\'xgi\\'\\n')","for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j]","xg out_file.write('\\n // jacobian com absolute positions\\n') out_file.write(' if (flag_jacob)\\n {\\n') flag_first =","G_i,\\n') out_file.write(' * expressed in the relative frame of the current body i\\n')","out_file.write('// joints enumeration\\n') out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)):","xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] =","flag_first = 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\n')","0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0,","x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r","der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string) elem_str = elem_str.replace('-","nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None]","'') # jacobian x out_file.write('\\n // jacobian anchor point positions\\n') out_file.write(' if (flag_jacob)\\n","vector of the COM G_i of body i # xpi : derivative of","# expressed in the relative frame of the current body i # Omi","out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\n\\n') out_file.write('/*! \\\\brief main kinematics computation\\n') out_file.write(' *\\n') out_file.write(' * \\\\param[in,out]","{};\\n'.format(i, R_l_foot[i])) out_file.write('\\n') out_file.write(' // right foot absolute orientation jacobian\\n') out_file.write(' if (flag_jacob)\\n","matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all","return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([])","x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right","for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac","// 19 14 23 //\\n') out_file.write(' * // 20 01 24 //\\n') out_file.write('","# variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization --","2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else:","0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0,","nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the vector elements for","if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f: #","i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for","x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for","G_i of body i\\n') out_file.write(' * xpi : derivative of xi\\n') out_file.write(' *","matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1: return","nb_bodies-1: out_file.write('\\n') else: out_file.write(' }\\n\\n') # from an orientation matrix, compute the roll,","'.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j] !=","= 1 elif not flag_print: flag_print = 1 out_file.write('\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]),","of \\'xi\\'\\n') out_file.write(' * xgji : jacobian of \\'xgi\\'\\n') out_file.write(' * Rji :","// torso orientation matrix as angles [rad]\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\n') out_file.write('","////////////////////////\\n') out_file.write(' * // //\\n') out_file.write(' * // 17 16 21 //\\n') out_file.write('","omega_{} = in_out.IMU_Angular_Rate[{}];\\n'.format(i+1, i)) out_file.write('\\n // joint cosines\\n') for i in range(1, nb_bodies):","xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\n","1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(',","arm 2, 1, 3, 2 # left arm ]) # parent index parent_body_index","R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] =","= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20]","R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1],","= r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation","= get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l","i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None]","out_file.write('\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\n\\n') out_file.write(' // left foot","out_file.write(';\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min, max):","//')) != 1: out_file.write(' // -- variables declaration -- //\\n\\n') print_all_declaration(in_temp, out_file, 100)","out_file.write(' * \\\\param[in,out] in_out inputs and outputs class\\n') out_file.write(' *\\n') out_file.write(' * computation","Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs","out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if","in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0:","== 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif","vector of the COM G_i of body i\\n') out_file.write(' * xpi : derivative"],"string":"[\n \"rotations # inertial frame: located at the origin (waist), but aligned with the\",\n \"[None] for i in range(0,3): if vector[i] == 0 or vector[i] == 1:\",\n \"= nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1]\",\n \"{}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write('\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get a string for the enumeration of\",\n \"# write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj,\",\n \"2, # right arm 2, 1, 3, 2 # left arm ]) #\",\n \"+ R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\\\n')\",\n \"range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n') for\",\n \"= nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] =\",\n \"leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm\",\n \"end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for i in\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot contact points absolute position\\\\n') for\",\n \"5, # right leg 0, 7, 8, 9, 10, 11, # left leg\",\n \"17, 18, # right arm 15, 20, 21, 22 # left arm ])\",\n \"mass) position and velocity\\\\n') out_file.write(' * feet position, velocity and orientation\\\\n') out_file.write(' *\",\n \"vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector =\",\n \"joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close()\",\n \"'_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M):\",\n \"0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'),\",\n \"nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for i in\",\n \"for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')])\",\n \"# # 17 16 21 # # 18 15 22 # # 19\",\n \": derivative of xgi # omi : absolute rotational vector of body i\",\n \"derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): #\",\n \"xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i,\",\n \"1.0]]) else: return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0],\",\n \"parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id]\",\n \"x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] =\",\n \"left foot contact points absolute position\\\\n') for i in range(0, nb_contacts): for j\",\n \"initialization -- //')) != 1: out_file.write(' // -- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp,\",\n \"= len(cur_term_split) # no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj,\",\n \"== 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif\",\n \"for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if\",\n \"[rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot orientation angle derivatives\",\n \"+ om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' ,\",\n \"(flag_jacob)\\\\n {\\\\n') for i in range(1, nb_bodies): for j in range(0, 3): out_file.write('\",\n \"1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count +=\",\n \"compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon):\",\n \"// left foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i,\",\n \"get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1,\",\n \"= nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] *\",\n \"def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0\",\n \"as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left foot orientation\",\n \"out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * // 17\",\n \"out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') # get tilde matrix\",\n \"Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\\\n') for i in\",\n \"0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1))\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg & xgp out_file.write('\\\\n //\",\n \"{};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix # save the\",\n \"sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont\",\n \"in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in range(0,\",\n \"and its predecessor # si : sine of the relative angle before body\",\n \"in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute orientation\",\n \"return true if it has a shape 'x%a_%b' (indexes %a, %b also returned)\",\n \"angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso orientation matrix as\",\n \"or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name =\",\n \"xgj, xj_print, x_vector, index): # loop on all the joints for i in\",\n \"i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot\",\n \"derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q))\",\n \"term_list[0] == '': term_list.pop(0) result = 0 # loop on all terms for\",\n \"all the joints for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1)\",\n \"point of body i to its COM (center of mass) G_i,\\\\n') out_file.write(' *\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute position\\\\n') for i in range(0,3):\",\n \"der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {}\",\n \"trunk 15, 16, 17, 18, # right arm 15, 20, 21, 22 #\",\n \"// global com absolute velocity\\\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] =\",\n \"joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' *\",\n \"right arm 2, 1, 3, 2 # left arm ]) # parent index\",\n \"nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0]\",\n \"R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb\",\n \"body to the current body i # (previous body is not always body\",\n \"= atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2],\",\n \"get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T\",\n \"Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close() # output file\",\n \"joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id',\",\n \"as f: # loop on all the lines for line in f: cut_line\",\n \"angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the wrists position,\",\n \"= sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] =\",\n \"for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count =\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg & xgp out_file.write('\\\\n\",\n \"# # 20 01 24 # # 02 08 # # 03 09\",\n \"out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs\",\n \"cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis def\",\n \"// //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin: in the waist,\",\n \"= 3 * [None] for i in range(0,3): if vector[i] == 0 or\",\n \"Ri : absolute rotational matrix\\\\n') out_file.write(' * Rti : transpose matrix of Ri\\\\n')\",\n \"xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T\",\n \"rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply\",\n \"= nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None]\",\n \"matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix # save the symbolic\",\n \"new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else:\",\n \"position vector from the anchor point of the previous body to the current\",\n \"initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt,\",\n \"i-1), expressed in the relative # frame of the previous body # DGi\",\n \"1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation\",\n \"two pitch hip rotations # inertial frame: located at the origin (waist), but\",\n \"si : sine of the relative angle before body i # ci :\",\n \"body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1],\",\n \"in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute orientation jacobian\\\\n') out_file.write('\",\n \"i # Omi : rotational vector from the previous body to the current\",\n \"'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an\",\n \"'') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n // jacobian\",\n \"j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] =\",\n \"returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a,\",\n \"mass) G_i, # expressed in the relative frame of the current body i\",\n \"range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute\",\n \"for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i]\",\n \"i in range (1, nb_bodies): flag_print = 0 for j in range(0, 3):\",\n \"right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18]\",\n \"1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0],\",\n \"= elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive neg_flag =\",\n \"-{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in,\",\n \"R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23]\",\n \"24 # # 02 08 # # 03 09 # # 04 10\",\n \"left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg,\",\n \"= 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n')\",\n \"if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] =\",\n \"omi : absolute rotational vector of body i # Ri : absolute rotational\",\n \"1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n')\",\n \"i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' //\",\n \"declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in range(min,\",\n \"in term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no\",\n \"xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0],\",\n \"'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n //\",\n \"= sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')])\",\n \"c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"= sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0,\",\n \"if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range (1, nb_bodies): flag_print\",\n \"of 'xgi' # Rji : jacobian of 'Ri' # return true if it\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute position\\\\n') for i in range(0,3):\",\n \"*\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * // 17 16\",\n \"write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\\\n //\",\n \"body is not always body i-1), expressed in the relative # frame of\",\n \"try: int(value) return True except: return False # return true if it has\",\n \"= int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except: return False,\",\n \"in_out inputs and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n') out_file.write(' *\",\n \"get vector axis def get_vector_axis(axis, direct, elem): if direct: if axis == 1:\",\n \"com absolute velocity\\\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first\",\n \"k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k],\",\n \"x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string,\",\n \"matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {}\",\n \"(1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the vector elements\",\n \"3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative of\",\n \"in range(1, nb_bodies): flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k]\",\n \"# left leg 1, 2, 3, # trunk 2, 1, 3, 2, #\",\n \"velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ')\",\n \"= {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute orientation jacobian\\\\n') out_file.write(' if\",\n \"{};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist absolute velocity\\\\n') for i in range(0,3):\",\n \"'./{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) #\",\n \"= elem_split[1] else: print('Error: {} instead of 1 or 2 in negative detection\",\n \"08 //\\\\n') out_file.write(' * // 03 09 //\\\\n') out_file.write(' * // 04 10\",\n \"// right wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i,\",\n \"compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var,\",\n \"def get_vector_axis(axis, direct, elem): if direct: if axis == 1: return np.array([[elem], [0.0],\",\n \"positions and velocities\\\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] *\",\n \"13 # # # ###################### # # origin: in the waist, middle point\",\n \"xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis,\",\n \"absolute rotational matrix # Rti : transpose matrix of Ri # xji :\",\n \"out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU - angles velocity\\\\n') for\",\n \"1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count +=\",\n \"R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file,\",\n \"# one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj,\",\n \"sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) #\",\n \"= r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] =\",\n \"# loop on all terms for cur_term in term_list: # detect products cur_term_split\",\n \"out_file.write(' * Rti : transpose matrix of Ri\\\\n') out_file.write(' * xji : jacobian\",\n \"9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write('\",\n \"jacobian rotation matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0:\",\n \"return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if direct: if\",\n \"the file def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f:\",\n \"out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on all\",\n \"sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'),\",\n \", 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id'\",\n \"xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error:\",\n \"xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1 elif\",\n \"= sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0,\",\n \"xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0],\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] =\",\n \"for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] =\",\n \"{};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i]))\",\n \"1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if\",\n \"j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac !=\",\n \"out_file.write(' * Di : position vector from the anchor point of the previous\",\n \"0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None]\",\n \"# omi : absolute rotational vector of body i # Ri : absolute\",\n \"and velocity\\\\n') out_file.write(' * feet position, velocity and orientation\\\\n') out_file.write(' * waist and\",\n \"\\\\\\\\param[in,out] in_out inputs and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n') out_file.write('\",\n \"len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count\",\n \">= 6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief\",\n \"out_file.write(' * // //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin: in\",\n \"i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all\",\n \"(info from IMU) # # Di : position vector from the anchor point\",\n \"absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0,\",\n \"xj, xgj, xgj_print, x_vector, index): # loop on all the joints for i\",\n \"body i\\\\n') out_file.write(' * xgi : absolute position vector of the COM G_i\",\n \"# trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29'\",\n \"i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in\",\n \"== 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1,\",\n \"in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ')\",\n \"= 9 * [None] for i in range(0,3): for j in range(0,3): if\",\n \"nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\\\n') out_file.write('\\\\n // IMU -\",\n \"Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0,\",\n \"the relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write(' * DGi :\",\n \"-- //')) != 1: out_file.write(' // -- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file,\",\n \"write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n // jacobian com absolute positions\\\\n')\",\n \"j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac !=\",\n \"def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b\",\n \"R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the\",\n \"-> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5],\",\n \"R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1]))\",\n \"for i in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} =\",\n \"[0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine,\",\n \"out_file.write(' // feet absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] =\",\n \"out_file.write(' * xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian of\",\n \"23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic\",\n \"flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j],\",\n \"of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{}\",\n \"if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print\",\n \"- angles velocity\\\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i))\",\n \"\\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center of mass position\",\n \"i == 2 and j == 2: out_file.write(';\\\\n') else: out_file.write(', ') # print\",\n \"'') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n // jacobian\",\n \"xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id]\",\n \"sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) #\",\n \"else: out_file.write(' + ') out_file.write(' // global com absolute position\\\\n') for i in\",\n \"xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\\\n // rotation matrices\\\\n') R\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not\",\n \"in range (1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac =\",\n \"{}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out\",\n \"to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q',''))\",\n \"sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')])\",\n \"generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg,\",\n \"xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if\",\n \"range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of\",\n \"|| (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write('\",\n \"i # (previous body is not always body i-1), expressed in the relative\",\n \"out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from an orientation\",\n \"in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{}\",\n \"16 21 //\\\\n') out_file.write(' * // 18 15 22 //\\\\n') out_file.write(' * //\",\n \"# compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x,\",\n \"* Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing\",\n \"'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print =\",\n \"c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot\",\n \"for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first\",\n \"R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot\",\n \"axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis\",\n \"out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from an orientation matrix, compute the roll, pitch,\",\n \"derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): #\",\n \"x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max\",\n \"Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i]\",\n \"out_file.write(' * origin: in the waist, middle point between the two pitch hip\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n')\",\n \"'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name,\",\n \"out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\\\n') out_file.write('\\\\n // IMU\",\n \"nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for\",\n \"sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right\",\n \"temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script #\",\n \"sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')])\",\n \"body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write('\",\n \"= int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len\",\n \"# rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result\",\n \"if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] =\",\n \"'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis,\",\n \"relative angle before body i\\\\n') out_file.write(' * ci : cosine of the relative\",\n \"arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name\",\n \"in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute orientation\",\n \"out_file.write(' * // 04 10 //\\\\n') out_file.write(' * // 05 11 //\\\\n') out_file.write('\",\n \"vector and replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name):\",\n \"out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics computation for the COMAN\",\n \"not always body i-1), expressed in the relative\\\\n') out_file.write(' * frame of the\",\n \"{};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot absolute velocity\\\\n') for i in range(0,3):\",\n \"= r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb)\",\n \"range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute\",\n \"right wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"= sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if vector[i] ==\",\n \"nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1)\",\n \"Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') ,\",\n \"1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n // -- Collecting results --\",\n \"\\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian\",\n \"= int(value.split('_')[1][1]) return True, a, b, c except: return False, -1, -1, -1\",\n \"symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \": absolute position vector of the COM G_i of body i\\\\n') out_file.write(' *\",\n \"= x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T\",\n \"range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for k in\",\n \"of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string =\",\n \"'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28',\",\n \"flag_print: out_file.write('\\\\n') return new_vector # write symbolic matrix and replace symbolic variable by\",\n \"sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] =\",\n \"sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot =\",\n \"R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l\",\n \"# right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')])\",\n \"also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c =\",\n \", only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result # write\",\n \"// left wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i,\",\n \"sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')])\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \"out_write.write(';\\\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with\",\n \"except: return False # return true if it has a shape 'R%a_%b%c' (indexes\",\n \"point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on all\",\n \"in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac !=\",\n \"k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac\",\n \"if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip())\",\n \"absolute rotational vector of body i # Ri : absolute rotational matrix #\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not\",\n \"Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i]\",\n \"3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0,\",\n \"if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(',\",\n \"body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of\",\n \"1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0]\",\n \"else: new_string = cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for\",\n \"for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac\",\n \"out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write('\",\n \"= sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist\",\n \"start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for i\",\n \"= xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb =\",\n \"elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] =\",\n \"xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i],\",\n \"write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '')\",\n \"sines\\\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n //\",\n \"result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag,\",\n \"in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot absolute velocity\\\\n') for i\",\n \"0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] =\",\n \"om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3,\",\n \"b = int(value.split('_')[1]) return True, a, b except: return False, -1, -1 #\",\n \"out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1]))\",\n \"= 0 with open(in_file,'r') as f: # loop on all the lines for\",\n \"end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for i in\",\n \"xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i]\",\n \"14 23 //\\\\n') out_file.write(' * // 20 01 24 //\\\\n') out_file.write(' * //\",\n \"0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write('\",\n \"sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')])\",\n \"# xpi : derivative of xi # xgpi : derivative of xgi #\",\n \"== 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1,\",\n \"on all the joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3,\",\n \"'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' ,\",\n \"sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name,\",\n \"the relative frame of the current body i # Omi : rotational vector\",\n \"if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] #\",\n \"Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] =\",\n \"# sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag,\",\n \"out_file.write(' // right foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] =\",\n \"{} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix #\",\n \"range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3,\",\n \"out_file.write('\\\\n') out_file.write(' // global com absolute velocity\\\\n') for i in range(0, 3): out_file.write('\",\n \"on all the vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj,\",\n \"xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id =\",\n \"## COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')])\",\n \"out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range (1, nb_bodies):\",\n \"# waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right\",\n \"to the current body i # (previous body is not always body i-1),\",\n \"else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n')\",\n \"1) out_file.write('\\\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot',\",\n \"of the anchor point of body i\\\\n') out_file.write(' * xgi : absolute position\",\n \"velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"out_file.write('\\\\n') return new_vector # write symbolic matrix and replace symbolic variable by its\",\n \"j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_',\",\n \": derivative of xi\\\\n') out_file.write(' * xgpi : derivative of xgi\\\\n') out_file.write(' *\",\n \"b, c except: return False, -1, -1, -1 # return true if it\",\n \"= '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j]\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"out_write.write(';\\\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0,\",\n \"* COM (center of mass) position and velocity\\\\n') out_file.write(' * feet position, velocity\",\n \"in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j,\",\n \"import re import os ###################### # # # 17 16 21 # #\",\n \"elif cur_len == 2: # negative neg_flag = 1 pos_str = elem_split[1] else:\",\n \"= cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot =\",\n \"out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies):\",\n \"3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n') for i in\",\n \"relative velocities\\\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) #\",\n \"out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector\",\n \"= sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i\",\n \"= R[i].T # jacobian rotation matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write(' if\",\n \"Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_',\",\n \"the waist, middle point between the two pitch hip rotations # inertial frame:\",\n \"elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as\",\n \"// 17 16 21 //\\\\n') out_file.write(' * // 18 15 22 //\\\\n') out_file.write('\",\n \"in the waist, middle point between the two pitch hip rotations # inertial\",\n \"out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get a string\",\n \"* expressed in the relative frame of the current body i\\\\n') out_file.write(' *\",\n \"xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] =\",\n \"nb_bodies): flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k] != None:\",\n \"2, 1, 3, 2, 1, 2, # right leg 2, 1, 3, 2,\",\n \"i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist\",\n \"in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for\",\n \"of 'Ri' # return true if it is a float def isInt(value): try:\",\n \"None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} =\",\n \"= sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j]\",\n \"xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] =\",\n \"save_matrix # write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj,\",\n \"om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1)\",\n \"inertial frame: located at the origin (waist), but aligned with the ground (info\",\n \"Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies,\",\n \"Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point\",\n \"= x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i]\",\n \"{\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot =\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] =\",\n \"open(in_file,'r') as f: # loop on all the lines for line in f:\",\n \"joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,\",\n \"out_file.write('\\\\n\\\\n // -- symbolic computation -- //\\\\n') # Rj, xj, xgj and xgj\",\n \"[elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([])\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n')\",\n \"Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'),\",\n \"R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian\",\n \"write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all the joints\",\n \"current body i # Omi : rotational vector from the previous body to\",\n \"matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j],\",\n \"= -result return result # compute the derivative of an expression (for jacobian)\",\n \"= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k]))\",\n \"out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center of mass position and\",\n \"0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] =\",\n \"= x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot\",\n \"1: count = 0 with open(in_file,'r') as f: # loop on all the\",\n \"range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0:\",\n \"der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len == 2: result\",\n \"range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else:\",\n \"range (1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj,\",\n \"matrix of Ri\\\\n') out_file.write(' * xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji\",\n \"the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2:\",\n \"\\\\\\\\brief forward kinematics computation for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n')\",\n \"3) for i in range(0, 3): for j in range(0, 3): R[0][i,j] =\",\n \"with open(in_file, 'r') as f: # loop on all the lines for line\",\n \"relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write(' * Rdi : rotational\",\n \"{};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute velocity\\\\n') for i in range(0,3):\",\n \"range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print\",\n \"flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write('\",\n \": sine of the relative angle before body i # ci : cosine\",\n \"declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line) out_file.close()\",\n \"Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')])\",\n \"= atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def\",\n \"'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' ,\",\n \"(from origin, expressed in the inertial frame) # of the anchor point of\",\n \"for i in range (1, nb_bodies): flag_print = 0 for j in range(0,9):\",\n \"# left leg 0, 13, 14, # trunk 15, 16, 17, 18, #\",\n \"new_matrix = sp.zeros(3, 3) # loop on all the matrix elements for j\",\n \"{} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic\",\n \"its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print =\",\n \"for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1)\",\n \"== 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else:\",\n \"Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'),\",\n \"* // 19 14 23 //\\\\n') out_file.write(' * // 20 01 24 //\\\\n')\",\n \"sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] =\",\n \"def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for\",\n \"0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else:\",\n \"//\\\\n') out_file.write(' * // 19 14 23 //\\\\n') out_file.write(' * // 20 01\",\n \"out_file.write(' * Rdi : rotational matrix between body i and its predecessor\\\\n') out_file.write('\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0,\",\n \"masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' ,\",\n \"Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] =\",\n \"if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(',\",\n \"the current body i \\\\n') out_file.write(' * (previous body is not always body\",\n \"out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0\",\n \"// left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop\",\n \"sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')])\",\n \"nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj,\",\n \"//\\\\n') out_file.write(' * // 04 10 //\\\\n') out_file.write(' * // 05 11 //\\\\n')\",\n \"out_file.write('}\\\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for\",\n \"sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12]\",\n \"waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'),\",\n \"points absolute position\\\\n') for i in range(0, nb_contacts): for j in range(0, 3):\",\n \"Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'),\",\n \"of the file def write_file_end(out_file): out_file.write('}\\\\n') # print matrix components declaration def write_matrix_declaration(out_file,\",\n \"sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] =\",\n \"sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')])\",\n \"COM G_i of body i\\\\n') out_file.write(' * xpi : derivative of xi\\\\n') out_file.write('\",\n \"# # 02 08 # # 03 09 # # 04 10 #\",\n \"out_file.write(' * Ri : absolute rotational matrix\\\\n') out_file.write(' * Rti : transpose matrix\",\n \"if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global\",\n \"R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] =\",\n \"= 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k],\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')])\",\n \"body i # ci : cosine of the relative angle before body i\",\n \"vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name,\",\n \"== 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else:\",\n \"leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21',\",\n \"s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write('\",\n \"out_file.write(' * Omi : rotational vector from the previous body to the current\",\n \"cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1):\",\n \"cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len == 1: result +=\",\n \"nb_contacts): for j in range (1, nb_bodies): flag_print = 0 for k in\",\n \"# Rti : transpose matrix of Ri # xji : jacobian of 'xi'\",\n \"-- Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file,\",\n \"flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*',''))\",\n \"if count >= nb_max_char: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n')\",\n \"range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print\",\n \"element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count =\",\n \"2, 1, 3, 2 # left arm ]) # parent index parent_body_index =\",\n \"'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', #\",\n \"= 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist\",\n \"range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if\",\n \"int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len ==\",\n \"Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the\",\n \"joints for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop\",\n \"[-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis ==\",\n \"the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] =\",\n \"outputs out_file.write(' // right foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}]\",\n \"rotational matrix\\\\n') out_file.write(' * Rti : transpose matrix of Ri\\\\n') out_file.write(' * xji\",\n \"for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i],\",\n \", 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', #\",\n \"xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def\",\n \"elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b,\",\n \"sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11]\",\n \", sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'),\",\n \"theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\\\n')\",\n \"len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' // -- variables declaration\",\n \"in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot contact points jacobian\\\\n')\",\n \"sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] =\",\n \"out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0\",\n \"body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the waist and the\",\n \"* (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] =\",\n \"sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] =\",\n \"17 16 21 # # 18 15 22 # # 19 14 23\",\n \"count = 0 with open(in_file,'r') as f: # loop on all the lines\",\n \"# computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T\",\n \"for i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j]\",\n \"from the anchor point of body i to its COM (center of mass)\",\n \"the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' *\",\n \"orientation\\\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write('\",\n \"range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0,\",\n \"R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045)\",\n \"out_file.write(' * frame of the previous body\\\\n') out_file.write(' * Rdi : rotational matrix\",\n \"= nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i\",\n \"def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min,\",\n \"Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0,\",\n \"in range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count\",\n \"out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n\",\n \"return False, -1, -1 # count the number of 'elem' in the file\",\n \"[-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([])\",\n \"= cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i in range(1,\",\n \"j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac\",\n \"(Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file,\",\n \"04 10 # # 05 11 # # 06 12 # # 07\",\n \"loop on all the lines for line in f: # declaration if len(line.split('//\",\n \"np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2:\",\n \"aligned with the ground (info from IMU) # # Di : position vector\",\n \"# print the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if\",\n \"orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for\",\n \"hip rotations # inertial frame: located at the origin (waist), but aligned with\",\n \"+ {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j\",\n \"= line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2)\",\n \"// global com absolute position\\\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] =\",\n \"j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]]\",\n \"Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w')\",\n \"Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'),\",\n \"f: # loop on all the lines for line in f: cut_line =\",\n \"end of the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w')\",\n \"x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of\",\n \"sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')])\",\n \"1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] =\",\n \"save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = '\",\n \"x & xp out_file.write('\\\\n // anchor point absolute positions and velocities\\\\n') x =\",\n \"Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M\",\n \"orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot\",\n \"matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix =\",\n \"= 0 if count != 0: out_write.write(';\\\\n') # get tilde matrix def get_tilde(v):\",\n \"21 //\\\\n') out_file.write(' * // 18 15 22 //\\\\n') out_file.write(' * // 19\",\n \"nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i]\",\n \"epsilon): if epsilon > 0: # epsilon = 1 -> pitch angle in\",\n \"if not flag_first: flag_first = 1 flag_print = 1 elif not flag_print: flag_print\",\n \"[vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)]\",\n \"in range (1, nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac\",\n \"out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if\",\n \"range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1)\",\n \"symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n // -- symbolic computation\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write('\",\n \"save_matrix = 9 * [None] for i in range(0,3): for j in range(0,3):\",\n \"= xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print,\",\n \"in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j\",\n \"write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i]\",\n \"'r') as f: # loop on all the lines for line in f:\",\n \"{};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot contact points absolute\",\n \"for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string\",\n \"def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in range(min, max+1):\",\n \"= write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i]\",\n \"x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot\",\n \"Rdi : rotational matrix between body i and its predecessor\\\\n') out_file.write(' * si\",\n \"and replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix\",\n \"an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop\",\n \"time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{}\",\n \"15 22 //\\\\n') out_file.write(' * // 19 14 23 //\\\\n') out_file.write(' * //\",\n \"range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n') out_file.write('\",\n \"R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_',\",\n \"i-1), expressed in the relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write('\",\n \"om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations\",\n \"out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n')\",\n \"%a, %b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b =\",\n \"0 pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag = 1\",\n \"relative angle before body i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi : absolute position\",\n \"R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1\",\n \"Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i]\",\n \"0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12]\",\n \"nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j]\",\n \"isInt(value): try: int(value) return True except: return False # return true if it\",\n \"all the lines for line in f: # declaration if len(line.split('// -- variables\",\n \"2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis\",\n \"om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id],\",\n \"= symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:\",\n \"[-elem]]) else: return np.array([]) # compute the derivative of an element (for jacobian)\",\n \"of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center of mass\",\n \": jacobian of 'Ri' # return true if it is a float def\",\n \"predecessor # si : sine of the relative angle before body i #\",\n \"}\\\\n\\\\n') out_file.write(' // right foot contact points absolute position\\\\n') for i in range(0,\",\n \"symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic\",\n \"nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] =\",\n \"== 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem],\",\n \"velocities\\\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0]\",\n \"'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk\",\n \"s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0],\",\n \"def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\\\n') out_file.write('\\\\n //\",\n \"new_vector = sp.zeros(3, 1) # loop on all the vector elements for j\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0]\",\n \"elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive neg_flag\",\n \"sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1))\",\n \"sympy as sp import re import os ###################### # # # 17 16\",\n \"in the relative # frame of the previous body # DGi : position\",\n \"(1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2, 1,\",\n \"nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts *\",\n \"len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip()))\",\n \"2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis\",\n \"for i in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print =\",\n \"-v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine):\",\n \"double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\\\n')\",\n \"[0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0,\",\n \"out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i ==\",\n \"* [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot\",\n \"None: if not flag_first: flag_first = 1 flag_print = 1 elif not flag_print:\",\n \"range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not\",\n \"cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n')\",\n \"# waist 0, 1, 2, 3, 4, 5, # right leg 0, 7,\",\n \"for i in range (1, nb_bodies): flag_print = 0 for j in range(0,\",\n \"= sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'),\",\n \"i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative\",\n \"range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot\",\n \"anchor point of body i to its COM (center of mass) G_i,\\\\n') out_file.write('\",\n \"flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if\",\n \"i # ci : cosine of the relative angle before body i #\",\n \"absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n')\",\n \"s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{}\",\n \"1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]])\",\n \"2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]])\",\n \"i\\\\n') out_file.write(' * Ri : absolute rotational matrix\\\\n') out_file.write(' * Rti : transpose\",\n \"in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] =\",\n \"symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix\",\n \"Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names,\",\n \"axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0],\",\n \"+ R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing\",\n \"# 19 14 23 # # 20 01 24 # # 02 08\",\n \"out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0])\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0])\",\n \"-sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get\",\n \"rotational matrix between body i and its predecessor\\\\n') out_file.write(' * si : sine\",\n \"right foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i]))\",\n \"}\\\\n') # omega out_file.write('\\\\n // joint absolute velocities\\\\n') Om = nb_bodies*[None] om =\",\n \"wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n')\",\n \"# declaration if len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' //\",\n \"R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot\",\n \"print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with\",\n \"out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n')\",\n \"*\\\\n') out_file.write(' * xi : absolute position vector (from origin, expressed in the\",\n \", sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'),\",\n \"previous body to the current body i # (previous body is not always\",\n \"sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist =\",\n \"open(in_file, 'r') as f: # loop on all the lines for line in\",\n \"frame of the previous body # Rdi : rotational matrix between body i\",\n \"np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return\",\n \"* [None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i]\",\n \"Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0,\",\n \"if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return\",\n \"omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out,\",\n \"in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] =\",\n \"# 17 16 21 # # 18 15 22 # # 19 14\",\n \"R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\\\n // jacobian\",\n \"matrix, compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix,\",\n \"else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name,\",\n \"sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'),\",\n \"* Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T *\",\n \"vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = '\",\n \"out_file.write(' * si : sine of the relative angle before body i\\\\n') out_file.write('\",\n \"Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' //\",\n \"out_file.write(' // right foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] =\",\n \"1 pos_str = elem_split[1] else: print('Error: {} instead of 1 or 2 in\",\n \"s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot\",\n \"matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_bodies):\",\n \"body i\\\\n') out_file.write(' * Omi : rotational vector from the previous body to\",\n \"out_file.write(' // right foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso orientation angle derivatives\",\n \"kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs and outputs class\\\\n') out_file.write('\",\n \"for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for\",\n \"= atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2],\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get\",\n \"the end of the file def write_file_end(out_file): out_file.write('}\\\\n') # print matrix components declaration\",\n \"0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name\",\n \"in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the\",\n \"# get vector axis def get_vector_axis(axis, direct, elem): if direct: if axis ==\",\n \"xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot +\",\n \"xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id]\",\n \"out_file.write('\\\\n // jacobian com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None]\",\n \"# left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')])\",\n \"the previous body # DGi : position vector from the anchor point of\",\n \"'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj,\",\n \"= Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj,\",\n \"Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13]\",\n \", 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id'\",\n \"//\\\\n') out_file.write(' * // 20 01 24 //\\\\n') out_file.write(' * // 02 08\",\n \"out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i in\",\n \"# save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix =\",\n \"rotation matrices\\\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] =\",\n \"np.array([0, # waist 2, 1, 3, 2, 1, 2, # right leg 2,\",\n \"# left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', #\",\n \"loop on all the joints for i in range (1, nb_bodies): new_matrix =\",\n \"(info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di : position vector from the\",\n \"in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist absolute\",\n \"x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_',\",\n \"om, 0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis,\",\n \"new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix # save the symbolic vector\",\n \"count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f: # loop on\",\n \"R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file,\",\n \"sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4]\",\n \"elem_split[1] else: print('Error: {} instead of 1 or 2 in negative detection !'.format(cur_len))\",\n \"def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None]\",\n \"om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for\",\n \"== 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else:\",\n \"roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon\",\n \"velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"= cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso =\",\n \"end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n') for i in range(0,\",\n \"= {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector # write\",\n \"position vector from the anchor point of body i to its COM (center\",\n \"== 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else:\",\n \"== 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis ==\",\n \"torso orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist\",\n \"xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names,\",\n \"absolute position vector (from origin, expressed in the inertial frame) # of the\",\n \"True except: return False # return true if it has a shape 'R%a_%b%c'\",\n \"-sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine,\",\n \"expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all\",\n \"result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj,\",\n \"'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write('\",\n \"Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1]\",\n \"before body i # # xi : absolute position vector (from origin, expressed\",\n \"prefix): out_file.write(' double ') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix,\",\n \"np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine],\",\n \"out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n // -- Collecting results -- //\\\\n\\\\n') com_compute(out_file,\",\n \"R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max):\",\n \"out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}]\",\n \"term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product\",\n \"xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables\",\n \"R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3)\",\n \"i+1, j+1)) if i == 2 and j == 2: out_file.write(';\\\\n') else: out_file.write(',\",\n \"* frame of the previous body\\\\n') out_file.write(' * DGi : position vector from\",\n \"range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n') for i\",\n \"of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): #\",\n \"return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis ==\",\n \"out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com absolute velocity\\\\n') for i in\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not\",\n \"'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg\",\n \"= symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:\",\n \"angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write('\",\n \"i\\\\n') out_file.write(' * xpi : derivative of xi\\\\n') out_file.write(' * xgpi : derivative\",\n \"[0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis == 1: return np.array([[1.0,\",\n \"x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute velocity\\\\n') for i in range(0,3): out_file.write('\",\n \"matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n')\",\n \"i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist\",\n \"if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_bodies): for j\",\n \"== 0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1\",\n \"of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut\",\n \"forward kinematics computation for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum\",\n \"Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) #\",\n \"np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if direct: if axis\",\n \"parent_body_index = np.array([ -1, # waist 0, 1, 2, 3, 4, 5, #\",\n \"np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id'\",\n \"range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min\",\n \"and orientation\\\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n')\",\n \"'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id]\",\n \"inertial frame) # of the anchor point of body i # xgi :\",\n \"}\\\\n\\\\n') out_file.write(' // left foot contact points absolute position\\\\n') for i in range(0,\",\n \"sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) #\",\n \"yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso orientation matrix as angles [rad]\\\\n')\",\n \"of mass) G_i,\\\\n') out_file.write(' * expressed in the relative frame of the current\",\n \"= sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'),\",\n \"out_file.write('\\\\n') out_file.write(' // right wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}]\",\n \"out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute\",\n \"# loop on all the vector elements for j in range(0, 3): new_vector[j]\",\n \"enumeration\\\\n') out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)): count +=\",\n \"# writing outputs out_file.write(' // right foot absolute position\\\\n') for i in range(0,3):\",\n \"left leg 0, 13, 14, # trunk 15, 16, 17, 18, # right\",\n \"'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23',\",\n \"foot contact points absolute position\\\\n') for i in range(0, nb_contacts): for j in\",\n \"Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i in\",\n \"23 //\\\\n') out_file.write(' * // 20 01 24 //\\\\n') out_file.write(' * // 02\",\n \"xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write('\",\n \"(1, nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj,\",\n \"== 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result +=\",\n \": absolute position vector (from origin, expressed in the inertial frame)\\\\n') out_file.write(' *\",\n \"= 0 # loop on all terms for cur_term in term_list: # detect\",\n \"= ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count ==\",\n \"out_file.write(' *\\\\n') out_file.write(' * Di : position vector from the anchor point of\",\n \"j == 2: out_file.write(';\\\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file,\",\n \"vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print =\",\n \"* Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i],\",\n \"# waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0,\",\n \"sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'),\",\n \"with open(in_temp, 'r') as f: # loop on all the lines for line\",\n \"def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f: # loop\",\n \"* origin: in the waist, middle point between the two pitch hip rotations\\\\n')\",\n \"range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0:\",\n \"###################### # # origin: in the waist, middle point between the two pitch\",\n \"of body i # Ri : absolute rotational matrix # Rti : transpose\",\n \"0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif\",\n \"Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1]\",\n \"body i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi : absolute position vector (from origin,\",\n \"out_file.write(' // right wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"elif count >= 6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n')\",\n \"15 22 # # 19 14 23 # # 20 01 24 #\",\n \"of xgi\\\\n') out_file.write(' * omi : absolute rotational vector of body i\\\\n') out_file.write('\",\n \"om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3,\",\n \"+ R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r =\",\n \"Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'),\",\n \"0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0])\",\n \"09 //\\\\n') out_file.write(' * // 04 10 //\\\\n') out_file.write(' * // 05 11\",\n \"xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T\",\n \"der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted ,\",\n \"matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"= write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies):\",\n \"jacobian x out_file.write('\\\\n // jacobian anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"= -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{},\",\n \"between the two pitch hip rotations # inertial frame: located at the origin\",\n \"= 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else:\",\n \"for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if\",\n \"(!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot\",\n \"16 21 # # 18 15 22 # # 19 14 23 #\",\n \"in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont =\",\n \"1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len\",\n \"* inertial frame: located at the origin (waist), but aligned with the ground\",\n \"') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\\\n')\",\n \"sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left\",\n \"/ c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation\",\n \"in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n')\",\n \"vector from the anchor point of the previous body to the current body\",\n \"[rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot =\",\n \"Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts\",\n \"= nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot =\",\n \"write_file_end(out_file): out_file.write('}\\\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ')\",\n \"leg 0, 13, 14, # trunk 15, 16, 17, 18, # right arm\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results\",\n \"i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] =\",\n \"//\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin: in the waist, middle\",\n \"om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the wrists position, velocity and orientation def\",\n \"out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write('\",\n \"float def isInt(value): try: int(value) return True except: return False # return true\",\n \"if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1\",\n \"// //\\\\n') out_file.write(' * // 17 16 21 //\\\\n') out_file.write(' * // 18\",\n \"for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >=\",\n \"= np.array([0, # waist 2, 1, 3, 2, 1, 2, # right leg\",\n \"derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) #\",\n \"y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for\",\n \"in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute\",\n \"range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0])\",\n \"'') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n')\",\n \"Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1)\",\n \"Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb\",\n \"[sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine,\",\n \"1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot\",\n \"in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j]\",\n \"k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n')\",\n \"pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag = 1 pos_str\",\n \"end_name) save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix,\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"# derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len\",\n \"Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont =\",\n \"0 for k in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first:\",\n \"trunk 2, 1, 3, 2, # right arm 2, 1, 3, 2 #\",\n \"# list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '':\",\n \"orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id,\",\n \"= symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) #\",\n \"np.array([ -1, # waist 0, 1, 2, 3, 4, 5, # right leg\",\n \"axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0],\",\n \"3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot contact\",\n \"2, 3, 4, 5, # right leg 0, 7, 8, 9, 10, 11,\",\n \"Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb\",\n \"for line in f: cut_line = line.split(elem) if len(cut_line) == 2: count +=\",\n \"line in f: cut_line = line.split(elem) if len(cut_line) == 2: count += 1\",\n \"cosine of the relative angle before body i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi\",\n \"points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0,\",\n \"[0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3:\",\n \"cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for i in range(1, nb_bodies): out_file.write(' s{}\",\n \"+= 1 if count >= nb_max_line: out_write.write(';\\\\n') count = 0 if count !=\",\n \"out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j\",\n \"* Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\\\n') for i\",\n \"# # 04 10 # # 05 11 # # 06 12 #\",\n \"point between the two pitch hip rotations # inertial frame: located at the\",\n \"xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i]\",\n \"if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com absolute velocity\\\\n') for\",\n \"f: # loop on all the lines for line in f: # declaration\",\n \"position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot\",\n \"sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')])\",\n \"xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else:\",\n \"arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint names joint_id_names\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0,\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \"# save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector =\",\n \"= int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return False, -1,\",\n \"1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n')\",\n \"velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute position\\\\n')\",\n \"xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None]\",\n \"new_vector # write symbolic matrix and replace symbolic variable by its name def\",\n \"sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] =\",\n \"= sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'),\",\n \"l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb =\",\n \": jacobian of 'xgi' # Rji : jacobian of 'Ri' # return true\",\n \"3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if\",\n \"main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs and outputs class\\\\n')\",\n \"[rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso orientation matrix as angles\",\n \"%b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0])\",\n \"range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix,\",\n \"= print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index,\",\n \"pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8]))\",\n \"0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) #\",\n \"from IMU) # # Di : position vector from the anchor point of\",\n \"atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0],\",\n \", # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id'\",\n \"om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute\",\n \"R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute\",\n \"position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name\",\n \"{};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix\",\n \"// joint cosines\\\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i]))\",\n \"of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element\",\n \"lines for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ')\",\n \"for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write('\",\n \"result # write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n')\",\n \"matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name,\",\n \"j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return\",\n \"// 05 11 //\\\\n') out_file.write(' * // 06 12 //\\\\n') out_file.write(' * //\",\n \"# loop on all the lines for line in f: # declaration if\",\n \"out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write(' // global\",\n \"right foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' //\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result +=\",\n \"* (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] =\",\n \"R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2]\",\n \"# symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end\",\n \"save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3,\",\n \"current body i # (previous body is not always body i-1), expressed in\",\n \"Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag:\",\n \", 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id'\",\n \": absolute rotational matrix # Rti : transpose matrix of Ri # xji\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name,\",\n \"1 !'.format(cur_len-1)) exit() return result # write the beginning of the file def\",\n \"def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n')\",\n \"'M_28', 'M_29' # left arm ]) # joint names joint_id_names = np.array(['0', #\",\n \"joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225)\",\n \"[rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and orientation\",\n \"enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string\",\n \"file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script # rotation\",\n \"-0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23,\",\n \"or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name =\",\n \"xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file,\",\n \"// right foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write('\",\n \"parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i]\",\n \"orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso orientation\",\n \"* Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj,\",\n \"out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\\\n') count = 0\",\n \"elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return\",\n \"= nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3,\",\n \"of body i # xpi : derivative of xi # xgpi : derivative\",\n \"xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj,\",\n \"der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only implemented for 0\",\n \"return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis ==\",\n \"left arm ]) # parent index parent_body_index = np.array([ -1, # waist 0,\",\n \"1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis\",\n \"'omega_torso', 'torso') # compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names,\",\n \"= -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb)\",\n \"save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix\",\n \"flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k] != None: if\",\n \"left leg 1, 2, 3, # trunk 2, 1, 3, 2, # right\",\n \"// jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25',\",\n \"# Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None]\",\n \"previous body # DGi : position vector from the anchor point of body\",\n \"if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine],\",\n \"with open(in_file,'r') as f: # loop on all the lines for line in\",\n \"in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"cur_jac != 0: if not flag_first: flag_first = 1 flag_print = 1 elif\",\n \"nb_max_char): count = 0 with open(in_file,'r') as f: # loop on all the\",\n \"symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write\",\n \"nb_bodies): flag_print = 0 for k in range(0, 3): if xj_print[i][j][k] != None:\",\n \"output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary\",\n \"right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if\",\n \"= {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1)\",\n \"'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17',\",\n \"velocities\\\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3,\",\n \"for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1)))\",\n \"return result # compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj,\",\n \"line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1)\",\n \"return new_vector, save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name,\",\n \"else: return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0,\",\n \"in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for k\",\n \"out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) #\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k,\",\n \"0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3: return\",\n \"angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso\",\n \"= 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id],\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \"9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index),\",\n \"'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id',\",\n \"x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file,\",\n \"in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0],\",\n \"// wrists absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i,\",\n \"i)) out_file.write('\\\\n // joint cosines\\\\n') for i in range(1, nb_bodies): out_file.write(' c{} =\",\n \"== 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count\",\n \"out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i in range(1, nb_bodies): for j in range(0,\",\n \"i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not\",\n \"ground (info from IMU) # # Di : position vector from the anchor\",\n \"arm 15, 20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index) ##\",\n \"int(value.split('_')[1][1]) return True, a, b, c except: return False, -1, -1, -1 #\",\n \"in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in\",\n \"position\\\\n') for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}]\",\n \"waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id],\",\n \"= r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] =\",\n \"out_file.write('\\\\n // joint absolute velocities\\\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde =\",\n \"results out_file.write('\\\\n // -- Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg,\",\n \"end of the file def write_file_end(out_file): out_file.write('}\\\\n') # print matrix components declaration def\",\n \"# main script # rotation axis for each joint before body i (1:x,\",\n \"r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb\",\n \"ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center of mass position and velocity def com_compute(out_file,\",\n \"# right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0,\",\n \"1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max\",\n \"pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8]))\",\n \"the current body i # Omi : rotational vector from the previous body\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n') for i\",\n \"theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and orientation def\",\n \"om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist\",\n \"}\\\\n') # results out_file.write('\\\\n // -- Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names,\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] =\",\n \"# 05 11 # # 06 12 # # 07 13 # #\",\n \"yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n')\",\n \"(R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for\",\n \"(s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) #\",\n \"return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis ==\",\n \"out_file.write(' * // 05 11 //\\\\n') out_file.write(' * // 06 12 //\\\\n') out_file.write('\",\n \"sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'),\",\n \"// right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i]\",\n \"j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write('\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}]\",\n \"# trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id'\",\n \"write symbolic vector and replace symbolic variable by its name def write_symb_vector(out_file, vector,\",\n \"((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n')\",\n \"i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2,\",\n \"x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies,\",\n \"if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if\",\n \"return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis\",\n \"sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'),\",\n \"xgpi : derivative of xgi\\\\n') out_file.write(' * omi : absolute rotational vector of\",\n \"nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write('\",\n \"if i == max: out_file.write(';\\\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file,\",\n \"orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' * // //\\\\n')\",\n \"a, b, c except: return False, -1, -1, -1 # return true if\",\n \"velocity and orientation\\\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\\\n') out_file.write('\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot contact points absolute position\\\\n') for i in\",\n \"0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if\",\n \"symb_var, der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0]\",\n \"= 0 for i in range(1, len(joint_id_names)): count += 1 if i ==\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id'\",\n \"x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}]\",\n \"# angles (position and derivative) of the waist and the torso def torso_waist_angles(out_file,\",\n \"open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp,\",\n \"out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') # print all declarations\",\n \"left arm ]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id',\",\n \"nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot\",\n \"# # 19 14 23 # # 20 01 24 # # 02\",\n \"def write_file_end(out_file): out_file.write('}\\\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double\",\n \"matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left foot\",\n \"omi : absolute rotational vector of body i\\\\n') out_file.write(' * Ri : absolute\",\n \"Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'),\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine],\",\n \"om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T *\",\n \"left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9]\",\n \"of body i to its COM (center of mass) G_i, # expressed in\",\n \"== 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count ==\",\n \"writing outputs out_file.write(' // right foot absolute position\\\\n') for i in range(0,3): out_file.write('\",\n \"sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) ||\",\n \"if count != 0: out_write.write(';\\\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char):\",\n \"neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead of 1 or\",\n \"symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \"out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line:\",\n \"Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0]\",\n \"= 0 for i in range(0,3): for j in range(0,3): if matrix[i,j] ==\",\n \"as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write('\",\n \"out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist\",\n \"\\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics computation for\",\n \"3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if\",\n \"left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22]\",\n \"v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct:\",\n \"the joints for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) #\",\n \"derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot orientation angle\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n') for i in range(0, 9):\",\n \"= om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot =\",\n \"out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute\",\n \"for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3\",\n \"count += 1 if count >= nb_max_line: out_write.write(';\\\\n') count = 0 if count\",\n \"'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot',\",\n \"inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1],\",\n \"return true if it has a shape 'R%a_%b%c' (indexes %a, %b, %c also\",\n \"out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write(' // global com absolute position\\\\n') for i\",\n \"sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19]\",\n \"len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\\\n') count = 0 if count\",\n \"(string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"its predecessor # si : sine of the relative angle before body i\",\n \"R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon\",\n \"= isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag: result\",\n \"body i and its predecessor # si : sine of the relative angle\",\n \"in the relative frame of the current body i\\\\n') out_file.write(' * Omi :\",\n \"x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot *\",\n \"= in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable by its\",\n \"v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis,\",\n \"# trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15]\",\n \"absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write('\",\n \"ci : cosine of the relative angle before body i\\\\n') out_file.write(' *\\\\n') out_file.write('\",\n \"= sp.zeros(3, 1) save_vector = 3 * [None] for i in range(0,3): if\",\n \"out_file.write(' // left wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] =\",\n \"Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)),\",\n \"sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses\",\n \"inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' //\",\n \"{} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector #\",\n \"vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for\",\n \"theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part,\",\n \"np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis\",\n \"= x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] =\",\n \"Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all the joints for\",\n \"xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right\",\n \"sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right\",\n \"# temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of\",\n \"(1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj,\",\n \"for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\\\n') else:\",\n \"// -- symbolic computation -- //\\\\n') # Rj, xj, xgj and xgj (jacobian)\",\n \"right leg 2, 1, 3, 2, 1, 2, # left leg 1, 2,\",\n \"0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'),\",\n \"07 13 # # # ###################### # # origin: in the waist, middle\",\n \"range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0:\",\n \"out_file.write(' // -- variables initialization -- //\\\\n') out_file.write('\\\\n // IMU - rotation matrices\\\\n')\",\n \"0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0,\",\n \"* // 07 13 //\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write('\",\n \"# return true if it is a float def isInt(value): try: int(value) return\",\n \"tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0],\",\n \"axis for each joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0,\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot contact points absolute position\\\\n')\",\n \"file def write_file_end(out_file): out_file.write('}\\\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write('\",\n \"vector from the previous body to the current body i \\\\n') out_file.write(' *\",\n \">= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string,\",\n \"compute the center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M,\",\n \"== 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] =\",\n \"the previous body to the current body i \\\\n') out_file.write(' * (previous body\",\n \"Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm\",\n \"len(cur_term_split) # no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj,\",\n \"0 or vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name\",\n \"Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for\",\n \"lines for line in f: # declaration if len(line.split('// -- variables initialization --\",\n \"the current body i\\\\n') out_file.write(' * Omi : rotational vector from the previous\",\n \"+ {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1],\",\n \"'{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the file def write_file_end(out_file):\",\n \"[rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\\\n')\",\n \"}\\\\n\\\\n') out_file.write(' // left foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}]\",\n \"neg_flag: result = -result return result # compute the derivative of an expression\",\n \"other else: print('Error: {} * counted , only implemented for 0 or 1\",\n \"range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i]\",\n \"sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'),\",\n \"in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first:\",\n \"get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj,\",\n \"str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first:\",\n \"= nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print =\",\n \"nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0,\",\n \"2, # left leg 1, 2, 3, # trunk 2, 1, 3, 2,\",\n \"symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write\",\n \"nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for\",\n \"detection !'.format(cur_len)) exit() # compute derivative result = 0 # cosine if pos_str\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute position\\\\n') for i in range(0,3): out_file.write('\",\n \"return true if it is a float def isInt(value): try: int(value) return True\",\n \"elem_split[0] elif cur_len == 2: # negative neg_flag = 1 pos_str = elem_split[1]\",\n \"19 14 23 //\\\\n') out_file.write(' * // 20 01 24 //\\\\n') out_file.write(' *\",\n \"symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix,\",\n \"leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] =\",\n \"body i # xgi : absolute position vector of the COM G_i of\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): #\",\n \"= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1))\",\n \"{}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\\\n') count = 0 if\",\n \"sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 ,\",\n \"= cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot =\",\n \"out_file.write('\\\\n') out_file.write(' // right foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}]\",\n \"torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1)\",\n \"and velocities\\\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0]\",\n \"cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis == 1:\",\n \"* Rdi : rotational matrix between body i and its predecessor\\\\n') out_file.write(' *\",\n \"for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a rotation matrix def\",\n \"cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1],\",\n \"else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' *\",\n \"out_file.write(' }\\\\n\\\\n') # get a string for the enumeration of joints def get_string_enum(cur_string):\",\n \"1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot =\",\n \"# return true if it has a shape 'x%a_%b' (indexes %a, %b also\",\n \"of the waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write('\",\n \"of the previous body\\\\n') out_file.write(' * Rdi : rotational matrix between body i\",\n \"= nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j\",\n \"only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result # write the\",\n \", # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm\",\n \"1 or 2 in negative detection !'.format(cur_len)) exit() # compute derivative result =\",\n \"sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in\",\n \"else: return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif\",\n \"Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3\",\n \"* Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj,\",\n \"isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True,\",\n \"right foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write('\",\n \"'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies,\",\n \"j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first =\",\n \"!= nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from an orientation matrix, compute the\",\n \"nb_max_char: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') # get tilde\",\n \"a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix\",\n \"direct, elem): if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif\",\n \"s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write('\",\n \"'waist') out_file.write('\\\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso')\",\n \"if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one\",\n \"out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k ==\",\n \"leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' ,\",\n \"xgj, der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative\",\n \"= sp.zeros(3, 3) # loop on all the matrix elements for j in\",\n \"new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name)\",\n \"Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3,\",\n \"always body i-1), expressed in the relative\\\\n') out_file.write(' * frame of the previous\",\n \"out_file.write('\\\\n // joint relative velocities\\\\n') for i in range(1, nb_bodies): out_file.write(' Om{} =\",\n \"names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', #\",\n \"+= sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d,\",\n \"between the two pitch hip rotations\\\\n') out_file.write(' * inertial frame: located at the\",\n \"= 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right\",\n \"= {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute velocity\\\\n') for i in\",\n \"left wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"[None] # rotation matrices out_file.write('\\\\n // rotation matrices\\\\n') R = nb_bodies*[None] Rt =\",\n \"out_file.write(' // wrists absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] =\",\n \"i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i]\",\n \"# inertial frame: located at the origin (waist), but aligned with the ground\",\n \"result # compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj,\",\n \"elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')\",\n \"matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i in\",\n \"for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj,\",\n \"symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \"elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return\",\n \"= xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' //\",\n \"of:\\\\n') out_file.write(' * COM (center of mass) position and velocity\\\\n') out_file.write(' * feet\",\n \"joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0]\",\n \"& xp out_file.write('\\\\n // anchor point absolute positions and velocities\\\\n') x = nb_bodies*[None]\",\n \"1, 2, # left leg 1, 2, 3, # trunk 2, 1, 3,\",\n \"\\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics computation for the COMAN model\\\\n') out_file.write('\",\n \"frame of the current body i # Omi : rotational vector from the\",\n \"count # print the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line):\",\n \"sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6',\",\n \"Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max\",\n \"= parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] *\",\n \"3, 4, 5, # right leg 0, 7, 8, 9, 10, 11, #\",\n \"(!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso\",\n \"else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first:\",\n \"axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]])\",\n \"== 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count\",\n \"'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\\\n')\",\n \"-1, -1, -1 # return true if it has a shape 'x%a_%b' (indexes\",\n \"current body i\\\\n') out_file.write(' * Omi : rotational vector from the previous body\",\n \"G_i of body i # xpi : derivative of xi # xgpi :\",\n \"and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id,\",\n \"velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j])))\",\n \"max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\\\n') else: out_file.write(', ') # variables\",\n \"sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8]\",\n \"]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id',\",\n \"nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i]\",\n \"in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first =\",\n \"'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\\\n // anchor point\",\n \"return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis ==\",\n \"range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j]\",\n \"out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n')\",\n \"body i\\\\n') out_file.write(' * ci : cosine of the relative angle before body\",\n \"out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n') out_file.write(' * COM (center of mass) position\",\n \"range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace\",\n \"out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable\",\n \"elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i]\",\n \"in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1,\",\n \"== nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write(' // global com absolute position\\\\n')\",\n \"parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close() # output\",\n \"body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3,\",\n \"Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r =\",\n \"feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08,\",\n \"i # Ri : absolute rotational matrix # Rti : transpose matrix of\",\n \"= sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist)\",\n \"range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute velocity\\\\n')\",\n \"com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on\",\n \"j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j]\",\n \"* // 02 08 //\\\\n') out_file.write(' * // 03 09 //\\\\n') out_file.write(' *\",\n \"class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n') out_file.write(' * COM (center of mass)\",\n \"previous body to the current body i \\\\n') out_file.write(' * (previous body is\",\n \"}\\\\n') # xg & xgp out_file.write('\\\\n // com absolute positions and velocities\\\\n') xg\",\n \"}\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n')\",\n \"sp.zeros(3, 1) save_vector = 3 * [None] for i in range(0,3): if vector[i]\",\n \"j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac !=\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n')\",\n \"Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj,\",\n \"1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x\",\n \"jacobian com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"//\\\\n') out_file.write(' * // 02 08 //\\\\n') out_file.write(' * // 03 09 //\\\\n')\",\n \"# results out_file.write('\\\\n // -- Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M,\",\n \"out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com absolute velocity\\\\n') for i in range(0, 3):\",\n \"xji : jacobian of 'xi' # xgji : jacobian of 'xgi' # Rji\",\n \"0: out_write.write(';\\\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2],\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute position\\\\n') for i\",\n \"y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot\",\n \"if neg_flag: result = -result return result # compute the derivative of an\",\n \"= symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:\",\n \"= np.array([ -1, # waist 0, 1, 2, 3, 4, 5, # right\",\n \"the vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj,\",\n \"also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True,\",\n \"= 0 for i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print\",\n \"to its COM (center of mass) G_i, # expressed in the relative frame\",\n \"out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_bodies): for\",\n \"'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg,\",\n \"matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print =\",\n \"|| (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write('\",\n \"a string for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if\",\n \"cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n')\",\n \"rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on\",\n \"in range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i]\",\n \"0, 13, 14, # trunk 15, 16, 17, 18, # right arm 15,\",\n \"* xgpi : derivative of xgi\\\\n') out_file.write(' * omi : absolute rotational vector\",\n \"and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print\",\n \"range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j))\",\n \"symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of\",\n \"isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag: result +=\",\n \"# compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write('\",\n \"'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print =\",\n \"= sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'),\",\n \"# 20 01 24 # # 02 08 # # 03 09 #\",\n \"== 2 and j == 2: out_file.write(';\\\\n') else: out_file.write(', ') # print variables\",\n \"in the relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write(' * Rdi\",\n \"* // 05 11 //\\\\n') out_file.write(' * // 06 12 //\\\\n') out_file.write(' *\",\n \"= 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({},\",\n \"sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] =\",\n \"}\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}]\",\n \"IMU - rotation matrices\\\\n') for i in range(0, 3): for j in range(0,\",\n \"angles (position and derivative) of the waist and the torso def torso_waist_angles(out_file, R,\",\n \"& xgp out_file.write('\\\\n // com absolute positions and velocities\\\\n') xg = nb_bodies*[None] xgp\",\n \"xgi : absolute position vector of the COM G_i of body i #\",\n \"= 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg & xgp out_file.write('\\\\n // com\",\n \"on all the lines for line in f: cut_line = line.split(' = ')\",\n \"joint absolute velocities\\\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0]\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"the center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg,\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script\",\n \"rotation matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count == 0:\",\n \"sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'),\",\n \": cosine of the relative angle before body i\\\\n') out_file.write(' *\\\\n') out_file.write(' *\",\n \"+ om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1)\",\n \"detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len\",\n \"'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i]\",\n \"[0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis ==\",\n \"out_file.write(' * waist and torso orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' *\",\n \"in range (1, nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac\",\n \"def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out,\",\n \"out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i ==\",\n \"1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0],\",\n \"out_file.write(' * xgpi : derivative of xgi\\\\n') out_file.write(' * omi : absolute rotational\",\n \"inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] =\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot contact points absolute position\\\\n') for i\",\n \"out_file.write(' * // //\\\\n') out_file.write(' * // 17 16 21 //\\\\n') out_file.write(' *\",\n \"nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n // -- symbolic computation --\",\n \"out_file.write('\\\\n') out_file.write(' // right wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"true if it is a float def isInt(value): try: int(value) return True except:\",\n \"pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result\",\n \"i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' +\",\n \"out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot orientation\",\n \"result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return result\",\n \"position vector of the COM G_i of body i # xpi : derivative\",\n \"cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) #\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get a string for the enumeration\",\n \"leg 0, 7, 8, 9, 10, 11, # left leg 0, 13, 14,\",\n \"xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file,\",\n \"sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4]\",\n \"i \\\\n') out_file.write(' * (previous body is not always body i-1), expressed in\",\n \"der_var) # one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj,\",\n \"sine of the relative angle before body i\\\\n') out_file.write(' * ci : cosine\",\n \"in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j])))\",\n \"= y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i\",\n \"= symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:\",\n \": cosine of the relative angle before body i # # xi :\",\n \"file os.remove(in_temp) # main script # rotation axis for each joint before body\",\n \", sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6]\",\n \"out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} +\",\n \"'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id'\",\n \"// left foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n')\",\n \"range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if\",\n \"mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write('\",\n \"R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives\",\n \"write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in\",\n \"= om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i]\",\n \"// right foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_contacts): for j\",\n \"len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6:\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'),\",\n \"{}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\\\n') count =\",\n \"// 07 13 //\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n')\",\n \"j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print,\",\n \"* Di : position vector from the anchor point of the previous body\",\n \"line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('['))\",\n \"the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if\",\n \"pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c]\",\n \"//\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin:\",\n \"in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] =\",\n \"relative angle before body i # # xi : absolute position vector (from\",\n \"point of the previous body to the current body i \\\\n') out_file.write(' *\",\n \"all the lines for line in f: cut_line = line.split(elem) if len(cut_line) ==\",\n \"(previous body is not always body i-1), expressed in the relative\\\\n') out_file.write(' *\",\n \"in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j))\",\n \"False, -1, -1 # count the number of 'elem' in the file def\",\n \"to its COM (center of mass) G_i,\\\\n') out_file.write(' * expressed in the relative\",\n \"xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id]\",\n \"symbolic computation -- //\\\\n') # Rj, xj, xgj and xgj (jacobian) Rj =\",\n \"= vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {}\",\n \"= None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} =\",\n \"Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm\",\n \"# (previous body is not always body i-1), expressed in the relative #\",\n \"# 18 15 22 # # 19 14 23 # # 20 01\",\n \"of mass) G_i, # expressed in the relative frame of the current body\",\n \"== 2: count += 1 return count # print the declaration of an\",\n \"x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist absolute velocity\\\\n') for i in range(0,3): out_file.write('\",\n \"23 # # 20 01 24 # # 02 08 # # 03\",\n \"in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot\",\n \"* Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i\",\n \"all the lines for line in f: cut_line = line.split(' = ') if\",\n \"9, 10, 11, # left leg 0, 13, 14, # trunk 15, 16,\",\n \"elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0,\",\n \"point of body i\\\\n') out_file.write(' * xgi : absolute position vector of the\",\n \"j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1:\",\n \"= 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n')\",\n \"R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"count != 0: out_write.write(';\\\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2],\",\n \"0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if\",\n \"= x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id]\",\n \"flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j],\",\n \"range(0,3): for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1:\",\n \", 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics'\",\n \"= get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb +\",\n \"{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0],\",\n \"# # Di : position vector from the anchor point of the previous\",\n \"cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n')\",\n \"i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name)\",\n \"* xpi : derivative of xi\\\\n') out_file.write(' * xgpi : derivative of xgi\\\\n')\",\n \"# parent index parent_body_index = np.array([ -1, # waist 0, 1, 2, 3,\",\n \"= atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] =\",\n \"body i\\\\n') out_file.write(' * xpi : derivative of xi\\\\n') out_file.write(' * xgpi :\",\n \"i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i]\",\n \"sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) #\",\n \"expressed in the relative # frame of the previous body # Rdi :\",\n \"x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id]\",\n \"count != 0: out_write.write(';\\\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count\",\n \"9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write('\",\n \"-- variables initialization -- //\\\\n') out_file.write('\\\\n // IMU - rotation matrices\\\\n') for i\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n') for i in range(0, 9):\",\n \"i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from an orientation matrix, compute\",\n \"return np.array([]) # compute the derivative of an element (for jacobian) def der_elem(elem_str,\",\n \"out_file.write(' // waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write('\",\n \"{};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative)\",\n \"all terms for cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len\",\n \"Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj,\",\n \"cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product\",\n \"out_file.write(' // global com absolute velocity\\\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}]\",\n \"atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0],\",\n \"inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file,\",\n \"sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9')\",\n \"9): if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print =\",\n \"point absolute positions and velocities\\\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] =\",\n \"# epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0]\",\n \"return False # return true if it has a shape 'R%a_%b%c' (indexes %a,\",\n \"c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation angle\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y\",\n \"= R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1)))\",\n \"# Ri : absolute rotational matrix # Rti : transpose matrix of Ri\",\n \"0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0,\",\n \"for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if\",\n \"') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count == 0:\",\n \"20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor point\",\n \"the relative # frame of the previous body # DGi : position vector\",\n \"= '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] =\",\n \": derivative of xgi\\\\n') out_file.write(' * omi : absolute rotational vector of body\",\n \"open(in_temp, 'r') as f: # loop on all the lines for line in\",\n \"derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon =\",\n \"def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for\",\n \"x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l\",\n \"= ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix #\",\n \"xpi : derivative of xi # xgpi : derivative of xgi # omi\",\n \"leg 2, 1, 3, 2, 1, 2, # left leg 1, 2, 3,\",\n \"= print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def\",\n \"range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the vector\",\n \"def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1: return np.array([[1.0,\",\n \"[xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names,\",\n \"of the current body i # Omi : rotational vector from the previous\",\n \"'xi' # xgji : jacobian of 'xgi' # Rji : jacobian of 'Ri'\",\n \"vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result =\",\n \"joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) #\",\n \"= sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts\",\n \"i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0]\",\n \"out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\\\n') else: out_file.write(', ') # variables initialization\",\n \"and derivative) of the waist and the torso def torso_waist_angles(out_file, R, om, waist_id,\",\n \"# copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main\",\n \"of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): #\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') #\",\n \"# # # ###################### # # origin: in the waist, middle point between\",\n \"xgi # omi : absolute rotational vector of body i # Ri :\",\n \"expressed in the inertial frame) # of the anchor point of body i\",\n \"but aligned with the ground (info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di\",\n \"out_file.write(' // waist orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n')\",\n \"orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the wrists\",\n \"r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id]\",\n \"range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print\",\n \"* (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist)\",\n \"om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0,\",\n \"omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part,\",\n \"in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] =\",\n \"the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp,\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] =\",\n \"G_i, # expressed in the relative frame of the current body i #\",\n \"sine): if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine,\",\n \"[0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0,\",\n \"str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i !=\",\n \"1) flag_print = 0 for i in range(0,3): if vector[i] == 0 or\",\n \"= {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix # save\",\n \"write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i]\",\n \"of the relative angle before body i\\\\n') out_file.write(' * ci : cosine of\",\n \"declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id]\",\n \"get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T\",\n \"if term_list[0] == '': term_list.pop(0) result = 0 # loop on all terms\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6,\",\n \"nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0])\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \"of the COM G_i of body i # xpi : derivative of xi\",\n \"right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') ,\",\n \"sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] =\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file,\",\n \"axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split)\",\n \"sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] =\",\n \"1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist =\",\n \"xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"jacobian of 'xgi' # Rji : jacobian of 'Ri' # return true if\",\n \"False # return true if it has a shape 'R%a_%b%c' (indexes %a, %b,\",\n \"Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for\",\n \"in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right\",\n \"in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n')\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"new_string = cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for i\",\n \"# jacobian xg out_file.write('\\\\n // jacobian com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) #\",\n \"computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the\",\n \"{};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot contact points jacobian\\\\n') out_file.write(' if\",\n \"= elem_split[0] elif cur_len == 2: # negative neg_flag = 1 pos_str =\",\n \"// left wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i]\",\n \"xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"= 0 for k in range(0, 3): if xgj_print[i][j][k] != None: if not\",\n \"9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute orientation\",\n \"= cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist =\",\n \"xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1 elif\",\n \"body i # Omi : rotational vector from the previous body to the\",\n \"out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n')\",\n \"from the anchor point of the previous body to the current body i\",\n \"out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics\",\n \"== nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com absolute\",\n \"body # DGi : position vector from the anchor point of body i\",\n \"= nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1)\",\n \"sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')])\",\n \"{\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso =\",\n \"R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3,\",\n \"write the end of the file def write_file_end(out_file): out_file.write('}\\\\n') # print matrix components\",\n \"return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative of an\",\n \"= r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb\",\n \"cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and\",\n \"3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis == 1:\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"IMU - angles velocity\\\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1,\",\n \"nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in\",\n \"'.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' +\",\n \"= sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a rotation matrix\",\n \"+ om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1)\",\n \"point of body i to its COM (center of mass) G_i, # expressed\",\n \"jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range (1,\",\n \"sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] =\",\n \"trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] =\",\n \"computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T *\",\n \"-sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else:\",\n \"[Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor\",\n \"rotational vector of body i # Ri : absolute rotational matrix # Rti\",\n \"05 11 //\\\\n') out_file.write(' * // 06 12 //\\\\n') out_file.write(' * // 07\",\n \"of the relative angle before body i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi :\",\n \"orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist =\",\n \"c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso\",\n \"position vector (from origin, expressed in the inertial frame) # of the anchor\",\n \"product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) #\",\n \"an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to\",\n \"has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try:\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute orientation jacobian\\\\n')\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n //\",\n \"r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb\",\n \"xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3,\",\n \"sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')])\",\n \"j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' //\",\n \"= filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string\",\n \"if i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from an orientation matrix,\",\n \"[-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] =\",\n \"main script # rotation axis for each joint before body i (1:x, 2:y,\",\n \"') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else:\",\n \"out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write('\",\n \"0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine,\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \"elem): count = 0; with open(in_file, 'r') as f: # loop on all\",\n \"+ {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1],\",\n \"x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot absolute velocity\\\\n') for i in range(0,3): out_file.write('\",\n \"0, 7, 8, 9, 10, 11, # left leg 0, 13, 14, #\",\n \"body i # Ri : absolute rotational matrix # Rti : transpose matrix\",\n \"cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n')\",\n \"out_file.write(' * DGi : position vector from the anchor point of body i\",\n \"# compute derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q): result\",\n \"sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0]\",\n \"out_file.write(' // torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute\",\n \"# write the end of the file def write_file_end(out_file): out_file.write('}\\\\n') # print matrix\",\n \"xi # xgpi : derivative of xgi # omi : absolute rotational vector\",\n \"for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i\",\n \"= sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0,\",\n \"== 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj,\",\n \"write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' *\",\n \"other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str)\",\n \"parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id]\",\n \"sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j]\",\n \"'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id'\",\n \"= 1 pos_str = elem_split[1] else: print('Error: {} instead of 1 or 2\",\n \"+= der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip())\",\n \"Dpt[0] xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print,\",\n \"}\\\\n\\\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id],\",\n \"//\\\\n') out_file.write(' * // 17 16 21 //\\\\n') out_file.write(' * // 18 15\",\n \"\\\\n') out_file.write(' * (previous body is not always body i-1), expressed in the\",\n \"Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0,\",\n \"nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com absolute velocity\\\\n')\",\n \"= sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29')\",\n \"out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n') out_file.write(' if\",\n \"in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write('\",\n \"') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization\",\n \"out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin: in the waist, middle point\",\n \"range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write('\",\n \"2 # left arm ]) # parent index parent_body_index = np.array([ -1, #\",\n \"symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \"* (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] =\",\n \"+= 1 return count # print the declaration of an element def print_declaration_elem(in_file,\",\n \"len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the\",\n \"print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt,\",\n \"xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb +\",\n \", 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16',\",\n \"i in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print = 0\",\n \"!= 0: out_write.write(';\\\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count =\",\n \"= ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n')\",\n \"if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first =\",\n \"= nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0,\",\n \"out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write('\",\n \"3) flag_print = 0 for i in range(0,3): for j in range(0,3): if\",\n \"1) save_vector = 3 * [None] for i in range(0,3): if vector[i] ==\",\n \"xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None] # rotation matrices\",\n \"3) R[0] = sp.zeros(3, 3) for i in range(0, 3): for j in\",\n \"return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([])\",\n \"'{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if\",\n \"= matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j]\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg & xgp\",\n \": derivative of xi # xgpi : derivative of xgi # omi :\",\n \"Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i]\",\n \"R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices\",\n \"= nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0]\",\n \"flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"rotational vector from the previous body to the current body i # (previous\",\n \"om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '')\",\n \"# cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str\",\n \"xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') #\",\n \"s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n')\",\n \"element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q =\",\n \"inertial frame)\\\\n') out_file.write(' * of the anchor point of body i\\\\n') out_file.write(' *\",\n \"write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i\",\n \"6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj,\",\n \"out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot contact\",\n \"x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] *\",\n \"= x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot =\",\n \"the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem))\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n') for i in range(0, 9): out_file.write('\",\n \"om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id =\",\n \"xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write('\",\n \"= {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist absolute velocity\\\\n') for i in\",\n \"= nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1):\",\n \"out_file.write(' * expressed in the relative frame of the current body i\\\\n') out_file.write('\",\n \"','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-')\",\n \"out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double\",\n \"one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip())\",\n \"for j in range(1, nb_bodies): flag_print = 0 for k in range(0, 9):\",\n \"= 0 if count != 0: out_write.write(';\\\\n') # print all declarations def print_all_declaration(in_file,\",\n \"[-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine,\",\n \"nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n')\",\n \"Rj, xj, xgj, xgj_print, x_vector, index): # loop on all the joints for\",\n \"com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om,\",\n \"0.0, 0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return\",\n \"xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot\",\n \"for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i\",\n \"sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'),\",\n \"IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di : position vector from the anchor point\",\n \"Rj_print, R_matrix, index): # loop on all the joints for i in range\",\n \"of mass) position and velocity\\\\n') out_file.write(' * feet position, velocity and orientation\\\\n') out_file.write('\",\n \"{};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i]))\",\n \"cur_len == 1: # positive neg_flag = 0 pos_str = elem_split[0] elif cur_len\",\n \"xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot\",\n \"for i in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on\",\n \"out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center of mass position and velocity def\",\n \"Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\\\n') for i in\",\n \"joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min,\",\n \"') # print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ')\",\n \"the anchor point of body i # xgi : absolute position vector of\",\n \"of Ri # xji : jacobian of 'xi' # xgji : jacobian of\",\n \"result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} *\",\n \"sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] =\",\n \"declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id]\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n // --\",\n \"1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3\",\n \"for i in range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i])))\",\n \"write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for j in range(0,3):\",\n \"out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] =\",\n \"waist and torso orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write('\",\n \"nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first\",\n \"'{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] =\",\n \"* Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot +\",\n \"write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg\",\n \"positions and velocities\\\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0,\",\n \"exit() return result # write the beginning of the file def write_file_beginning(out_file, joint_id_names):\",\n \"[pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] =\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] =\",\n \"feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max,\",\n \"3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU - angles velocity\\\\n')\",\n \"-sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0],\",\n \"if epsilon > 0: # epsilon = 1 -> pitch angle in [-pi/2\",\n \"print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point def write_symb_xgj(nb_bodies,\",\n \"write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print,\",\n \"for j in range (1, nb_bodies): flag_print = 0 for k in range(0,\",\n \"'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg\",\n \"+ R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot)\",\n \"cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i in\",\n \"# symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id]\",\n \"sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a,\",\n \"c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position\",\n \"sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None]\",\n \"out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute velocity\\\\n') for\",\n \"'_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj,\",\n \"= elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split\",\n \"absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write('\",\n \"3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n\",\n \"om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i\",\n \"= sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14]\",\n \"out_file.write(' }\\\\n') # results out_file.write('\\\\n // -- Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies,\",\n \"matrix\\\\n') out_file.write(' * Rti : transpose matrix of Ri\\\\n') out_file.write(' * xji :\",\n \"position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] +\",\n \"# write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write('\",\n \"{') count = 0 for i in range(1, len(joint_id_names)): count += 1 if\",\n \"}\\\\n\\\\n') # from an orientation matrix, compute the roll, pitch, yaw angles (and\",\n \"out_file.write(';\\\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' //\",\n \"x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max\",\n \"= sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') ,\",\n \"= R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16')\",\n \"= x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] =\",\n \"Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot orientation matrix as\",\n \"if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis ==\",\n \": absolute position vector (from origin, expressed in the inertial frame) # of\",\n \"of the COM G_i of body i\\\\n') out_file.write(' * xpi : derivative of\",\n \"the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix\",\n \", 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies,\",\n \"torso orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' * //\",\n \"return new_vector # write symbolic matrix and replace symbolic variable by its name\",\n \"its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print =\",\n \"= x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot\",\n \"point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0,\",\n \"# write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj,\",\n \"computation of:\\\\n') out_file.write(' * COM (center of mass) position and velocity\\\\n') out_file.write(' *\",\n \"sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')])\",\n \"if cur_len == 1: # positive neg_flag = 0 pos_str = elem_split[0] elif\",\n \"+ om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T\",\n \"save_vector = 3 * [None] for i in range(0,3): if vector[i] == 0\",\n \"Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0, 3):\",\n \"om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file,\",\n \"= '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string\",\n \"print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 *\",\n \"rotational vector from the previous body to the current body i \\\\n') out_file.write('\",\n \"cut_line = line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1:\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega\",\n \"as f: # loop on all the lines for line in f: #\",\n \"xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian\",\n \"'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id'\",\n \"// left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation\",\n \"R, om, 0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies,\",\n \"right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0,\",\n \"out_file.write(' * COM (center of mass) position and velocity\\\\n') out_file.write(' * feet position,\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"* si : sine of the relative angle before body i\\\\n') out_file.write(' *\",\n \"*/\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center of mass position and velocity\",\n \"in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2]\",\n \"M, xg, xgp, xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies):\",\n \"out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute\",\n \"sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10]\",\n \"nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3,\",\n \"IMU) # # Di : position vector from the anchor point of the\",\n \"{};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0, 3): for j\",\n \"pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{}\",\n \"COM (center of mass) G_i,\\\\n') out_file.write(' * expressed in the relative frame of\",\n \"(previous body is not always body i-1), expressed in the relative # frame\",\n \"# output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: #\",\n \"xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"= '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j]\",\n \"in range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first = 1\",\n \"in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print = 0 for\",\n \"on all the lines for line in f: # declaration if len(line.split('// --\",\n \"at the origin (waist), but aligned with the ground (info from IMU) #\",\n \"= om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist\",\n \"'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies,\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] =\",\n \"left foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\\\n') else: out_file.write(', ') #\",\n \"# right arm 2, 1, 3, 2 # left arm ]) # parent\",\n \"1, 2, 3, # trunk 2, 1, 3, 2, # right arm 2,\",\n \"range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in range(0, 9):\",\n \"out_file.write(' * computation of:\\\\n') out_file.write(' * COM (center of mass) position and velocity\\\\n')\",\n \"out_file.write(' // -- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy\",\n \"# get a string for the enumeration of joints def get_string_enum(cur_string): cur_split =\",\n \"flag_print: out_file.write('\\\\n') return new_matrix # save the symbolic vector for print def print_save_symb_vector(vector,\",\n \"i # xgi : absolute position vector of the COM G_i of body\",\n \"for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0\",\n \"out_file.write(' * // 20 01 24 //\\\\n') out_file.write(' * // 02 08 //\\\\n')\",\n \"on all the matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj,\",\n \"\\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward\",\n \"frame)\\\\n') out_file.write(' * of the anchor point of body i\\\\n') out_file.write(' * xgi\",\n \"start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i in range(0,3):\",\n \"xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] =\",\n \"anchor point of body i # xgi : absolute position vector of the\",\n \"trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] =\",\n \"on all the joints for i in range (1, nb_bodies): new_vector = sp.zeros(3,\",\n \"[0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0,\",\n \"= 0 for i in range (1, nb_bodies): flag_print = 0 for j\",\n \"pos_str = elem_split[1] else: print('Error: {} instead of 1 or 2 in negative\",\n \"R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso',\",\n \"sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] =\",\n \"= symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first:\",\n \"Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x,\",\n \"M) # end of the file write_file_end(file_temp) file_temp.close() # output file out_file =\",\n \"torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet\",\n \"Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj,\",\n \"nb_bodies): flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj,\",\n \"of the relative angle before body i # # xi : absolute position\",\n \"out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write('\",\n \"+ R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r =\",\n \"1, 3, 2, 1, 2, # left leg 1, 2, 3, # trunk\",\n \": absolute rotational vector of body i\\\\n') out_file.write(' * Ri : absolute rotational\",\n \"name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0\",\n \"the origin (waist), but aligned with the ground (info from IMU)\\\\n') out_file.write(' *\\\\n')\",\n \"10, 11, # left leg 0, 13, 14, # trunk 15, 16, 17,\",\n \", 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18',\",\n \"= write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i\",\n \"nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj,\",\n \"frame of the previous body # DGi : position vector from the anchor\",\n \"x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb\",\n \"for i in range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1,\",\n \"sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] =\",\n \"') out_file.write(' // global com absolute position\\\\n') for i in range(0, 3): out_file.write('\",\n \"# rotation matrices out_file.write('\\\\n // rotation matrices\\\\n') R = nb_bodies*[None] Rt = nb_bodies*[None]\",\n \"-1 # return true if it has a shape 'x%a_%b' (indexes %a, %b\",\n \"new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for i in range(0,3):\",\n \"// IMU - angles velocity\\\\n') for i in range(0, 3): out_file.write(' omega_{} =\",\n \">= nb_max_line: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') # print\",\n \"out_file.write(' // right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"out_file.write(' *\\\\n') out_file.write(' * origin: in the waist, middle point between the two\",\n \"om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot\",\n \"Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T #\",\n \"Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg\",\n \"vector axis def get_vector_axis(axis, direct, elem): if direct: if axis == 1: return\",\n \"else: return np.array([]) # compute the derivative of an element (for jacobian) def\",\n \"computation for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count\",\n \"flag_first = 0 for i in range(0, nb_bodies): for j in range(1, nb_bodies):\",\n \"get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n') for i in\",\n \"variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1)\",\n \"cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str ==\",\n \"Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x &\",\n \"out_file.write('\\\\n // -- Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp,\",\n \"{}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0]))\",\n \"}\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n')\",\n \"leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' ,\",\n \"return True except: return False # return true if it has a shape\",\n \"k in range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first =\",\n \"in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont =\",\n \"= sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M =\",\n \"filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string =\",\n \"axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the\",\n \"R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1,\",\n \"x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute\",\n \"except: return False, -1, -1 # count the number of 'elem' in the\",\n \"// 03 09 //\\\\n') out_file.write(' * // 04 10 //\\\\n') out_file.write(' * //\",\n \"for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint\",\n \"xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write\",\n \"the lines for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' =\",\n \"return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine,\",\n \"# loop on all the lines for line in f: cut_line = line.split(elem)\",\n \"the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3)\",\n \"def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all the\",\n \"joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for i in range(0,\",\n \"// right foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n')\",\n \"point of body i # xgi : absolute position vector of the COM\",\n \"out_file.write(' * \\\\\\\\brief forward kinematics computation for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('//\",\n \"out_file.write(' // global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i in range(1,\",\n \"i\\\\n') out_file.write(' * xgi : absolute position vector of the COM G_i of\",\n \"om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T *\",\n \"xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) #\",\n \"and j == 2: out_file.write(';\\\\n') else: out_file.write(', ') # print variables declaration def\",\n \"end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return\",\n \"'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13',\",\n \"absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write('\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i],\",\n \"positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive\",\n \"= nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i]\",\n \"{}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0]))\",\n \"elem): if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis\",\n \"* Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x\",\n \"sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) #\",\n \"body i # xpi : derivative of xi # xgpi : derivative of\",\n \"arm ]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id',\",\n \"is a float def isInt(value): try: int(value) return True except: return False #\",\n \"right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id'\",\n \"sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix,\",\n \"(and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon\",\n \"R_matrix[0])) else: # epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2]\",\n \"0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1))\",\n \"i in range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] =\",\n \"= sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] =\",\n \"'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24',\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write('\",\n \"initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\\\n') out_file.write('\\\\n\",\n \"result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) #\",\n \"new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string))\",\n \"cur_len = len(cur_term_split) # no product if cur_len == 1: result += der_elem(cur_term_split[0],\",\n \"index parent_body_index = np.array([ -1, # waist 0, 1, 2, 3, 4, 5,\",\n \"= nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1)\",\n \"i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split\",\n \"np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) #\",\n \"for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left\",\n \"arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M)\",\n \"= 3 * [None] # rotation matrices out_file.write('\\\\n // rotation matrices\\\\n') R =\",\n \"]) # parent index parent_body_index = np.array([ -1, # waist 0, 1, 2,\",\n \"= om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist =\",\n \"M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,\",\n \"1, 2, # right leg 2, 1, 3, 2, 1, 2, # left\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute position\\\\n') for\",\n \"Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg\",\n \"# left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0])\",\n \"it has a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try:\",\n \"joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2,\",\n \"xpi : derivative of xi\\\\n') out_file.write(' * xgpi : derivative of xgi\\\\n') out_file.write('\",\n \"but aligned with the ground (info from IMU) # # Di : position\",\n \"= sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'),\",\n \"body i\\\\n') out_file.write(' * Ri : absolute rotational matrix\\\\n') out_file.write(' * Rti :\",\n \"-1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name,\",\n \"= inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1]\",\n \"int(value.split('_')[1]) return True, a, b except: return False, -1, -1 # count the\",\n \"orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot =\",\n \"Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') #\",\n \"# vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result\",\n \"sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11]\",\n \"range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n') out_file.write('\",\n \"out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r')\",\n \"out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write('\",\n \"if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 /\",\n \"double ') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1))\",\n \"absolute positions and velocities\\\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0]\",\n \"new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector # write symbolic matrix and\",\n \"lines for line in f: cut_line = line.split(' = ') if len(cut_line) ==\",\n \"line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) ==\",\n \"# detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1:\",\n \"out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n') out_file.write(' if\",\n \"out_file.write(' // left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"at the origin (waist), but aligned with the ground (info from IMU)\\\\n') out_file.write('\",\n \"flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k] != None: if\",\n \"body # Rdi : rotational matrix between body i and its predecessor #\",\n \"= sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21]\",\n \"out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet\",\n \"of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj):\",\n \"in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i]\",\n \"'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\\\n // jacobian rotation\",\n \"3): if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print =\",\n \"isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag:\",\n \"in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0:\",\n \"mass) G_i,\\\\n') out_file.write(' * expressed in the relative frame of the current body\",\n \"0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if\",\n \"it has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value):\",\n \"in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n') for i in range(1, nb_bodies): out_file.write(' c{}\",\n \"[v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct,\",\n \"for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1)\",\n \"sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n\",\n \"Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'),\",\n \"# si : sine of the relative angle before body i # ci\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot contact points\",\n \"sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n')\",\n \"//\\\\n') out_file.write(' * // 05 11 //\\\\n') out_file.write(' * // 06 12 //\\\\n')\",\n \"vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i in\",\n \"shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try: a =\",\n \"1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] =\",\n \"-- symbolic computation -- //\\\\n') # Rj, xj, xgj and xgj (jacobian) Rj\",\n \"np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2:\",\n \"terms for cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len =\",\n \"rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0,\",\n \"sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 ,\",\n \"ci : cosine of the relative angle before body i # # xi\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists\",\n \"'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm\",\n \"xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i]\",\n \"out_file.write(' // left foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1)\",\n \"xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] =\",\n \"= 0 for k in range(0, 3): if xj_print[i][j][k] != None: if not\",\n \"sp.zeros(3, 3) # loop on all the matrix elements for j in range(0,\",\n \"# left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt,\",\n \"foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the\",\n \"count = 0 for i in range(1, len(joint_id_names)): count += 1 if i\",\n \"Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0, 3):\",\n \"###################### # # # 17 16 21 # # 18 15 22 #\",\n \"body i to its COM (center of mass) G_i,\\\\n') out_file.write(' * expressed in\",\n \"the inertial frame)\\\\n') out_file.write(' * of the anchor point of body i\\\\n') out_file.write('\",\n \"om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3,\",\n \"position\\\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0\",\n \"x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] *\",\n \"# xgi : absolute position vector of the COM G_i of body i\",\n \"= x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb =\",\n \"}\\\\n\\\\n') # get a string for the enumeration of joints def get_string_enum(cur_string): cur_split\",\n \"x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min\",\n \"jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index):\",\n \"om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] =\",\n \"file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f:\",\n \"xj, xgj, xj_print, x_vector, index): # loop on all the joints for i\",\n \"in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot absolute\",\n \"2, 1, 3, 2, # right arm 2, 1, 3, 2 # left\",\n \"and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n') out_file.write(' * COM (center\",\n \"1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if\",\n \"xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id]\",\n \"(R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\\\n') for\",\n \"// com absolute positions and velocities\\\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for\",\n \"e] = isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector\",\n \"count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if\",\n \"= cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len == 1: result\",\n \"out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n') out_file.write(' if\",\n \"* ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin: in the waist, middle point between\",\n \"nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] =\",\n \"# symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id]\",\n \"'Lfoot') out_file.write('\\\\n') # compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names,\",\n \"= 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot =\",\n \"# loop on all the lines for line in f: cut_line = line.split('\",\n \"0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] =\",\n \"i in range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1))\",\n \"out_file.write(' * // 19 14 23 //\\\\n') out_file.write(' * // 20 01 24\",\n \"def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon = 1\",\n \"str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 # loop on\",\n \"if flag_print: out_file.write('\\\\n') return new_vector # write symbolic matrix and replace symbolic variable\",\n \"a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return False,\",\n \"0 # loop on all terms for cur_term in term_list: # detect products\",\n \"or vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name =\",\n \"out_file.write('\\\\n // jacobian anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"max: out_file.write(';\\\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write('\",\n \"j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac !=\",\n \"* // 18 15 22 //\\\\n') out_file.write(' * // 19 14 23 //\\\\n')\",\n \"18 15 22 # # 19 14 23 # # 20 01 24\",\n \"ground (info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di : position vector from\",\n \"out_file.write(' * omi : absolute rotational vector of body i\\\\n') out_file.write(' * Ri\",\n \"angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write('\",\n \"x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot =\",\n \"(1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the matrix elements\",\n \"-v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] =\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}]\",\n \"rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n // -- symbolic computation -- //\\\\n') #\",\n \"out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot\",\n \"# write symbolic vector and replace symbolic variable by its name def write_symb_vector(out_file,\",\n \"-- //\\\\n') # Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj\",\n \"i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot\",\n \"\\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n')\",\n \"03 09 //\\\\n') out_file.write(' * // 04 10 //\\\\n') out_file.write(' * // 05\",\n \"string for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split)\",\n \"!= None: if not flag_first: flag_first = 1 flag_print = 1 elif not\",\n \"i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] =\",\n \"1, 3, 2, 1, 2, # right leg 2, 1, 3, 2, 1,\",\n \"2, 1, 2, # left leg 1, 2, 3, # trunk 2, 1,\",\n \"int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except: return False, -1,\",\n \"= x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] =\",\n \"vector (from origin, expressed in the inertial frame) # of the anchor point\",\n \"foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n')\",\n \"out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n // joint absolute velocities\\\\n') Om = nb_bodies*[None]\",\n \"for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None]\",\n \"= R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x\",\n \"x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id]\",\n \"range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0:\",\n \"origin, expressed in the inertial frame)\\\\n') out_file.write(' * of the anchor point of\",\n \"= str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 # loop\",\n \"out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part,\",\n \"COM G_i of body i # xpi : derivative of xi # xgpi\",\n \"jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void\",\n \"relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write(' * DGi : position\",\n \"vector (from origin, expressed in the inertial frame)\\\\n') out_file.write(' * of the anchor\",\n \"[sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis\",\n \"= sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for\",\n \"range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n') out_file.write('\",\n \"i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\\\n') else: out_file.write(',\",\n \"// joint absolute velocities\\\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None]\",\n \"4, 5, # right leg 0, 7, 8, 9, 10, 11, # left\",\n \"0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0,\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}]\",\n \"R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): #\",\n \"in the file def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as\",\n \"= symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) #\",\n \"range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id]\",\n \"- rotation matrices\\\\n') for i in range(0, 3): for j in range(0, 3):\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"term_list.pop(0) result = 0 # loop on all terms for cur_term in term_list:\",\n \"velocity\\\\n') out_file.write(' * feet position, velocity and orientation\\\\n') out_file.write(' * waist and torso\",\n \"xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n // jacobian anchor\",\n \"100) out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp)\",\n \"i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] *\",\n \"nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies,\",\n \"{\\\\n') for i in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}]\",\n \"os.remove(in_temp) # main script # rotation axis for each joint before body i\",\n \"trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' #\",\n \"j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont\",\n \"# variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis,\",\n \"declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >=\",\n \"// left foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') #\",\n \"1: # positive neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2:\",\n \"sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n\",\n \"# write symbolic matrix and replace symbolic variable by its name def write_symb_matrix(out_file,\",\n \"== 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif\",\n \"out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} +\",\n \"in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1\",\n \"0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'),\",\n \"right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint names\",\n \"+ Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i])\",\n \"+ s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out,\",\n \"get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get\",\n \"xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om,\",\n \"for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left\",\n \"* computation of:\\\\n') out_file.write(' * COM (center of mass) position and velocity\\\\n') out_file.write('\",\n \"COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) #\",\n \"'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ])\",\n \"sp import re import os ###################### # # # 17 16 21 #\",\n \"file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on\",\n \"j+1)) if i == 2 and j == 2: out_file.write(';\\\\n') else: out_file.write(', ')\",\n \"of xi\\\\n') out_file.write(' * xgpi : derivative of xgi\\\\n') out_file.write(' * omi :\",\n \"/ c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist')\",\n \"out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot\",\n \"the lines for line in f: # declaration if len(line.split('// -- variables initialization\",\n \"3 * [None] # rotation matrices out_file.write('\\\\n // rotation matrices\\\\n') R = nb_bodies*[None]\",\n \"variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in\",\n \"write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on all the joints\",\n \"orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for\",\n \"j in range (1, nb_bodies): flag_print = 0 for k in range(0, 3):\",\n \"(R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) #\",\n \"1: new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name)\",\n \"in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp,\",\n \"as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write('\",\n \"= cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for i in\",\n \"[xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com\",\n \"else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\\\n') count =\",\n \"absolute position\\\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first =\",\n \"the origin (waist), but aligned with the ground (info from IMU) # #\",\n \"# loop on all the lines for line in f: cut_line_1 = line.split(elem)\",\n \"2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else:\",\n \"out_file.write(' * // 02 08 //\\\\n') out_file.write(' * // 03 09 //\\\\n') out_file.write('\",\n \"the ground (info from IMU) # # Di : position vector from the\",\n \"omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0]))\",\n \"in range(0,3): for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] ==\",\n \"absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write('\",\n \"-1 # count the number of 'elem' in the file def count_elem(in_file, elem):\",\n \"# no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj,\",\n \"joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names,\",\n \"= sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for print def\",\n \"waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg\",\n \"1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name,\",\n \"= xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot =\",\n \"1 if count >= nb_max_line: out_write.write(';\\\\n') count = 0 if count != 0:\",\n \"sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine,\",\n \", 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , #\",\n \"for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac\",\n \"7, 8, 9, 10, 11, # left leg 0, 13, 14, # trunk\",\n \"len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double\",\n \"c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot\",\n \"3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies):\",\n \"R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i]\",\n \"14, # trunk 15, 16, 17, 18, # right arm 15, 20, 21,\",\n \"waist, middle point between the two pitch hip rotations # inertial frame: located\",\n \"def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp\",\n \"// left foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"+ {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j ==\",\n \"* xi : absolute position vector (from origin, expressed in the inertial frame)\\\\n')\",\n \"22 //\\\\n') out_file.write(' * // 19 14 23 //\\\\n') out_file.write(' * // 20\",\n \"// right wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"out_file.write(' * inertial frame: located at the origin (waist), but aligned with the\",\n \"the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M):\",\n \"3, 2, 1, 2, # left leg 1, 2, 3, # trunk 2,\",\n \"before body i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi : absolute position vector (from\",\n \", sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'),\",\n \"its predecessor\\\\n') out_file.write(' * si : sine of the relative angle before body\",\n \"!= 0: out_write.write(';\\\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]],\",\n \"declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for j\",\n \"i-1), expressed in the relative # frame of the previous body # Rdi\",\n \"*\\\\n') out_file.write(' * computation of:\\\\n') out_file.write(' * COM (center of mass) position and\",\n \"3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{}\",\n \"om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist',\",\n \"= open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on all the\",\n \"range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector,\",\n \"Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14]\",\n \"out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17]\",\n \"s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso\",\n \"0: if not flag_first: flag_first = 1 flag_print = 1 elif not flag_print:\",\n \"in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac !=\",\n \"end_name) save_vector[i] = ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector,\",\n \"{};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute velocity\\\\n') for i in range(0,3):\",\n \"in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch\",\n \"-0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output\",\n \"of body i to its COM (center of mass) G_i,\\\\n') out_file.write(' * expressed\",\n \"center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp,\",\n \"(center of mass) G_i, # expressed in the relative frame of the current\",\n \"out_file.write(' // right foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n')\",\n \"return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation\",\n \"out_file.write(' * feet position, velocity and orientation\\\\n') out_file.write(' * waist and torso orientaion\",\n \"loop on all terms for cur_term in term_list: # detect products cur_term_split =\",\n \"for i in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i]\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot orientation matrix as angles\",\n \"absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write('\",\n \"sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19]\",\n \"np import sympy as sp import re import os ###################### # # #\",\n \"len(cut_line) == 2: count += 1 return count # print the declaration of\",\n \"(waist), but aligned with the ground (info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' *\",\n \"1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] =\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg &\",\n \"the COM G_i of body i # xpi : derivative of xi #\",\n \"compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0]\",\n \"derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len =\",\n \"in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split =\",\n \"3*i+j)) out_file.write('\\\\n // IMU - angles velocity\\\\n') for i in range(0, 3): out_file.write('\",\n \"// jacobian anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"com absolute positions and velocities\\\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i\",\n \"0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] =\",\n \"count = 0 if count != 0: out_write.write(';\\\\n') # print all declarations def\",\n \"line in f: cut_line = line.split(' = ') if len(cut_line) == 2: if\",\n \"13, 14, # trunk 15, 16, 17, 18, # right arm 15, 20,\",\n \"out_file.write(' * of the anchor point of body i\\\\n') out_file.write(' * xgi :\",\n \"out_file.write('\\\\n') out_file.write(' // left foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id],\",\n \"to the current body i \\\\n') out_file.write(' * (previous body is not always\",\n \"= sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0,\",\n \"range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None,\",\n \"= sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0]\",\n \"# get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis\",\n \"out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for i in range(1,\",\n \"xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic\",\n \">= 1: count = 0 with open(in_file,'r') as f: # loop on all\",\n \"file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp,\",\n \"'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15',\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"+= 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count =\",\n \"Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2]\",\n \"//\\\\n') out_file.write(' * // 18 15 22 //\\\\n') out_file.write(' * // 19 14\",\n \"16, 17, 18, # right arm 15, 20, 21, 22 # left arm\",\n \"pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon >\",\n \"* Ri : absolute rotational matrix\\\\n') out_file.write(' * Rti : transpose matrix of\",\n \"s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part,\",\n \"= sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0,\",\n \"sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9\",\n \"sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0,\",\n \"i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '')\",\n \"a, b except: return False, -1, -1 # count the number of 'elem'\",\n \"Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i\",\n \"sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] =\",\n \"= Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None] #\",\n \"the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var):\",\n \"# left arm ]) # joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id',\",\n \"body i # # xi : absolute position vector (from origin, expressed in\",\n \"i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector\",\n \"# frame of the previous body # DGi : position vector from the\",\n \"0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'),\",\n \"# compute the derivative of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj,\",\n \"om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] =\",\n \"out_file.write(' // left wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] =\",\n \"-0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,\",\n \"xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06,\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]]\",\n \"R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i],\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"= cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for i in range(1, nb_bodies): out_file.write('\",\n \"xgj, der_var) # one product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj,\",\n \"Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions\",\n \"left wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i]))\",\n \"= symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first:\",\n \"write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print,\",\n \"flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n')\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac))\",\n \"out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write('\",\n \"# negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead of\",\n \"out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count = 0 for i in\",\n \"','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 # loop on all\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute position\\\\n') for\",\n \"xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"cosines\\\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n //\",\n \"name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print = 0\",\n \"(c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} -\",\n \"+ Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i])\",\n \"sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8' ,\",\n \"transpose matrix of Ri # xji : jacobian of 'xi' # xgji :\",\n \"+ Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i])\",\n \"/ c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle\",\n \"b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except: return\",\n \"= sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'),\",\n \"0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0,\",\n \"out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write('\",\n \"the previous body\\\\n') out_file.write(' * DGi : position vector from the anchor point\",\n \"script # rotation axis for each joint before body i (1:x, 2:y, 3:z)\",\n \"Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation\",\n \"left foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute\",\n \"'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' ,\",\n \"i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j\",\n \"anchor point absolute positions and velocities\\\\n') x = nb_bodies*[None] xp = nb_bodies*[None] x[0]\",\n \"leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] =\",\n \"def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms term_list =\",\n \"out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of\",\n \"write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print,\",\n \"# right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19',\",\n \"positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_bodies):\",\n \"or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name =\",\n \"// feet absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i,\",\n \"j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j ==\",\n \"flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j],\",\n \"angle before body i # # xi : absolute position vector (from origin,\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] =\",\n \"0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'),\",\n \"12 //\\\\n') out_file.write(' * // 07 13 //\\\\n') out_file.write(' * // //\\\\n') out_file.write('\",\n \"end_name): new_matrix = sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for\",\n \"== '': term_list.pop(0) result = 0 # loop on all terms for cur_term\",\n \"out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write('\",\n \"= sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0,\",\n \"absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write('\",\n \"int(value) return True except: return False # return true if it has a\",\n \"the COM G_i of body i\\\\n') out_file.write(' * xpi : derivative of xi\\\\n')\",\n \"os ###################### # # # 17 16 21 # # 18 15 22\",\n \"index): # loop on all the joints for i in range (1, nb_bodies):\",\n \"15, 16, 17, 18, # right arm 15, 20, 21, 22 # left\",\n \"relative # frame of the previous body # DGi : position vector from\",\n \"out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist absolute velocity\\\\n') for\",\n \"waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id'\",\n \"[0.0], [-elem]]) else: return np.array([]) # compute the derivative of an element (for\",\n \"compute the derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj,\",\n \"in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1\",\n \"<NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics computation for the\",\n \"out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n')\",\n \"of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic\",\n \"matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] =\",\n \"return True, a, b, c except: return False, -1, -1, -1 # return\",\n \"out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute\",\n \"atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle in\",\n \"// 06 12 //\\\\n') out_file.write(' * // 07 13 //\\\\n') out_file.write(' * //\",\n \"out_file.write(' *\\\\n') out_file.write(' * xi : absolute position vector (from origin, expressed in\",\n \"sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] =\",\n \"24 //\\\\n') out_file.write(' * // 02 08 //\\\\n') out_file.write(' * // 03 09\",\n \"jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_contacts):\",\n \"else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com absolute velocity\\\\n') for i in range(0,\",\n \"Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')])\",\n \"out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot absolute velocity\\\\n') for\",\n \"nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first =\",\n \"out_file.write(' // right foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None\",\n \"cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len == 1:\",\n \"0.0], [0.0, cosine, sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine,\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i\",\n \"l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot\",\n \"if it has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def\",\n \"0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) #\",\n \"== 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem],\",\n \"0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name\",\n \"in f: # declaration if len(line.split('// -- variables initialization -- //')) != 1:\",\n \"# positive neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2: #\",\n \"2: # negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead\",\n \"* Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0],\",\n \"absolute position vector (from origin, expressed in the inertial frame)\\\\n') out_file.write(' * of\",\n \"= xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts):\",\n \"re import os ###################### # # # 17 16 21 # # 18\",\n \"// right wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i,\",\n \": sine of the relative angle before body i\\\\n') out_file.write(' * ci :\",\n \"start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i in range(0,3):\",\n \"'') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\\\n // anchor point absolute\",\n \"Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1),\",\n \"results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R,\",\n \"R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of\",\n \": jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n')\",\n \"x_vector, index): # loop on all the joints for i in range (1,\",\n \"20 01 24 # # 02 08 # # 03 09 # #\",\n \"R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i\",\n \"and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id,\",\n \"Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')])\",\n \"om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist =\",\n \"# 02 08 # # 03 09 # # 04 10 # #\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right\",\n \"compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp,\",\n \"3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj,\",\n \"out_file.write('\\\\n // anchor point absolute positions and velocities\\\\n') x = nb_bodies*[None] xp =\",\n \"np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis == 1: return np.array([[-elem],\",\n \"elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0,\",\n \"05 11 # # 06 12 # # 07 13 # # #\",\n \"true if it has a shape 'R%a_%b%c' (indexes %a, %b, %c also returned)\",\n \"on all terms for cur_term in term_list: # detect products cur_term_split = cur_term.split('*')\",\n \"symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot\",\n \"of the current body i\\\\n') out_file.write(' * Omi : rotational vector from the\",\n \"Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15)\",\n \"range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] +\",\n \"= atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle\",\n \"angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity\",\n \"out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_contacts): for\",\n \"xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k] !=\",\n \"2 and j == 2: out_file.write(';\\\\n') else: out_file.write(', ') # print variables declaration\",\n \"12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,\",\n \"// torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the\",\n \"out_file.write(' // left foot contact points absolute position\\\\n') for i in range(0, nb_contacts):\",\n \"Dpt, Dg, M): out_file.write('\\\\n\\\\n // -- symbolic computation -- //\\\\n') # Rj, xj,\",\n \"sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] =\",\n \"of xi # xgpi : derivative of xgi # omi : absolute rotational\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] =\",\n \", 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14',\",\n \"* (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist absolute position\\\\n')\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] =\",\n \"= 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i]))\",\n \"= sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16]\",\n \"# return true if it has a shape 'R%a_%b%c' (indexes %a, %b, %c\",\n \"elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative\",\n \"{};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"prefix, min, max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i))\",\n \"0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine,\",\n \"0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left\",\n \"out_file.close() # remove temporary file os.remove(in_temp) # main script # rotation axis for\",\n \"out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation angle derivatives\",\n \"np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative of an element\",\n \"no product if cur_len == 1: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)\",\n \"sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] =\",\n \"0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1))\",\n \"= nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print =\",\n \"leg 1, 2, 3, # trunk 2, 1, 3, 2, # right arm\",\n \"xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id]\",\n \"direct, cosine, sine): if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0],\",\n \"2: count += 1 return count # print the declaration of an element\",\n \"= {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot contact points jacobian\\\\n') out_file.write('\",\n \"out_file.write(' * frame of the previous body\\\\n') out_file.write(' * DGi : position vector\",\n \"for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id]\",\n \"x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb\",\n \"= cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot =\",\n \"= sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0,\",\n \"# origin: in the waist, middle point between the two pitch hip rotations\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \", 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' ,\",\n \"the waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' //\",\n \"for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac\",\n \"xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n // jacobian com absolute positions\\\\n') out_file.write('\",\n \"out_file.write(' * // 07 13 //\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * ////////////////////////\\\\n')\",\n \"theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the wrists position, velocity and orientation\",\n \"in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end\",\n \"om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7]\",\n \"if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1\",\n \"Rti : transpose matrix of Ri\\\\n') out_file.write(' * xji : jacobian of \\\\'xi\\\\'\\\\n')\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i,\",\n \"file def count_elem(in_file, elem): count = 0; with open(in_file, 'r') as f: #\",\n \"range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU - angles\",\n \"matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso orientation\",\n \"symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \"of 'xi' # xgji : jacobian of 'xgi' # Rji : jacobian of\",\n \"18 15 22 //\\\\n') out_file.write(' * // 19 14 23 //\\\\n') out_file.write(' *\",\n \"the previous body to the current body i # (previous body is not\",\n \"= x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot =\",\n \"Rj, xj, xgj, xj_print, x_vector, index): # loop on all the joints for\",\n \"global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i in range(1, nb_bodies): for\",\n \"nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for k in range(0,\",\n \"out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left\",\n \"out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute\",\n \"list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0)\",\n \"out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary file\",\n \"* (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2]))\",\n \"loop on all the lines for line in f: cut_line = line.split(' =\",\n \"xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i],\",\n \"1, 3, 2 # left arm ]) # parent index parent_body_index = np.array([\",\n \"in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute orientation jacobian\\\\n') out_file.write('\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\\\n')\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot contact points absolute position\\\\n') for i in\",\n \"print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i\",\n \"= matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write('\",\n \"xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for\",\n \"i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i])))\",\n \"outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n') out_file.write(' * COM (center of\",\n \"== nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n')\",\n \"def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for j in\",\n \"= {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i,\",\n \"as f: # loop on all the lines for line in f: cut_line_1\",\n \"or 1 !'.format(cur_len-1)) exit() return result # write the beginning of the file\",\n \"nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic\",\n \"# jacobian rotation matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"waist 2, 1, 3, 2, 1, 2, # right leg 2, 1, 3,\",\n \"0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'),\",\n \"right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ])\",\n \"new_vector[i] = vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i]\",\n \"Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj,\",\n \"right foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"True, a, b, c except: return False, -1, -1, -1 # return true\",\n \"in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n')\",\n \"Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable by\",\n \"cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper()\",\n \"def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for\",\n \"jacobian xg out_file.write('\\\\n // jacobian com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"// -- Collecting results -- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj)\",\n \"symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the file\",\n \"0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return\",\n \"out_file.write(' double ') for i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1,\",\n \"of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write('\",\n \"variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb =\",\n \"and replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector\",\n \"kinematics computation for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {')\",\n \"waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist\",\n \"0 for i in range (1, nb_bodies): flag_print = 0 for j in\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}]\",\n \"nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n')\",\n \"new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write('\",\n \"is not always body i-1), expressed in the relative\\\\n') out_file.write(' * frame of\",\n \"'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso',\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"a float def isInt(value): try: int(value) return True except: return False # return\",\n \"range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for i\",\n \"'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies):\",\n \"body i-1), expressed in the relative\\\\n') out_file.write(' * frame of the previous body\\\\n')\",\n \"R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle in [pi/2 ;\",\n \"out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1]\",\n \"out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"{};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector # write symbolic\",\n \"= 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else:\",\n \"== 1: # positive neg_flag = 0 pos_str = elem_split[0] elif cur_len ==\",\n \"# frame of the previous body # Rdi : rotational matrix between body\",\n \"; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{},\",\n \"# loop on all the joints for i in range (1, nb_bodies): new_vector\",\n \"= sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] =\",\n \"torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as\",\n \"jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index):\",\n \"22 # # 19 14 23 # # 20 01 24 # #\",\n \"absolute velocities\\\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] =\",\n \"= sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg\",\n \"the lines for line in f: cut_line = line.split(' = ') if len(cut_line)\",\n \"= get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj,\",\n \"# end of the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name),\",\n \"= sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')])\",\n \"in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot\",\n \"+= -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other\",\n \"out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i\",\n \"11, # left leg 0, 13, 14, # trunk 15, 16, 17, 18,\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0] =\",\n \"yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n')\",\n \"xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian\",\n \"0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index,\",\n \"[0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine,\",\n \"save_vector[i] = ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector\",\n \"// right foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"body\\\\n') out_file.write(' * DGi : position vector from the anchor point of body\",\n \"wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range\",\n \"= sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write('\",\n \"out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute orientation jacobian\\\\n')\",\n \"located at the origin (waist), but aligned with the ground (info from IMU)\",\n \"* [None] for i in range(0,3): if vector[i] == 0 or vector[i] ==\",\n \"elem_str.replace('- ','-').strip() # derivative axis der_q = int(der_var.replace('q','')) # detect positive/negative elem_split =\",\n \"in the waist, middle point between the two pitch hip rotations\\\\n') out_file.write(' *\",\n \"out_file.write('\\\\n // rotation matrices\\\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None]\",\n \"for j in range(1, nb_bodies): flag_print = 0 for k in range(0, 3):\",\n \"its COM (center of mass) G_i,\\\\n') out_file.write(' * expressed in the relative frame\",\n \"j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n')\",\n \"cur_split[i].upper()) return new_string # write the end of the file def write_file_end(out_file): out_file.write('}\\\\n')\",\n \"= sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot =\",\n \"nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None]\",\n \"3, 2, # right arm 2, 1, 3, 2 # left arm ])\",\n \"and its predecessor\\\\n') out_file.write(' * si : sine of the relative angle before\",\n \"-{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0],\",\n \"i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n')\",\n \"range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first\",\n \"return count # print the declaration of an element def print_declaration_elem(in_file, out_write, elem,\",\n \"out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\\\n') count\",\n \"rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp\",\n \"write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on all the joints\",\n \"jacobian anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n') for i in range(0,\",\n \"* (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot)\",\n \"out_file.write(' }\\\\n') # omega out_file.write('\\\\n // joint absolute velocities\\\\n') Om = nb_bodies*[None] om\",\n \"nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp)\",\n \"elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute\",\n \"the joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) #\",\n \"* Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i],\",\n \"rot_axis = np.array([0, # waist 2, 1, 3, 2, 1, 2, # right\",\n \"1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1,\",\n \"= sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts *\",\n \"range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] =\",\n \"an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count\",\n \"out_file.write('\\\\n') out_file.write(' // left foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}]\",\n \"if len(cut_line) == 2: count += 1 return count # print the declaration\",\n \"xp out_file.write('\\\\n // anchor point absolute positions and velocities\\\\n') x = nb_bodies*[None] xp\",\n \"'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n // jacobian com absolute positions\\\\n') out_file.write(' if\",\n \"= 0 pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag =\",\n \"elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1))\",\n \"s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if\",\n \"'_d{}'.format(i+1)) # write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj,\",\n \"out_file.write(' // right foot contact points absolute position\\\\n') for i in range(0, nb_contacts):\",\n \"i in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1,\",\n \"sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')])\",\n \"out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write('\",\n \"for i in range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i]\",\n \"derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} *\",\n \"temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the\",\n \"%a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1])\",\n \"for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write\",\n \"absolute velocity\\\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first =\",\n \"xgj_print, x_vector, index): # loop on all the joints for i in range\",\n \"= './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names)\",\n \"0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute orientation jacobian\\\\n') out_file.write(' if\",\n \": position vector from the anchor point of body i to its COM\",\n \"om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\\\n // anchor point absolute positions\",\n \"0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine,\",\n \"write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print = 0 for i\",\n \"flag_print = 0 for i in range(0,3): for j in range(0,3): if matrix[i,j]\",\n \"1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] =\",\n \"save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {}\",\n \"') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1:\",\n \"// left wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i,\",\n \"range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the matrix\",\n \"position and velocity\\\\n') out_file.write(' * feet position, velocity and orientation\\\\n') out_file.write(' * waist\",\n \"the two pitch hip rotations # inertial frame: located at the origin (waist),\",\n \"06 12 # # 07 13 # # # ###################### # # origin:\",\n \"8, 9, 10, 11, # left leg 0, 13, 14, # trunk 15,\",\n \"negative detection !'.format(cur_len)) exit() # compute derivative result = 0 # cosine if\",\n \"r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb =\",\n \"np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12',\",\n \"with the ground (info from IMU) # # Di : position vector from\",\n \"end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return\",\n \"out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg & xgp out_file.write('\\\\n // com absolute positions\",\n \"its COM (center of mass) G_i, # expressed in the relative frame of\",\n \"c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n')\",\n \"cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string =\",\n \"right leg 0, 7, 8, 9, 10, 11, # left leg 0, 13,\",\n \"in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k,\",\n \"parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp =\",\n \"foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if\",\n \"'Ri' # return true if it is a float def isInt(value): try: int(value)\",\n \"cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine,\",\n \"orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso\",\n \"matrix # Rti : transpose matrix of Ri # xji : jacobian of\",\n \"foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n')\",\n \"for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left\",\n \"of 'elem' in the file def count_elem(in_file, elem): count = 0; with open(in_file,\",\n \"axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]])\",\n \"Rj, xj, xgj, der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip()\",\n \"out_file.write(' * // 06 12 //\\\\n') out_file.write(' * // 07 13 //\\\\n') out_file.write('\",\n \"3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if\",\n \"xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j]\",\n \"sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'),\",\n \"1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i],\",\n \"= sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13]\",\n \"the relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write(' * Rdi :\",\n \"cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n')\",\n \"for line in f: # declaration if len(line.split('// -- variables initialization -- //'))\",\n \"* // 04 10 //\\\\n') out_file.write(' * // 05 11 //\\\\n') out_file.write(' *\",\n \"out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n')\",\n \"the relative # frame of the previous body # Rdi : rotational matrix\",\n \"{}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if\",\n \"cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string #\",\n \"Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z #\",\n \"10 # # 05 11 # # 06 12 # # 07 13\",\n \"symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector\",\n \"# waist 2, 1, 3, 2, 1, 2, # right leg 2, 1,\",\n \"r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb +\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}]\",\n \"if xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1\",\n \"compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp,\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg =\",\n \"s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if\",\n \"len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\",\",\n \"# 03 09 # # 04 10 # # 05 11 # #\",\n \"+ {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and\",\n \"Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other\",\n \"out_file.write(' // left foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] =\",\n \"x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r =\",\n \"* Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l =\",\n \"nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0]\",\n \"range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in\",\n \"xj, xgj, Rj_print, R_matrix, index): # loop on all the joints for i\",\n \"cosine of the relative angle before body i # # xi : absolute\",\n \"= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1))\",\n \"symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb\",\n \"com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i in range(1, nb_bodies): for j\",\n \"out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if\",\n \"inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' //\",\n \"= len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] =\",\n \"sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) ||\",\n \"parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n // -- symbolic computation -- //\\\\n') # Rj,\",\n \"* ////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * // 17 16 21 //\\\\n')\",\n \"Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm\",\n \"'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right\",\n \"joints enumeration\\\\n') out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)): count\",\n \"sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] =\",\n \"as sp import re import os ###################### # # # 17 16 21\",\n \"'{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f: # loop on\",\n \"loop on all the lines for line in f: cut_line = line.split(elem) if\",\n \"02 08 //\\\\n') out_file.write(' * // 03 09 //\\\\n') out_file.write(' * // 04\",\n \"the anchor point of body i to its COM (center of mass) G_i,\\\\n')\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute orientation jacobian\\\\n') out_file.write('\",\n \"in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first = 1\",\n \"return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif\",\n \"R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\\\n') for\",\n \"in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"xj_print, x_vector, index): # loop on all the joints for i in range\",\n \"# # 18 15 22 # # 19 14 23 # # 20\",\n \"xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] =\",\n \"'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) #\",\n \"cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0,\",\n \"out_file.write(' * xi : absolute position vector (from origin, expressed in the inertial\",\n \"1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis\",\n \"+ ') out_file.write(' // global com absolute position\\\\n') for i in range(0, 3):\",\n \"(flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range (1, nb_bodies): flag_print =\",\n \"cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0,\",\n \"* // //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin: in the\",\n \"write symbolic matrix and replace symbolic variable by its name def write_symb_matrix(out_file, matrix,\",\n \"= nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0]\",\n \"wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n')\",\n \"'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names,\",\n \"xgpi : derivative of xgi # omi : absolute rotational vector of body\",\n \"# detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if\",\n \"3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if\",\n \"of xgi # omi : absolute rotational vector of body i # Ri\",\n \"relative frame of the current body i\\\\n') out_file.write(' * Omi : rotational vector\",\n \"f: cut_line = line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) ==\",\n \"= sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7'\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg\",\n \"* (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot +\",\n \"Rti : transpose matrix of Ri # xji : jacobian of 'xi' #\",\n \"xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12,\",\n \"x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id'\",\n \"sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2]\",\n \"nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] =\",\n \"right foot contact points absolute position\\\\n') for i in range(0, nb_contacts): for j\",\n \"components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3): for\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i,\",\n \"right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3]\",\n \"xg, xgp, xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i]))\",\n \"term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0 #\",\n \"Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None]\",\n \"c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso\",\n \"if cur_jac != 0: if not flag_first: flag_first = 1 flag_print = 1\",\n \"variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3)\",\n \"!= 0: if not flag_first: flag_first = 1 flag_print = 1 elif not\",\n \"not always body i-1), expressed in the relative # frame of the previous\",\n \"== 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2\",\n \"i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist\",\n \"xp[i] = xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj,\",\n \"sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8'\",\n \"# right arm 'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint\",\n \"count = 0 if count != 0: out_write.write(';\\\\n') # get tilde matrix def\",\n \"for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] =\",\n \"//\\\\n') out_file.write(' * // 07 13 //\\\\n') out_file.write(' * // //\\\\n') out_file.write(' *\",\n \"else: print('Error: {} instead of 1 or 2 in negative detection !'.format(cur_len)) exit()\",\n \"= nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] =\",\n \"x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T *\",\n \"# from an orientation matrix, compute the roll, pitch, yaw angles (and derivative)\",\n \"else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')\",\n \"middle point between the two pitch hip rotations # inertial frame: located at\",\n \"# # 03 09 # # 04 10 # # 05 11 #\",\n \"theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file,\",\n \"-result return result # compute the derivative of an expression (for jacobian) def\",\n \"Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'),\",\n \"if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print\",\n \"# of the anchor point of body i # xgi : absolute position\",\n \"* ci : cosine of the relative angle before body i\\\\n') out_file.write(' *\\\\n')\",\n \"# generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt,\",\n \"x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r\",\n \"print('Error: {} * counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit()\",\n \"return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 /\",\n \"sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0],\",\n \"; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{},\",\n \"orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left\",\n \"* Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute\",\n \"# Di : position vector from the anchor point of the previous body\",\n \": absolute position vector of the COM G_i of body i # xpi\",\n \"as np import sympy as sp import re import os ###################### # #\",\n \"'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q))\",\n \"i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the\",\n \"== 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0],\",\n \"replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector =\",\n \"matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n')\",\n \"rotation matrices out_file.write('\\\\n // rotation matrices\\\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd\",\n \"jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"* [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] =\",\n \"matrix between body i and its predecessor\\\\n') out_file.write(' * si : sine of\",\n \"nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in\",\n \"rotations\\\\n') out_file.write(' * inertial frame: located at the origin (waist), but aligned with\",\n \"out_file.write(' // global com absolute position\\\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}]\",\n \"body i-1), expressed in the relative # frame of the previous body #\",\n \"* xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji : jacobian of \\\\'Ri\\\\'\\\\n')\",\n \"new_vector, save_vector # save the symbolic matrix for print def print_save_symb_matrix(matrix, start_name, end_name):\",\n \"in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute velocity\\\\n') for i\",\n \"line in f: # declaration if len(line.split('// -- variables initialization -- //')) !=\",\n \"= sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write('\",\n \"out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n\",\n \"the relative angle before body i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi : absolute\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg\",\n \"out_file.write('\\\\n // IMU - rotation matrices\\\\n') for i in range(0, 3): for j\",\n \"torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output file def gen_symbolic_out(out_file_name,\",\n \"for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic\",\n \"xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] *\",\n \"R[i].T # jacobian rotation matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write('\",\n \"result = -result return result # compute the derivative of an expression (for\",\n \"out_file.write(' * ci : cosine of the relative angle before body i\\\\n') out_file.write('\",\n \"j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU\",\n \"-> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5],\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n //\",\n \"b except: return False, -1, -1 # count the number of 'elem' in\",\n \"= sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 * [None]\",\n \"in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist absolute velocity\\\\n') for i\",\n \"in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0:\",\n \"i in range (1, nb_bodies): flag_print = 0 for j in range(0,9): cur_jac\",\n \"wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y,\",\n \"sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i],\",\n \"nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None]\",\n \"flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n')\",\n \"out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot\",\n \"i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0]\",\n \"left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right\",\n \"range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot\",\n \"Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist\",\n \"# left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')])\",\n \"xp[0] = om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0],\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n // -- symbolic computation -- //\\\\n')\",\n \"out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\\\n')\",\n \"torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\\\n')\",\n \"out_file.write(' * xgi : absolute position vector of the COM G_i of body\",\n \"out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"the previous body # Rdi : rotational matrix between body i and its\",\n \"range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else:\",\n \"COM (center of mass) position and velocity\\\\n') out_file.write(' * feet position, velocity and\",\n \"waist, middle point between the two pitch hip rotations\\\\n') out_file.write(' * inertial frame:\",\n \"x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T *\",\n \"variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot = x[l_foot_id] xp_r_foot =\",\n \"17 16 21 //\\\\n') out_file.write(' * // 18 15 22 //\\\\n') out_file.write(' *\",\n \"foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"= nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0]\",\n \"computation -- //\\\\n') # Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None]\",\n \"= sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0]\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"= ') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) ==\",\n \"= om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot\",\n \"0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0],\",\n \"origin (waist), but aligned with the ground (info from IMU) # # Di\",\n \"0: out_write.write(';\\\\n') # print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0\",\n \"Omi : rotational vector from the previous body to the current body i\",\n \"= xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb\",\n \"xj, xgj, der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip() #\",\n \"def get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for\",\n \"== 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else:\",\n \"if it is a float def isInt(value): try: int(value) return True except: return\",\n \"* Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write('\",\n \"R_r_foot[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n')\",\n \"'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1])\",\n \"positive neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2: # negative\",\n \"]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist\",\n \", 'LeftElbPitch_id' # left arm ]) out_file_name = 'forward_kinematics' gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index,\",\n \"sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9')\",\n \"out_file.write(' // left foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n')\",\n \"for j in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n //\",\n \"declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f: #\",\n \"i in range(0,3): for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j]\",\n \"new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1))\",\n \"0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >=\",\n \"axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]])\",\n \"//\\\\n') out_file.write('\\\\n // IMU - rotation matrices\\\\n') for i in range(0, 3): for\",\n \"elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result\",\n \"for k in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first\",\n \"negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {} instead of 1\",\n \"sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1),\",\n \"# beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names)\",\n \"get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot contact points absolute position\\\\n')\",\n \"[rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the wrists position, velocity and\",\n \"in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] =\",\n \"Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb\",\n \"[0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else:\",\n \"x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R,\",\n \"for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for\",\n \"right wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i]))\",\n \"xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"j)) flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0:\",\n \"out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute\",\n \"!'.format(cur_len-1)) exit() return result # write the beginning of the file def write_file_beginning(out_file,\",\n \"angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot\",\n \"velocities\\\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] =\",\n \"symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms term_list = str(symb_var).replace('-\",\n \"* \\\\\\\\param[in,out] in_out inputs and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n')\",\n \"out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n') for i in range(1,\",\n \"nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i]\",\n \"try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a,\",\n \"of the previous body to the current body i \\\\n') out_file.write(' * (previous\",\n \"feet absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i]))\",\n \"omega out_file.write('\\\\n // joint absolute velocities\\\\n') Om = nb_bodies*[None] om = nb_bodies*[None] om_tilde\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left\",\n \"0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if\",\n \"Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0,\",\n \"{};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of\",\n \"0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] =\",\n \"frame of the previous body\\\\n') out_file.write(' * DGi : position vector from the\",\n \"Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1]\",\n \"= y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16')\",\n \"range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n') for i in range(0, 9):\",\n \"xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return result # compute\",\n \"return True, a, b except: return False, -1, -1 # count the number\",\n \"position\\\\n') for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}]\",\n \"out_file.write('\\\\n') # compute the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R,\",\n \"m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1:\",\n \"{} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the\",\n \"sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'),\",\n \"cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)):\",\n \"out_file.write('\\\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot')\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result\",\n \"c except: return False, -1, -1, -1 # return true if it has\",\n \"yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left foot orientation matrix as angles\",\n \"out_file.write('\\\\n // joint sines\\\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1,\",\n \"* [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot\",\n \"# right leg 0, 7, 8, 9, 10, 11, # left leg 0,\",\n \"# right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0])\",\n \"get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file,\",\n \"or 2 in negative detection !'.format(cur_len)) exit() # compute derivative result = 0\",\n \"sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist 'M_7' ,\",\n \"sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')])\",\n \"in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in range(0,\",\n \"sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0],\",\n \"return False, -1, -1, -1 # return true if it has a shape\",\n \"compute derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q): result +=\",\n \"Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', # waist\",\n \"i+1, end_name) save_vector[i] = ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return\",\n \"'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the wrists position, velocity and orientation def wrists_compute(out_file,\",\n \"def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None]\",\n \"left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j],\",\n \"Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0])\",\n \"= open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization\",\n \"jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji\",\n \"+= der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"the anchor point of the previous body to the current body i \\\\n')\",\n \"j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp =\",\n \"vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return\",\n \"# right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' ,\",\n \"sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')])\",\n \"a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b,\",\n \"2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj,\",\n \"# right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0])\",\n \"om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and orientation def feet_compute(out_file,\",\n \"0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1))\",\n \"= {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot absolute velocity\\\\n') for i in\",\n \"get a string for the enumeration of joints def get_string_enum(cur_string): cur_split = cur_string.split('_')\",\n \"axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0],\",\n \"02 08 # # 03 09 # # 04 10 # # 05\",\n \"the anchor point of body i\\\\n') out_file.write(' * xgi : absolute position vector\",\n \"point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) #\",\n \"except: return False, -1, -1, -1 # return true if it has a\",\n \"0.0, 1.0]]) else: return np.array([]) else: if axis == 1: return np.array([[1.0, 0.0,\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"= '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if\",\n \"nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all the matrix elements for\",\n \"21 # # 18 15 22 # # 19 14 23 # #\",\n \"in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part,\",\n \"3, 2, 1, 2, # right leg 2, 1, 3, 2, 1, 2,\",\n \"flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') #\",\n \"'{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print:\",\n \"* \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics computation\",\n \"out_file.write(' * // 18 15 22 //\\\\n') out_file.write(' * // 19 14 23\",\n \"i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j\",\n \"= sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j]\",\n \"anchor point of body i\\\\n') out_file.write(' * xgi : absolute position vector of\",\n \"x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] #\",\n \"= sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2]\",\n \"i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on all\",\n \"0 or 1 !'.format(cur_len-1)) exit() return result # write the beginning of the\",\n \"= 0 for i in range(0,3): if vector[i] == 0 or vector[i] ==\",\n \"range(1, nb_bodies): flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k] !=\",\n \"0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) #\",\n \"def isInt(value): try: int(value) return True except: return False # return true if\",\n \"matrices\\\\n') for i in range(0, 3): for j in range(0, 3): out_file.write(' IMU{}{}\",\n \"* // 03 09 //\\\\n') out_file.write(' * // 04 10 //\\\\n') out_file.write(' *\",\n \"exit() # compute derivative result = 0 # cosine if pos_str == 'c{}'.format(der_q):\",\n \"sp.zeros(3, 3) for i in range(0, 3): for j in range(0, 3): R[0][i,j]\",\n \"range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]]\",\n \"on all the lines for line in f: cut_line = line.split(elem) if len(cut_line)\",\n \"in f: cut_line = line.split(elem) if len(cut_line) == 2: count += 1 return\",\n \"vector of body i # Ri : absolute rotational matrix # Rti :\",\n \"of 1 or 2 in negative detection !'.format(cur_len)) exit() # compute derivative result\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic\",\n \"* Omi : rotational vector from the previous body to the current body\",\n \"nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1,\",\n \"angles velocity\\\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n\",\n \"for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop on\",\n \"= nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right leg Dpt[1]\",\n \"'M_10', 'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', #\",\n \"* // //\\\\n') out_file.write(' * // 17 16 21 //\\\\n') out_file.write(' * //\",\n \"Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0,\",\n \"0; with open(in_file, 'r') as f: # loop on all the lines for\",\n \"joint cosines\\\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n\",\n \"(1, nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj,\",\n \"1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1,\",\n \"before body i # ci : cosine of the relative angle before body\",\n \"com absolute position\\\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first\",\n \"22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \"left foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"out_file.write('\\\\n') return new_matrix # save the symbolic vector for print def print_save_symb_vector(vector, start_name,\",\n \"Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file,\",\n \"if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine elif pos_str == 's{}'.format(der_q):\",\n \"&in_out)\\\\n{\\\\n') # compute the center of mass position and velocity def com_compute(out_file, nb_bodies,\",\n \"result = 0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) #\",\n \"= sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n') for i in range(1, nb_bodies):\",\n \"out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write('\",\n \"else: # epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write('\",\n \"[0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3:\",\n \"shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1])\",\n \"replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix =\",\n \"j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for\",\n \"else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2:\",\n \"nb_bodies*[None] om = nb_bodies*[None] om_tilde = nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] =\",\n \"previous body\\\\n') out_file.write(' * Rdi : rotational matrix between body i and its\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"xi : absolute position vector (from origin, expressed in the inertial frame) #\",\n \"Rj, xj, xgj, der_var) # one product elif cur_len == 2: result +=\",\n \"not flag_first: flag_first = 1 flag_print = 1 elif not flag_print: flag_print =\",\n \"= 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1,\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double\",\n \"def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) #\",\n \"3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index),\",\n \"velocities\\\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i]\",\n \"wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n')\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"# right leg 2, 1, 3, 2, 1, 2, # left leg 1,\",\n \"sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ##\",\n \"rotation matrices\\\\n') for i in range(0, 3): for j in range(0, 3): out_file.write('\",\n \"axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]])\",\n \"R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon =\",\n \"-0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output file\",\n \"* \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics computation for the COMAN model\\\\n')\",\n \"xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb\",\n \"= write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies):\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') #\",\n \"01 24 # # 02 08 # # 03 09 # # 04\",\n \"from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di : position vector from the anchor\",\n \"epsilon = 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] =\",\n \"in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n')\",\n \"* // 20 01 24 //\\\\n') out_file.write(' * // 02 08 //\\\\n') out_file.write('\",\n \"nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot)\",\n \"R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist\",\n \"frame of the previous body\\\\n') out_file.write(' * Rdi : rotational matrix between body\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n\",\n \", sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] =\",\n \"[None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3,\",\n \"* Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l =\",\n \"sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10]\",\n \"derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * //\",\n \"x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb\",\n \"get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1],\",\n \"09 # # 04 10 # # 05 11 # # 06 12\",\n \"x_l_foot = x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot\",\n \"= {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot contact points jacobian\\\\n') out_file.write('\",\n \"point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on all\",\n \"'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , #\",\n \"elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: # positive neg_flag = 0\",\n \"in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1]\",\n \"in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies): if\",\n \"flag_first = 0 for i in range (1, nb_bodies): flag_print = 0 for\",\n \"= symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:\",\n \"d, e] = isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] #\",\n \"rotational matrix # Rti : transpose matrix of Ri # xji : jacobian\",\n \"'s{}'.format(der_q): result += sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str)\",\n \"Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3,\",\n \"// right foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i,\",\n \"the lines for line in f: cut_line = line.split(elem) if len(cut_line) == 2:\",\n \"xgp, xgj): out_file.write(' m_tot = ') for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if\",\n \"and velocities\\\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies):\",\n \"print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f: # loop on\",\n \"# # ###################### # # origin: in the waist, middle point between the\",\n \"in the inertial frame) # of the anchor point of body i #\",\n \"else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) #\",\n \"flag_print = 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j],\",\n \"sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if\",\n \"Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in\",\n \"for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left\",\n \"body i to its COM (center of mass) G_i, # expressed in the\",\n \"# symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n\",\n \"= int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c\",\n \"expressed in the relative frame of the current body i # Omi :\",\n \"x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] + om_tilde[parent_id] *\",\n \"all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result =\",\n \"i\\\\n') out_file.write(' * Omi : rotational vector from the previous body to the\",\n \"-- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary file\",\n \"if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1,\",\n \"get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l =\",\n \"%c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c\",\n \"if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i != nb_bodies-1:\",\n \"product elif cur_len == 2: result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result\",\n \"* waist and torso orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n')\",\n \"flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1))\",\n \"out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write('\",\n \"1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] = R[0].T for i in\",\n \"of the previous body to the current body i # (previous body is\",\n \"temporary file os.remove(in_temp) # main script # rotation axis for each joint before\",\n \"get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis ==\",\n \"2:y, 3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2, 1, 2,\",\n \"arm ]) # parent index parent_body_index = np.array([ -1, # waist 0, 1,\",\n \"foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n')\",\n \"= line.split(elem) if len(cut_line) == 2: count += 1 return count # print\",\n \"all the matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj,\",\n \"01 24 //\\\\n') out_file.write(' * // 02 08 //\\\\n') out_file.write(' * // 03\",\n \"range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0,\",\n \"'torso') # compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R,\",\n \"1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont\",\n \"i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist\",\n \"sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2]\",\n \"i to its COM (center of mass) G_i, # expressed in the relative\",\n \"# omega out_file.write('\\\\n // joint absolute velocities\\\\n') Om = nb_bodies*[None] om = nb_bodies*[None]\",\n \"= y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in\",\n \"the inertial frame) # of the anchor point of body i # xgi\",\n \"= {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian\",\n \"Rt[i] * Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies,\",\n \"contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"leg Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] =\",\n \"out_file.write(' // right wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] =\",\n \"print('Error: {} instead of 1 or 2 in negative detection !'.format(cur_len)) exit() #\",\n \"x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') #\",\n \"+= len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\\\n') count = 0 if\",\n \"* [None] for i in range(0,3): for j in range(0,3): if matrix[i,j] ==\",\n \"x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot)\",\n \"(from origin, expressed in the inertial frame)\\\\n') out_file.write(' * of the anchor point\",\n \"in range (1, nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the\",\n \"anchor point of body i to its COM (center of mass) G_i, #\",\n \"new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic matrix for print\",\n \"# remove temporary file os.remove(in_temp) # main script # rotation axis for each\",\n \"out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write('\",\n \"for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right\",\n \"instead of 1 or 2 in negative detection !'.format(cur_len)) exit() # compute derivative\",\n \"in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in\",\n \"= Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] =\",\n \"xgp out_file.write('\\\\n // com absolute positions and velocities\\\\n') xg = nb_bodies*[None] xgp =\",\n \"the relative frame of the current body i\\\\n') out_file.write(' * Omi : rotational\",\n \": rotational vector from the previous body to the current body i #\",\n \"xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot *\",\n \"in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute position\\\\n')\",\n \"in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1]\",\n \"k in range(0, 9): if Rj_print[i][j][k] != None: if not flag_first: flag_first =\",\n \"'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' ,\",\n \"= 0; with open(in_file, 'r') as f: # loop on all the lines\",\n \"0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write('\",\n \"the ground (info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di : position vector\",\n \"R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time\",\n \": jacobian of 'xi' # xgji : jacobian of 'xgi' # Rji :\",\n \"c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist\",\n \"matrices\\\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3,\",\n \"* \\\\\\\\brief forward kinematics computation for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints\",\n \"== 1: new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1,\",\n \"absolute position vector of the COM G_i of body i # xpi :\",\n \"= om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1)\",\n \"0 if count != 0: out_write.write(';\\\\n') # get tilde matrix def get_tilde(v): return\",\n \"c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for i in range(1, nb_bodies):\",\n \"R[0] = sp.zeros(3, 3) for i in range(0, 3): for j in range(0,\",\n \"* // 06 12 //\\\\n') out_file.write(' * // 07 13 //\\\\n') out_file.write(' *\",\n \"# compute the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x,\",\n \"i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n')\",\n \"= {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n') for\",\n \"{};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n') for i\",\n \"in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot contact points jacobian\\\\n')\",\n \"write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg,\",\n \"x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot + R_l_foot.T *\",\n \"jacobian of 'Ri' # return true if it is a float def isInt(value):\",\n \"// right foot contact points absolute position\\\\n') for i in range(0, nb_contacts): for\",\n \"der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] ==\",\n \"x_l[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"== 2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip()))\",\n \"!= 1: out_file.write(' // -- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n')\",\n \"*/\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count = 0 for i in range(1,\",\n \"sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n') for i in range(1, nb_bodies): out_file.write('\",\n \"-sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name,\",\n \"= x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist\",\n \"if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine,\",\n \"number of 'elem' in the file def count_elem(in_file, elem): count = 0; with\",\n \"nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T\",\n \"xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb =\",\n \"# waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg 'LeftHipPitch_id' ,\",\n \"of the relative angle before body i # ci : cosine of the\",\n \"12 # # 07 13 # # # ###################### # # origin: in\",\n \"(indexes %a, %b, %c also returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b\",\n \"in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first:\",\n \"body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) +\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') #\",\n \"sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector # write symbolic matrix and replace symbolic\",\n \"'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n // jacobian anchor point positions\\\\n') out_file.write(' if\",\n \"returned) def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1])\",\n \"sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'),\",\n \"1 return count # print the declaration of an element def print_declaration_elem(in_file, out_write,\",\n \"np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif axis == 3:\",\n \"i+1, j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print:\",\n \"in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2]\",\n \"joint relative velocities\\\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i]))\",\n \"right arm 15, 20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index)\",\n \"flag_first: flag_first = 1 flag_print = 1 elif not flag_print: flag_print = 1\",\n \"03 09 # # 04 10 # # 05 11 # # 06\",\n \"if count >= nb_max_line: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n')\",\n \"i in range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif\",\n \"i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for\",\n \"// left foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i,\",\n \"= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1))\",\n \"in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i]\",\n \"range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n // joint absolute\",\n \"'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk 'M_22',\",\n \"R_r_elb[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n')\",\n \"origin: in the waist, middle point between the two pitch hip rotations #\",\n \"before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist 2, 1,\",\n \"range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if\",\n \"pitch hip rotations # inertial frame: located at the origin (waist), but aligned\",\n \"direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0,\",\n \"if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 /\",\n \"}\\\\n\\\\n') out_file.write(' // left wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get a\",\n \"1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont =\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get a string for the enumeration of joints\",\n \"// 18 15 22 //\\\\n') out_file.write(' * // 19 14 23 //\\\\n') out_file.write('\",\n \"if len(cur_split) >= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string\",\n \"# # origin: in the waist, middle point between the two pitch hip\",\n \"0 for i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print =\",\n \"COM (center of mass) G_i, # expressed in the relative frame of the\",\n \"sine], [0.0, -sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0,\",\n \"count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f: # loop\",\n \"the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation\",\n \"symbolic variable by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3,\",\n \": position vector from the anchor point of the previous body to the\",\n \"out_file.write(line) out_file.close() # remove temporary file os.remove(in_temp) # main script # rotation axis\",\n \"symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3,\",\n \"result = 0 # loop on all terms for cur_term in term_list: #\",\n \"global com absolute position\\\\n') for i in range(0, 3): out_file.write(' in_out.r_COM[{}] = '.format(i))\",\n \"20 01 24 //\\\\n') out_file.write(' * // 02 08 //\\\\n') out_file.write(' * //\",\n \"write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\\\n //\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"07 13 //\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write('\",\n \"//\\\\n') # Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj =\",\n \"2 if count >= nb_max_char: out_write.write(';\\\\n') count = 0 if count != 0:\",\n \"by its name def write_symb_vector(out_file, vector, start_name, end_name): new_vector = sp.zeros(3, 1) flag_print\",\n \"for k in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first\",\n \"each joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, # waist\",\n \"body is not always body i-1), expressed in the relative\\\\n') out_file.write(' * frame\",\n \"in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n') for\",\n \"xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] =\",\n \"> 0: # epsilon = 1 -> pitch angle in [-pi/2 ; pi/2]\",\n \"in f: cut_line = line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('['))\",\n \"vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} =\",\n \"'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27',\",\n \"(for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string)\",\n \"xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i],\",\n \"'elem' in the file def count_elem(in_file, elem): count = 0; with open(in_file, 'r')\",\n \"= symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first:\",\n \"as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso orientation matrix\",\n \"flag_print = 0 for i in range(0,3): if vector[i] == 0 or vector[i]\",\n \"for i in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] =\",\n \"write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '')\",\n \"range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j]))\",\n \"14 23 # # 20 01 24 # # 02 08 # #\",\n \"i to its COM (center of mass) G_i,\\\\n') out_file.write(' * expressed in the\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute\",\n \"expressed in the inertial frame)\\\\n') out_file.write(' * of the anchor point of body\",\n \"r_wrist_y Dpt_r_wrist[2] = r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z\",\n \"# left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt =\",\n \"position vector of the COM G_i of body i\\\\n') out_file.write(' * xpi :\",\n \"xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] =\",\n \"xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only implemented\",\n \"'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} +\",\n \"xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] =\",\n \"variables initialization -- //\\\\n') out_file.write('\\\\n // IMU - rotation matrices\\\\n') for i in\",\n \"arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] #\",\n \"beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n')\",\n \"range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j]))\",\n \"save_matrix[3*i+j] = ' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix\",\n \"0 for k in range(0, 3): if xj_print[i][j][k] != None: if not flag_first:\",\n \"* counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result\",\n \"epsilon > 0: # epsilon = 1 -> pitch angle in [-pi/2 ;\",\n \"flag_print = 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj,\",\n \"write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write('\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n // -- Collecting\",\n \"vector from the anchor point of body i to its COM (center of\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute position\\\\n') for i in range(0,3): out_file.write('\",\n \"expressed in the relative # frame of the previous body # DGi :\",\n \"j+1, 3*i+j)) out_file.write('\\\\n // IMU - angles velocity\\\\n') for i in range(0, 3):\",\n \"in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and replace symbolic variable by its name\",\n \"om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute orientation jacobian\\\\n')\",\n \"joint_id_names[i])) out_file.write('\\\\n // joint sines\\\\n') for i in range(1, nb_bodies): out_file.write(' s{} =\",\n \"xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id]\",\n \"x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1,\",\n \"xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om,\",\n \"# # # 17 16 21 # # 18 15 22 # #\",\n \"for line in f: cut_line = line.split(' = ') if len(cut_line) == 2:\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac))\",\n \"print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in\",\n \"right wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i]))\",\n \"nb_bodies): flag_print = 0 for k in range(0, 9): if Rj_print[i][j][k] != None:\",\n \"# trunk 2, 1, 3, 2, # right arm 2, 1, 3, 2\",\n \"forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief forward kinematics computation for the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n')\",\n \"out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i])))\",\n \"position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,\",\n \"if count != 0: out_write.write(';\\\\n') # get tilde matrix def get_tilde(v): return np.array([[0.0,\",\n \"= nb_bodies*[None] Rd = nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3)\",\n \"matrix[i,j]).replace('1.0*','') new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a\",\n \"M): out_file.write('\\\\n\\\\n // -- symbolic computation -- //\\\\n') # Rj, xj, xgj and\",\n \"*\\\\n') out_file.write(' * origin: in the waist, middle point between the two pitch\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"Dpt_r_foot x_l = x_l_foot + R_l_foot.T * Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot\",\n \"symbolic matrix and replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name,\",\n \"return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [elem], [0.0]]) elif\",\n \"= 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j],\",\n \"x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute velocity\\\\n') for i in range(0,3): out_file.write('\",\n \"= sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix # save the symbolic vector for\",\n \"R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3,\",\n \"# loop on all the joints for i in range (1, nb_bodies): new_matrix\",\n \"vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] save_vector[i] = None\",\n \"= sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')]) Dpt[14] = sp.Matrix([0.0, 0.0, 0.0]) Dpt[15] = sp.Matrix([0.0, 0.0,\",\n \"# element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis der_q\",\n \"position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj,\",\n \"xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"* Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb +\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"for i in range(0,3): for j in range(0,3): if matrix[i,j] == 0 or\",\n \"absolute rotational matrix\\\\n') out_file.write(' * Rti : transpose matrix of Ri\\\\n') out_file.write(' *\",\n \"= sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17]\",\n \"= xp[parent_id] + om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print,\",\n \"body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the waist\",\n \"+ c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles\",\n \"out_file.write(' in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n') out_file.write(' if\",\n \"np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else:\",\n \"for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right\",\n \"range (1, nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac =\",\n \"axis def get_vector_axis(axis, direct, elem): if direct: if axis == 1: return np.array([[elem],\",\n \"write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id\",\n \"count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics\",\n \"= cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso = cos(in_out.theta_torso[1]);\\\\n') out_file.write(' c_z_waist = cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso =\",\n \"-1, # waist 0, 1, 2, 3, 4, 5, # right leg 0,\",\n \"der_elem(cur_term_split[0], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[1].strip()) result += der_elem(cur_term_split[1], Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) #\",\n \"1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0\",\n \"x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T *\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') #\",\n \"xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector,\",\n \"start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for i\",\n \"Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2] =\",\n \": rotational matrix between body i and its predecessor # si : sine\",\n \"sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')])\",\n \"-sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine], [0.0,\",\n \"om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot\",\n \"in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] =\",\n \"symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \"joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z):\",\n \"sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')])\",\n \"= write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1), '') Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\\\n\",\n \"rotation axis for each joint before body i (1:x, 2:y, 3:z) rot_axis =\",\n \"= c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} *\",\n \"position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \"R[0], 'R1_', '') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] =\",\n \"Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i]\",\n \"nb_bodies*[None] xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] *\",\n \"com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right wrist\",\n \"2: out_file.write(';\\\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min,\",\n \"in_out.rp_Lwrist[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"the two pitch hip rotations\\\\n') out_file.write(' * inertial frame: located at the origin\",\n \"xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk\",\n \"Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0]\",\n \"# right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8')\",\n \"int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return False, -1, -1\",\n \"# other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] =\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot contact points absolute position\\\\n') for i\",\n \"== 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]]) elif\",\n \"joint_id_names): out_file.write(' // -- variables initialization -- //\\\\n') out_file.write('\\\\n // IMU - rotation\",\n \"len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip()))\",\n \"range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist absolute velocity\\\\n')\",\n \"nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j],\",\n \"Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5]\",\n \"yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0:\",\n \"if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write('\",\n \"flag_print = 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj,\",\n \"9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute orientation\",\n \"range (1, nb_bodies): flag_print = 0 for j in range(0, 3): cur_jac =\",\n \"3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return\",\n \"range(0, nb_contacts): for j in range (1, nb_bodies): flag_print = 0 for k\",\n \"i)) if i == max: out_file.write(';\\\\n') else: out_file.write(', ') # variables initialization def\",\n \"sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] =\",\n \"k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac\",\n \"a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a =\",\n \"c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle derivatives\",\n \"# epsilon = 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write(' in_out.{}[0]\",\n \"import sympy as sp import re import os ###################### # # # 17\",\n \"cos(in_out.theta_waist[2]);\\\\n') out_file.write(' c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n')\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left\",\n \"in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l[j], 'q{}'.format(i+1)) if cur_jac !=\",\n \"flag_first = 0 for i in range(0, nb_contacts): for j in range (1,\",\n \"{};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"cosine]]) elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine,\",\n \"out_file.write(' // left foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"xgj, xgj_print, x_vector, index): # loop on all the joints for i in\",\n \"symbolic computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n //\",\n \"# x & xp out_file.write('\\\\n // anchor point absolute positions and velocities\\\\n') x\",\n \"in range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xj_print[i][j][k]\",\n \"Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0,\",\n \"'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20', 'M_21', # trunk\",\n \"def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on all the\",\n \"-- //\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x,\",\n \"// 20 01 24 //\\\\n') out_file.write(' * // 02 08 //\\\\n') out_file.write(' *\",\n \"3 * [None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\\\n //\",\n \"# trunk 15, 16, 17, 18, # right arm 15, 20, 21, 22\",\n \"{}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j ==\",\n \"i # xpi : derivative of xi # xgpi : derivative of xgi\",\n \"# compute the derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj,\",\n \"joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3) # loop\",\n \"// anchor point absolute positions and velocities\\\\n') x = nb_bodies*[None] xp = nb_bodies*[None]\",\n \"+ 2 if count >= nb_max_char: out_write.write(';\\\\n') count = 0 if count !=\",\n \"21, 22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions\",\n \"* Rti : transpose matrix of Ri\\\\n') out_file.write(' * xji : jacobian of\",\n \"relative angle before body i # ci : cosine of the relative angle\",\n \"and velocity def com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot =\",\n \"((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n')\",\n \"'M_11', 'M_12', # right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left\",\n \"True, a, b except: return False, -1, -1 # count the number of\",\n \"in f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) ==\",\n \"out_file.write(' * Rji : jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') #\",\n \"derivative of xi\\\\n') out_file.write(' * xgpi : derivative of xgi\\\\n') out_file.write(' * omi\",\n \"def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector, index): # loop on all the\",\n \"right foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i]))\",\n \"= sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0\",\n \"always body i-1), expressed in the relative # frame of the previous body\",\n \"vector from the previous body to the current body i # (previous body\",\n \"flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name,\",\n \"of the anchor point of body i # xgi : absolute position vector\",\n \"'{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name)\",\n \"R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else:\",\n \"3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0]\",\n \"write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '')\",\n \"if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, sine], [0.0, -sine,\",\n \"* [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None] #\",\n \"atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({},\",\n \"sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] =\",\n \"= 0 for k in range(0, 9): if Rj_print[i][j][k] != None: if not\",\n \"parent index parent_body_index = np.array([ -1, # waist 0, 1, 2, 3, 4,\",\n \"feet position, velocity and orientation\\\\n') out_file.write(' * waist and torso orientaion angles and\",\n \"vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector # write symbolic matrix\",\n \"= sp.zeros(3, 1) # loop on all the vector elements for j in\",\n \"c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if rot_flag:\",\n \"matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1]\",\n \"xgj, Rj_print, R_matrix, index): # loop on all the joints for i in\",\n \"np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis\",\n \"0 for i in range(0, nb_contacts): for j in range (1, nb_bodies): flag_print\",\n \"omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{})\",\n \"= {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot contact points\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}]\",\n \"foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' //\",\n \"out_file.write(' + ') out_file.write(' // global com absolute position\\\\n') for i in range(0,\",\n \"// left foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0\",\n \"Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7')\",\n \"Rj, xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj\",\n \"= nb_bodies*[None] Rd[0] = sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in\",\n \"positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0]) # right\",\n \"body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part,\",\n \"1, 3, 2, # right arm 2, 1, 3, 2 # left arm\",\n \"# waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7')\",\n \"Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk\",\n \"= get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l\",\n \"for k in range(0, 3): if xj_print[i][j][k] != None: if not flag_first: flag_first\",\n \"(R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T\",\n \"out_file.write(' }\\\\n') # xg & xgp out_file.write('\\\\n // com absolute positions and velocities\\\\n')\",\n \"// IMU - rotation matrices\\\\n') for i in range(0, 3): for j in\",\n \"= sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot)\",\n \"j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]]\",\n \"1) # loop on all the vector elements for j in range(0, 3):\",\n \"for i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] =\",\n \"out_file.write(' // right wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rwrist[{}] =\",\n \"xp_l = xp_l_foot + om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0,\",\n \"jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index):\",\n \"write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix,\",\n \"i == max: out_file.write(';\\\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies,\",\n \"atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{},\",\n \"velocity\\\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0\",\n \"= nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r\",\n \"== 2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line_2[0].strip()))\",\n \"= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('\",\n \"nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write(' // global com absolute position\\\\n') for\",\n \"# write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj,\",\n \"has a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a\",\n \"np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]]) # get rotation matrix\",\n \"{};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"Rj[i][j] = sp.zeros(3, 3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j]\",\n \"xgj[k][i][j] != 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n\",\n \"0, 1, 2, 3, 4, 5, # right leg 0, 7, 8, 9,\",\n \"//\\\\n') out_file.write(' * // 06 12 //\\\\n') out_file.write(' * // 07 13 //\\\\n')\",\n \"out_file.write('\\\\n') out_file.write(' // right foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] = sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'),\",\n \"= 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file,\",\n \"= {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"(position and derivative) of the waist and the torso def torso_waist_angles(out_file, R, om,\",\n \"cosine, sine): if direct: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0,\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}]\",\n \"1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n // joint absolute velocities\\\\n') Om\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}]\",\n \"def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_elb_id, l_elb_id, r_wrist_x,\",\n \"IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU - angles velocity\\\\n') for i\",\n \"nb_bodies*[None] xgp = nb_bodies*[None] for i in range(0, nb_bodies): xg[i] = x[i] +\",\n \"a shape 'R%a_%b%c' (indexes %a, %b, %c also returned) def isRot(value): try: a\",\n \"else: if axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0,\",\n \"foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n')\",\n \"get_string_enum(cur_string): cur_split = cur_string.split('_') if len(cur_split) >= 2: new_string = cur_split[0] for i\",\n \"9 * [None] for i in range(0,3): for j in range(0,3): if matrix[i,j]\",\n \"of body i # xgi : absolute position vector of the COM G_i\",\n \"true if it has a shape 'x%a_%b' (indexes %a, %b also returned) def\",\n \"elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if\",\n \"0.0], [sine, 0.0, cosine]]) elif axis == 3: return np.array([[cosine, sine, 0.0], [-sine,\",\n \"anchor point of the previous body to the current body i \\\\n') out_file.write('\",\n \"cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n')\",\n \"= R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1)\",\n \"i in range(1, nb_bodies): for j in range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]),\",\n \"for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts):\",\n \"in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j == 2:\",\n \"om[l_foot_id] R_r_foot = R[r_foot_id] R_l_foot = R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot =\",\n \"# rotation axis for each joint before body i (1:x, 2:y, 3:z) rot_axis\",\n \"19 14 23 # # 20 01 24 # # 02 08 #\",\n \"left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"-0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the symbolic output file def\",\n \"= sp.zeros(3, 1) for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file,\",\n \"= sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'), sp.Symbol('L_3_10')]) Dg[5] =\",\n \"!'.format(cur_len)) exit() # compute derivative result = 0 # cosine if pos_str ==\",\n \"return np.array([[0.0], [0.0], [elem]]) else: return np.array([]) else: if axis == 1: return\",\n \"om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i]\",\n \"out_file.write('\\\\n') out_file.write(' // left wrist absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lwrist[{}]\",\n \"0 with open(in_file,'r') as f: # loop on all the lines for line\",\n \"-- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line) out_file.close() #\",\n \"x out_file.write('\\\\n // jacobian anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"cur_len == 2: # negative neg_flag = 1 pos_str = elem_split[1] else: print('Error:\",\n \"rotational vector of body i\\\\n') out_file.write(' * Ri : absolute rotational matrix\\\\n') out_file.write('\",\n \"matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name,\",\n \"sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] =\",\n \"'w') with open(in_temp, 'r') as f: # loop on all the lines for\",\n \"else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\\\n')\",\n \"xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic\",\n \"left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9]\",\n \"* frame of the previous body\\\\n') out_file.write(' * Rdi : rotational matrix between\",\n \"origin (waist), but aligned with the ground (info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write('\",\n \"= 0 # cosine if pos_str == 'c{}'.format(der_q): result += -sp.Symbol('s{}'.format(der_q)) # sine\",\n \"detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if cur_len == 1: #\",\n \"# # 07 13 # # # ###################### # # origin: in the\",\n \"out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)): count += 1\",\n \"nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f:\",\n \"and torso orientaion angles and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\",\n \"j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else: flag_first\",\n \"the COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count = 0\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}]\",\n \"new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1))\",\n \"nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M)\",\n \"= cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist =\",\n \"body_part, body_part, omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the\",\n \"contact points absolute position\\\\n') for i in range(0, nb_contacts): for j in range(0,\",\n \"out_file.write(' // torso orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n')\",\n \"joint_id_names[i])) # write symbolic vector and replace symbolic variable by its name def\",\n \"y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2]\",\n \"for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i],\",\n \"cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0, sine],\",\n \"vector of body i\\\\n') out_file.write(' * Ri : absolute rotational matrix\\\\n') out_file.write(' *\",\n \"xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print =\",\n \"'M_29' # left arm ]) # joint names joint_id_names = np.array(['0', # waist\",\n \"(jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None]\",\n \"* Dg[i] xgp[i] = xp[i] + om_tilde[i] * (Rt[i] * Dg[i]) write_symb_xgj(nb_bodies, Rj,\",\n \"output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop\",\n \"om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] = get_tilde(om[0]) for i in range(1,\",\n \"//\\\\n') out_file.write(' * // 03 09 //\\\\n') out_file.write(' * // 04 10 //\\\\n')\",\n \"+ om_tilde_l_foot * (R_l_foot.T * Dpt_l_foot) for i in range(0, nb_contacts): x_r_cont[i] =\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n') for i in range(0, 9): out_file.write('\",\n \"in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n') for\",\n \"try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except: return\",\n \"isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return True, a, b except:\",\n \"elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1))\",\n \"write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '')\",\n \"3:z) rot_axis = np.array([0, # waist 2, 1, 3, 2, 1, 2, #\",\n \"'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left\",\n \"der_var): # element to derive (string) elem_str = elem_str.replace('- ','-').strip() # derivative axis\",\n \"located at the origin (waist), but aligned with the ground (info from IMU)\\\\n')\",\n \"beginning of the file write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) #\",\n \": absolute rotational vector of body i # Ri : absolute rotational matrix\",\n \"in negative detection !'.format(cur_len)) exit() # compute derivative result = 0 # cosine\",\n \"position vector (from origin, expressed in the inertial frame)\\\\n') out_file.write(' * of the\",\n \"= nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3) xj[i][j] =\",\n \"[rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation\",\n \"18, # right arm 15, 20, 21, 22 # left arm ]) nb_bodies\",\n \"right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18]\",\n \"i # # xi : absolute position vector (from origin, expressed in the\",\n \"matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i in range(0,3):\",\n \"i in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i,\",\n \"= cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string\",\n \": jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' *\",\n \"out_file.write('\\\\n // IMU - angles velocity\\\\n') for i in range(0, 3): out_file.write(' omega_{}\",\n \"'') for i in range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] +\",\n \"# Rji : jacobian of 'Ri' # return true if it is a\",\n \"si : sine of the relative angle before body i\\\\n') out_file.write(' * ci\",\n \"angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot orientation\",\n \"= nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None]\",\n \"print the declaration of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file,\",\n \"'') om_tilde[0] = get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i]\",\n \"products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split) # no product if cur_len ==\",\n \"1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg & xgp out_file.write('\\\\n // com absolute\",\n \"= sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20]\",\n \"matrix of Ri # xji : jacobian of 'xi' # xgji : jacobian\",\n \"of Ri\\\\n') out_file.write(' * xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji :\",\n \"matrix between body i and its predecessor # si : sine of the\",\n \"else: return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem): if direct:\",\n \"omega_in[1], body_part, omega_in[0], omega_in[2])) # angles (position and derivative) of the waist and\",\n \"point between the two pitch hip rotations\\\\n') out_file.write(' * inertial frame: located at\",\n \"velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,\",\n \"= 0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first:\",\n \"R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i], 'R{}_'.format(i+1),\",\n \"numpy as np import sympy as sp import re import os ###################### #\",\n \"in the relative frame of the current body i # Omi : rotational\",\n \"Rj, xj, xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only\",\n \"i and its predecessor\\\\n') out_file.write(' * si : sine of the relative angle\",\n \"jacobian of 'xi' # xgji : jacobian of 'xgi' # Rji : jacobian\",\n \"= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xj_print[i][j][k]))\",\n \"sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'),\",\n \"range(1, len(joint_id_names)): count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >=\",\n \"om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist\",\n \"j in range(1, nb_bodies): flag_print = 0 for k in range(0, 9): if\",\n \"sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'),\",\n \"# right arm 15, 20, 21, 22 # left arm ]) nb_bodies =\",\n \"3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index),\",\n \"Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb =\",\n \"* [None] # rotation matrices out_file.write('\\\\n // rotation matrices\\\\n') R = nb_bodies*[None] Rt\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not\",\n \"3, # trunk 2, 1, 3, 2, # right arm 2, 1, 3,\",\n \"np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else: return\",\n \"nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0, nb_contacts): for j in range(0,\",\n \"2, # right leg 2, 1, 3, 2, 1, 2, # left leg\",\n \"Ri : absolute rotational matrix # Rti : transpose matrix of Ri #\",\n \"xi\\\\n') out_file.write(' * xgpi : derivative of xgi\\\\n') out_file.write(' * omi : absolute\",\n \"positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right\",\n \"return new_matrix, save_matrix # write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies,\",\n \"if flag_print: out_file.write('\\\\n') return new_matrix # save the symbolic vector for print def\",\n \"M = np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10',\",\n \"xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb\",\n \"0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM\",\n \"body i \\\\n') out_file.write(' * (previous body is not always body i-1), expressed\",\n \"body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} * (s_z_{}*{} +\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"initialization -- //\\\\n') out_file.write('\\\\n // IMU - rotation matrices\\\\n') for i in range(0,\",\n \"}\\\\n\\\\n') out_file.write(' // left wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}]\",\n \"# xgpi : derivative of xgi # omi : absolute rotational vector of\",\n \"Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0,\",\n \"the relative angle before body i\\\\n') out_file.write(' * ci : cosine of the\",\n \"range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and\",\n \"out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n')\",\n \"out_file.write('\\\\n') out_file.write(' // left foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon = 1 -> pitch\",\n \"def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f: # loop\",\n \"3): out_file.write(' in_out.r_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies):\",\n \"jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string) elem_str\",\n \"hip rotations\\\\n') out_file.write(' * inertial frame: located at the origin (waist), but aligned\",\n \"file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file\",\n \"Dg, M): out_file.write('\\\\n\\\\n // -- symbolic computation -- //\\\\n') # Rj, xj, xgj\",\n \"absolute positions and velocities\\\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None] for i in\",\n \"for i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"# apply negative if neg_flag: result = -result return result # compute the\",\n \"sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] =\",\n \"nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] =\",\n \"= sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0,\",\n \"om_l_elb = om[l_elb_id] R_r_elb = R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1)\",\n \"1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id],\",\n \"right leg 'M_13', 'M_14', 'M_15', 'M_16', 'M_17', 'M_18', # left leg 'M_19', 'M_20',\",\n \"Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm\",\n \"== 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1,\",\n \"None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] = ' {} = {};\\\\n'.format(elem_name,\",\n \"k in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first: flag_first =\",\n \"the number of 'elem' in the file def count_elem(in_file, elem): count = 0;\",\n \"= sp.zeros(3, 3) save_matrix = 9 * [None] for i in range(0,3): for\",\n \"der_q = int(der_var.replace('q','')) # detect positive/negative elem_split = elem_str.split('-') cur_len = len(elem_split) if\",\n \"the current body i # (previous body is not always body i-1), expressed\",\n \"y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id] x_l_foot =\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0,\",\n \"the relative angle before body i # # xi : absolute position vector\",\n \"atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file,\",\n \"out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write('\",\n \"absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot\",\n \"= int(value.split('_')[1]) return True, a, b except: return False, -1, -1 # count\",\n \"range(0, nb_bodies): if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j],\",\n \"xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation\",\n \"range(0, nb_contacts): x_r_cont[i] = x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot +\",\n \"{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 -> pitch angle in [pi/2\",\n \"else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name,\",\n \"model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count = 0 for i\",\n \"i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '')\",\n \"{} * counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit() return\",\n \"= {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute orientation jacobian\\\\n') out_file.write(' if\",\n \"outputs out_file.write(' // right wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}]\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not\",\n \"1.0 / c_y_torso;\\\\n\\\\n') out_file.write(' // waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist',\",\n \"x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb *\",\n \"= xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id] R_r_elb =\",\n \"of the file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with\",\n \"nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name)\",\n \"flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com absolute velocity\\\\n') for i\",\n \"in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max: out_file.write(';\\\\n') else: out_file.write(', ')\",\n \"int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return True, a, b, c except:\",\n \"angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write('\",\n \"from the previous body to the current body i # (previous body is\",\n \"derivative) of the waist and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id):\",\n \"joint_id_names, R, x, xp, om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045,\",\n \"joint sines\\\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n\",\n \"(waist), but aligned with the ground (info from IMU) # # Di :\",\n \"get_vector_axis(axis, direct, elem): if direct: if axis == 1: return np.array([[elem], [0.0], [0.0]])\",\n \"epsilon = -1 -> pitch angle in [pi/2 ; 3*pi/2] out_file.write(' in_out.{}[0] =\",\n \"of the previous body # DGi : position vector from the anchor point\",\n \"// -- variables initialization -- //\\\\n') out_file.write('\\\\n // IMU - rotation matrices\\\\n') for\",\n \"[rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left foot orientation matrix as\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"-- variables initialization -- //')) != 1: out_file.write(' // -- variables declaration --\",\n \"(flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_contacts): for j in\",\n \"// 04 10 //\\\\n') out_file.write(' * // 05 11 //\\\\n') out_file.write(' * //\",\n \"= sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')])\",\n \"elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i] = sp.Symbol(elem_name)\",\n \"1) out_file.write('\\\\n') out_file.write(' // torso orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id],\",\n \"computation def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n // --\",\n \"x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1)) # symbolic computation def symbolic_computation(out_file,\",\n \"# count the number of 'elem' in the file def count_elem(in_file, elem): count\",\n \"absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write('\",\n \"if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return\",\n \"loop on all the vector elements for j in range(0, 3): new_vector[j] =\",\n \"elif axis == 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0,\",\n \"0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02,\",\n \"////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' * origin: in the waist, middle point between the\",\n \"== 0 or matrix[i,j] == 1: new_matrix[i,j] = matrix[i,j] else: flag_print = 1\",\n \"sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5]\",\n \"= sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0\",\n \"xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"omega_in[0], omega_in[2])) # angles (position and derivative) of the waist and the torso\",\n \"# left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' ,\",\n \"r_elb_id, l_elb_id, r_wrist_x, r_wrist_y, r_wrist_z): # symbolic variables declarations x_r_elb = x[r_elb_id] x_l_elb\",\n \"!= 0: if flag_first: out_file.write(' + {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k],\",\n \"0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if\",\n \"[0.0, cosine, -sine], [0.0, sine, cosine]]) elif axis == 2: return np.array([[cosine, 0.0,\",\n \"# ###################### # # origin: in the waist, middle point between the two\",\n \"// right foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Rfoot[{}] = {};\\\\n'.format(i,\",\n \"direct: if axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2:\",\n \"%b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b = int(value.split('_')[1]) return\",\n \"xg & xgp out_file.write('\\\\n // com absolute positions and velocities\\\\n') xg = nb_bodies*[None]\",\n \"print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as\",\n \"// left foot contact points absolute position\\\\n') for i in range(0, nb_contacts): for\",\n \"from the previous body to the current body i \\\\n') out_file.write(' * (previous\",\n \"Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0,\",\n \"2, 3, # trunk 2, 1, 3, 2, # right arm 2, 1,\",\n \"(flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range(0, nb_bodies): for j in\",\n \"jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms term_list\",\n \": rotational vector from the previous body to the current body i \\\\n')\",\n \"for cur_term in term_list: # detect products cur_term_split = cur_term.split('*') cur_len = len(cur_term_split)\",\n \"axis == 1: return np.array([[1.0, 0.0, 0.0], [0.0, cosine, -sine], [0.0, sine, cosine]])\",\n \"sine of the relative angle before body i # ci : cosine of\",\n \"2 in negative detection !'.format(cur_len)) exit() # compute derivative result = 0 #\",\n \"xgj, der_var)*sp.Symbol(cur_term_split[0].strip()) # other else: print('Error: {} * counted , only implemented for\",\n \"R_matrix, epsilon): if epsilon > 0: # epsilon = 1 -> pitch angle\",\n \"+ om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write(' // right wrist\",\n \"def der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string) elem_str =\",\n \"expressed in the relative frame of the current body i\\\\n') out_file.write(' * Omi\",\n \"in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot orientation matrix\",\n \"range(0,3): if vector[i] == 0 or vector[i] == 1: new_vector[i] = vector[i] else:\",\n \"loop on all the lines for line in f: cut_line_1 = line.split(elem) cut_line_2\",\n \"[0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]]) elif axis ==\",\n \"* xgi : absolute position vector of the COM G_i of body i\\\\n')\",\n \"= 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k]))\",\n \"= sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'),\",\n \"out_file.write(' // right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"}\\\\n\\\\n') out_file.write(' // left foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n') for i in\",\n \"Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"0 for i in range(1, len(joint_id_names)): count += 1 if i == 1:\",\n \"= None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) save_matrix[3*i+j] = ' {}\",\n \"* R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i] = write_symb_matrix(out_file, R[i],\",\n \"matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]], [-v[1], v[0], 0.0]])\",\n \"False, -1, -1, -1 # return true if it has a shape 'x%a_%b'\",\n \"Rji : jacobian of 'Ri' # return true if it is a float\",\n \"pitch hip rotations\\\\n') out_file.write(' * inertial frame: located at the origin (waist), but\",\n \"19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate the\",\n \"= nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0] *\",\n \"and the torso def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation\",\n \"Dg, M) # end of the file write_file_end(file_temp) file_temp.close() # output file out_file\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"3 * [None] for i in range(0,3): if vector[i] == 0 or vector[i]\",\n \"3): if xj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print =\",\n \"/ c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot',\",\n \"of body i\\\\n') out_file.write(' * Ri : absolute rotational matrix\\\\n') out_file.write(' * Rti\",\n \"[0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) #\",\n \"global com absolute velocity\\\\n') for i in range(0, 3): out_file.write(' in_out.rp_COM[{}] = '.format(i))\",\n \"// -- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary\",\n \"symbolic_jacob_der(Rj, xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i == max:\",\n \"joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n') for i in range(1, nb_bodies): out_file.write(' Om{}\",\n \"sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] =\",\n \"x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb = om[r_elb_id] om_l_elb = om[l_elb_id]\",\n \"joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right\",\n \"out_file.write(' * xpi : derivative of xi\\\\n') out_file.write(' * xgpi : derivative of\",\n \", 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' #\",\n \"out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write('\",\n \"np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]]) elif axis ==\",\n \"= get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot x_l = x_l_foot +\",\n \"if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n') out_file.write(' if\",\n \"// waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' //\",\n \"for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j], j+1, i+1)) else:\",\n \"jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center of\",\n \"new_matrix = sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for j\",\n \"1: new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name)\",\n \"0 for k in range(0, 3): if xgj_print[i][j][k] != None: if not flag_first:\",\n \"in the relative # frame of the previous body # Rdi : rotational\",\n \"if (flag_jacob)\\\\n {\\\\n') for i in range(1, nb_bodies): for j in range(0, 3):\",\n \"return new_matrix # save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name):\",\n \"f: # loop on all the lines for line in f: cut_line_1 =\",\n \"an orientation matrix, compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file,\",\n \"# joint names joint_id_names = np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id',\",\n \"xgi\\\\n') out_file.write(' * omi : absolute rotational vector of body i\\\\n') out_file.write(' *\",\n \"out_file.write(' // left wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"write_file_beginning(file_temp, joint_id_names) # variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies,\",\n \"# print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write(' double ') for\",\n \"max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if i\",\n \"origin: in the waist, middle point between the two pitch hip rotations\\\\n') out_file.write('\",\n \"'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id],\",\n \"i in range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont\",\n \"wrists absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i]))\",\n \"np.array([]) # compute the derivative of an element (for jacobian) def der_elem(elem_str, Rj,\",\n \"= write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian\",\n \"(Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file,\",\n \"= isVec(pos_str) # rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif\",\n \"implemented for 0 or 1 !'.format(cur_len-1)) exit() return result # write the beginning\",\n \"'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' ,\",\n \"symbolic jacobian of an anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector,\",\n \"the relative angle before body i # ci : cosine of the relative\",\n \"in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"between body i and its predecessor\\\\n') out_file.write(' * si : sine of the\",\n \"in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] =\",\n \"sp.Symbol('L_3_23')]) Dg[18] = sp.Matrix([sp.Symbol('L_1_24'), sp.Symbol('L_2_24'), sp.Symbol('L_3_24')]) Dg[19] = sp.Matrix([sp.Symbol('L_1_25'), sp.Symbol('L_2_25'), sp.Symbol('L_3_25')]) # left\",\n \"sp.zeros(3, 1) # loop on all the vector elements for j in range(0,\",\n \"angle before body i # ci : cosine of the relative angle before\",\n \"in range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies,\",\n \"sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([\",\n \"write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as\",\n \"count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) +\",\n \"(center of mass) position and velocity\\\\n') out_file.write(' * feet position, velocity and orientation\\\\n')\",\n \"range(0,3): out_file.write(' in_out.rp_Rwrist[{}] = {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n') out_file.write('\",\n \"* of the anchor point of body i\\\\n') out_file.write(' * xgi : absolute\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left\",\n \"out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // wrists absolute orientation\\\\n')\",\n \"line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if count\",\n \"left foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write('\",\n \"in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2\",\n \"sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n')\",\n \"else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min, max): out_file.write('\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"omega_in[2])) # angles (position and derivative) of the waist and the torso def\",\n \"i in range(0,3): for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i ==\",\n \"= parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id] +\",\n \"'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11', 'M_12', #\",\n \"the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author\",\n \"= x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T\",\n \"the feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om,\",\n \"* DGi : position vector from the anchor point of body i to\",\n \"aligned with the ground (info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di :\",\n \"body to the current body i \\\\n') out_file.write(' * (previous body is not\",\n \"3) xj[i][j] = sp.zeros(3, 1) xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 *\",\n \"= '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write('\",\n \"= sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min\",\n \"absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rwrist_or[{}] = {};\\\\n'.format(i, R_r_elb[i])) out_file.write('\\\\n')\",\n \"sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'),\",\n \", 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id' , 'LeftFootPitch_id' , # left\",\n \"}\\\\n\\\\n') out_file.write(' // feet absolute orientation\\\\n') for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}]\",\n \"in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1, joint_id_names[i])) # write symbolic vector and\",\n \"* (previous body is not always body i-1), expressed in the relative\\\\n') out_file.write('\",\n \"waist 0, 1, 2, 3, 4, 5, # right leg 0, 7, 8,\",\n \"Di : position vector from the anchor point of the previous body to\",\n \"out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i\",\n \"out_file.write(' // right foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1)\",\n \"# masses M = np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9'\",\n \"= 9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 *\",\n \"orientation matrix, compute the roll, pitch, yaw angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name,\",\n \"anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0, 0.0])\",\n \"out_file.write('/*! \\\\n') out_file.write(' * \\\\\\\\author <NAME>\\\\n') out_file.write(' * \\\\\\\\file forward_kinematics.cc\\\\n') out_file.write(' * \\\\\\\\brief\",\n \"[None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None] # rotation\",\n \", 'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' ,\",\n \"xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point\",\n \"'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj,\",\n \"print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 * [None] for\",\n \"15) # generate the symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names,\",\n \"body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part,\",\n \"cur_len = len(elem_split) if cur_len == 1: # positive neg_flag = 0 pos_str\",\n \"in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i] = sp.zeros(3, 1) Dpt_r_foot_cont[0][0] =\",\n \"in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '') om_tilde[0] =\",\n \"= 1 out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n // -- Collecting results\",\n \"i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in\",\n \"c_z_torso = cos(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' s_y_waist = sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist\",\n \"absolute rotational vector of body i\\\\n') out_file.write(' * Ri : absolute rotational matrix\\\\n')\",\n \"for i in range(0,3): out_file.write(' in_out.r_Rfoot[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right\",\n \"left foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i]))\",\n \"M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning\",\n \"\\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs and outputs\",\n \"cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string = cur_split[0].upper() for i in range(1,\",\n \"sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23]\",\n \"if len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' // -- variables\",\n \"axis == 1: return np.array([[elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0],\",\n \", sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] =\",\n \"0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1))\",\n \"3, 2 # left arm ]) # parent index parent_body_index = np.array([ -1,\",\n \"= {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i,\",\n \"out_write, nb_max_char): count = 0 with open(in_file,'r') as f: # loop on all\",\n \"out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix\",\n \"of body i\\\\n') out_file.write(' * xgi : absolute position vector of the COM\",\n \"1: out_file.write(' // -- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') #\",\n \"xgj, symb_var, der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if\",\n \"else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // --\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] =\",\n \"xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n // jacobian com\",\n \"0.0, 1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis, direct, elem):\",\n \"predecessor\\\\n') out_file.write(' * si : sine of the relative angle before body i\\\\n')\",\n \"f: # declaration if len(line.split('// -- variables initialization -- //')) != 1: out_file.write('\",\n \"c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot')\",\n \"0 for i in range(0,3): if vector[i] == 0 or vector[i] == 1:\",\n \"if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n')\",\n \"om_tilde[0] * (Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0]\",\n \"frame) # of the anchor point of body i # xgi : absolute\",\n \"axis == 3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]])\",\n \">= nb_max_char: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') # get\",\n \"= sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_vector # write symbolic matrix and replace\",\n \"nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None]\",\n \"+= der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len == 2:\",\n \"2, 1, 2, # right leg 2, 1, 3, 2, 1, 2, #\",\n \"result += der_elem(cur_term_split[0], Rj, xj, xgj, der_var) # one product elif cur_len ==\",\n \"out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write('\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lfoot_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n')\",\n \"frame of the current body i\\\\n') out_file.write(' * Omi : rotational vector from\",\n \"13 //\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' *\\\\n') out_file.write(' *\",\n \"0.0, cosine]]) elif axis == 3: return np.array([[cosine, -sine, 0.0], [sine, cosine, 0.0],\",\n \"R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1],\",\n \"np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3:\",\n \"the wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om,\",\n \"== 2: # negative neg_flag = 1 pos_str = elem_split[1] else: print('Error: {}\",\n \"two pitch hip rotations\\\\n') out_file.write(' * inertial frame: located at the origin (waist),\",\n \"out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot',\",\n \"sp.zeros(3, 3) save_matrix = 9 * [None] for i in range(0,3): for j\",\n \"from an orientation matrix, compute the roll, pitch, yaw angles (and derivative) def\",\n \"(center of mass) G_i,\\\\n') out_file.write(' * expressed in the relative frame of the\",\n \"+ {}*{}'.format(M[k], str(xgj[k][i][j]))) else: flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1:\",\n \"j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac\",\n \"flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n // joint\",\n \": transpose matrix of Ri # xji : jacobian of 'xi' # xgji\",\n \"= 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n // joint absolute velocities\\\\n')\",\n \"parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i] = xp[parent_id]\",\n \"out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n') for i in\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute orientation jacobian\\\\n') out_file.write(' if\",\n \"[None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] =\",\n \"-- //\\\\n') out_file.write('\\\\n // IMU - rotation matrices\\\\n') for i in range(0, 3):\",\n \"x[l_foot_id] xp_r_foot = xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id]\",\n \"in the inertial frame)\\\\n') out_file.write(' * of the anchor point of body i\\\\n')\",\n \"get_tilde(om[i]) # x & xp out_file.write('\\\\n // anchor point absolute positions and velocities\\\\n')\",\n \"i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi : absolute position vector (from origin, expressed\",\n \"len(cur_split) >= 2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string =\",\n \"'') # jacobian xg out_file.write('\\\\n // jacobian com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"body i # (previous body is not always body i-1), expressed in the\",\n \"# compute the center of mass position and velocity def com_compute(out_file, nb_bodies, joint_id_names,\",\n \"R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) #\",\n \"counted , only implemented for 0 or 1 !'.format(cur_len-1)) exit() return result #\",\n \"out_file.write(' * // 03 09 //\\\\n') out_file.write(' * // 04 10 //\\\\n') out_file.write('\",\n \"= sin(in_out.theta_waist[1]);\\\\n') out_file.write(' s_y_torso = sin(in_out.theta_torso[1]);\\\\n') out_file.write(' s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso =\",\n \"x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot\",\n \"in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n')\",\n \"xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n // jacobian anchor point positions\\\\n') out_file.write('\",\n \"between body i and its predecessor # si : sine of the relative\",\n \"0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif axis == 3: return\",\n \"print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line) out_file.close() # remove temporary\",\n \"len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0,\",\n \"// jacobian com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"out_file.write('\\\\n') out_file.write('{}'.format(xgj_print[i][j][k])) out_file.write(' }\\\\n') # results out_file.write('\\\\n // -- Collecting results -- //\\\\n\\\\n')\",\n \"variables initialization write_intialization(file_temp, nb_bodies, joint_id_names) # symbolic computation symbolic_computation(file_temp, nb_bodies, joint_id_names, rot_axis, parent_body_index,\",\n \"sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'),\",\n \"jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i in range(1, nb_bodies): for j in\",\n \"out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: # epsilon = -1 ->\",\n \"arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] =\",\n \"get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1: return np.array([[1.0, 0.0,\",\n \"= x_r_foot + R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i]\",\n \"range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot absolute velocity\\\\n')\",\n \"waist orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' //\",\n \"for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right\",\n \"count += len(cut_line[0].strip()) + 2 if count >= nb_max_char: out_write.write(';\\\\n') count = 0\",\n \"* omi : absolute rotational vector of body i\\\\n') out_file.write(' * Ri :\",\n \"= {};\\\\n'.format(i, xp_l[i])) out_file.write('\\\\n') out_file.write(' // left foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: # epsilon = 1 ->\",\n \"= atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] =\",\n \"previous body # Rdi : rotational matrix between body i and its predecessor\",\n \"xj, xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"= {};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"xgj, xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file,\",\n \"= 0 for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_elb[j], 'q{}'.format(i+1))\",\n \"of the previous body # Rdi : rotational matrix between body i and\",\n \"range(0, nb_contacts): for j in range(0, 3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts\",\n \"= {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot contact points absolute\",\n \": jacobian of \\\\'Ri\\\\'\\\\n') out_file.write(' */\\\\n') out_file.write('void ForwardKinematics::main_kinematics(KinematicsInOut &in_out)\\\\n{\\\\n') # compute the center\",\n \"nb_bodies): flag_print = 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj,\",\n \"in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k], 'q{}'.format(j+1)) if cur_jac !=\",\n \"0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8] = sp.Matrix([0.0,\",\n \"left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22]\",\n \"j == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com\",\n \"derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position, velocity and\",\n \"+ R_l_elb.T * Dpt_l_wrist xp_r = xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist)\",\n \"Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0,\",\n \"// rotation matrices\\\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd = nb_bodies*[None] Rd[0]\",\n \"= cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i]) else: new_string\",\n \"return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0 /\",\n \"== 2: out_file.write(';\\\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix,\",\n \"1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print,\",\n \"0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis,\",\n \"orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') # compute the feet position,\",\n \"out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables\",\n \"= nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in\",\n \"line.split(elem) if len(cut_line) == 2: count += 1 return count # print the\",\n \"all the joints for i in range (1, nb_bodies): new_matrix = sp.zeros(3, 3)\",\n \"out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' s_z_Rfoot = sin(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' s_z_Lfoot = sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write('\",\n \"file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): # temporary file\",\n \"3) save_matrix = 9 * [None] for i in range(0,3): for j in\",\n \"0: # epsilon = 1 -> pitch angle in [-pi/2 ; pi/2] out_file.write('\",\n \"absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write('\",\n \"3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot contact\",\n \"* Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right foot absolute position\\\\n') for i\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right\",\n \"by its name def write_symb_matrix(out_file, matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) flag_print\",\n \"# ci : cosine of the relative angle before body i # #\",\n \"= vector[i] save_vector[i] = None else: elem_name = '{}{}{}'.format(start_name, i+1, end_name) save_vector[i] =\",\n \"R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out,\",\n \"# # 06 12 # # 07 13 # # # ###################### #\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left wrist absolute position\\\\n') for i in\",\n \"= 3 * [None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\\\n\",\n \"xgj, 19, 23, -0.02, -0.005, -0.225) torso_waist_angles(out_file, R, om, 0, 15) # generate\",\n \"sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i\",\n \"* (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0], body_part, omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{}\",\n \"nb_max_line: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') # print all\",\n \"j in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' //\",\n \"# writing outputs out_file.write(' // right wrist absolute position\\\\n') for i in range(0,3):\",\n \"Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb =\",\n \"the matrix elements for j in range(0, 9): new_matrix[j] = symbolic_jacob_der(Rj, xj, xgj,\",\n \"f: cut_line = line.split(elem) if len(cut_line) == 2: count += 1 return count\",\n \"out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute velocity\\\\n') for\",\n \"= print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a com point def\",\n \"s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n') for i in range(1,\",\n \"return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]]) else:\",\n \"j in range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if\",\n \"y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in\",\n \"*\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' *\",\n \"vector[i] == 1: new_vector[i] = vector[i] else: flag_print = 1 elem_name = '{}{}{}'.format(start_name,\",\n \"in range(0, nb_bodies): xg[i] = x[i] + Rt[i] * Dg[i] xgp[i] = xp[i]\",\n \"i+1)) else: flag_first = 1 out_file.write('({}*xgp{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if\",\n \"* [None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\\\n // rotation\",\n \"om, Rj, xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R,\",\n \"xgji : jacobian of 'xgi' # Rji : jacobian of 'Ri' # return\",\n \"file write_file_end(file_temp) file_temp.close() # output file out_file = open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r')\",\n \"+= Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative if\",\n \"' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save\",\n \"else: flag_print = 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} =\",\n \"Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')])\",\n \"and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if count == 0: out_write.write('\",\n \"out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0\",\n \"derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[l_foot_id], 'omega_Lfoot', 'Lfoot') out_file.write('\\\\n') # compute the wrists position, velocity\",\n \"for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j,\",\n \"0.0, sp.Symbol('DPT_3_8')]) Dpt[4] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_10')]) Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6]\",\n \"Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1),\",\n \"out_file.write(' * (previous body is not always body i-1), expressed in the relative\\\\n')\",\n \"if len(cut_line_1) == 2 and len(cut_line_2) == 2: if len(cut_line_2[0].split('[')) == 1: if\",\n \"sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] #\",\n \"(Rt[0] * Dpt[0]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[0], 1) x[0] = write_symb_vector(out_file,\",\n \"out_file.write(' * xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian of\",\n \"out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from\",\n \"xj, xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first =\",\n \"i and its predecessor # si : sine of the relative angle before\",\n \"sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')])\",\n \"count the number of 'elem' in the file def count_elem(in_file, elem): count =\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_24')]) Dpt[12] =\",\n \"{}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1:\",\n \"flag_first = 1 out_file.write('({}*{}'.format(M[k], str(xgj[k][i][j]))) if k == nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else:\",\n \"loop on all the joints for i in range (1, nb_bodies): new_vector =\",\n \"//\\\\n\\\\n') com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj) feet_compute(out_file, joint_id_names, R, x, xp,\",\n \"flag_print = 0 for k in range(0, 3): if xj_print[i][j][k] != None: if\",\n \"sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'),\",\n \"3) # loop on all the matrix elements for j in range(0, 9):\",\n \"velocity\\\\n') for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n //\",\n \"[None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot +\",\n \"sp.zeros(3, 1) Dpt_r_foot_cont[0][0] = x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] =\",\n \"= np.array([ 'M_6', # waist 'M_7' , 'M_8' , 'M_9' , 'M_10', 'M_11',\",\n \"= sp.zeros(3, 3) R[0] = sp.zeros(3, 3) for i in range(0, 3): for\",\n \"new_string # write the end of the file def write_file_end(out_file): out_file.write('}\\\\n') # print\",\n \"R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] =\",\n \"sp.Symbol(elem_name) if flag_print: out_file.write('\\\\n') return new_matrix # save the symbolic vector for print\",\n \"04 10 //\\\\n') out_file.write(' * // 05 11 //\\\\n') out_file.write(' * // 06\",\n \"out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write(' // right foot contact points\",\n \"[0.0, 0.0, 1.0]]) else: return np.array([]) # get vector axis def get_vector_axis(axis, direct,\",\n \"f: cut_line_1 = line.split(elem) cut_line_2 = line.split(' = ') if len(cut_line_1) == 2\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute position\\\\n') for i\",\n \"xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs out_file.write('\",\n \"new_matrix[i,j] = matrix[i,j] save_matrix[3*i+j] = None else: elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name)\",\n \"# other else: print('Error: {} * counted , only implemented for 0 or\",\n \"in_out.{}[2] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles'\",\n \"r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4\",\n \"in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] = Rd[i] *\",\n \"i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write(' // global com absolute\",\n \"= x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] =\",\n \"9 * [None] xj_print[i][j] = 3 * [None] xgj_print[i][j] = 3 * [None]\",\n \"in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]), k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot contact\",\n \"new_matrix[i,j] = sp.Symbol(elem_name) return new_matrix, save_matrix # write symbolic jacobian of a rotation\",\n \"sp.Symbol('L_3_25')]) # left arm Dg[20] = sp.Matrix([sp.Symbol('L_1_26'), sp.Symbol('L_2_26'), sp.Symbol('L_3_26')]) Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'),\",\n \"R_matrix[0])) # compute the time derivatives of 'yaw_pitch_roll_angles' def theta_dot_compute(out_file, omega_in, omega_out, body_part):\",\n \"[None] x_l_cont = nb_contacts * [None] # computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot =\",\n \"sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_50')])\",\n \"the previous body\\\\n') out_file.write(' * Rdi : rotational matrix between body i and\",\n \"out_file.write(' }\\\\n\\\\n') # from an orientation matrix, compute the roll, pitch, yaw angles\",\n \"sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if vector[i] == 0\",\n \"out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j == 2: out_file.write(';\\\\n') else:\",\n \"count >= nb_max_line: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') #\",\n \"an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of\",\n \"print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for\",\n \"nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i]\",\n \"rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1:\",\n \"terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+') if term_list[0] == '': term_list.pop(0) result = 0\",\n \"left arm ]) nb_bodies = len(parent_body_index) ## anchor point positions Dpt = nb_bodies*[None]\",\n \"count = 0; with open(in_file, 'r') as f: # loop on all the\",\n \"feet position, velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj,\",\n \"# xi : absolute position vector (from origin, expressed in the inertial frame)\",\n \"if it has a shape 'x%a_%b' (indexes %a, %b also returned) def isVec(value):\",\n \"cut_line_2 = line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) == 2:\",\n \"previous body\\\\n') out_file.write(' * DGi : position vector from the anchor point of\",\n \"== 2: return np.array([[cosine, 0.0, sine], [0.0, 1.0, 0.0], [-sine, 0.0, cosine]]) elif\",\n \"COMAN model\\\\n') out_file.write(' */\\\\n\\\\n') out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count = 0 for\",\n \"if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write(' // global com\",\n \"= 0 for j in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_r[j],\",\n \"= {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector # save the symbolic\",\n \"for i in range(0,3): om[0][i] = sp.Symbol('omega_{}'.format(i+1)) om[0] = write_symb_vector(out_file, om[0], 'om1_', '')\",\n \"nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None] for i in range(0, nb_contacts):\",\n \"Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] =\",\n \"= inv_c_y_{} * s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part,\",\n \"c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot\",\n \"{};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute orientation jacobian\\\\n') out_file.write('\",\n \"1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j]\",\n \"om_tilde[parent_id] * (Rt[parent_id] * Dpt[i]) write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x[i], i+1) x[i]\",\n \"Dpt_r_foot_cont[1][1] = y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0,\",\n \"'xgi' # Rji : jacobian of 'Ri' # return true if it is\",\n \"# jacobian x out_file.write('\\\\n // jacobian anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n')\",\n \"computation om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T *\",\n \"x_max, y_min, y_max): # symbolic variables declarations nb_contacts = 4 x_r_foot = x[r_foot_id]\",\n \"xgj[i][j] = sp.zeros(3, 1) Rj_print[i][j] = 9 * [None] xj_print[i][j] = 3 *\",\n \"elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac))\",\n \"1, 2, 3, 4, 5, # right leg 0, 7, 8, 9, 10,\",\n \"out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot\",\n \"i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' // left foot\",\n \"for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_r_cont[i][j])) out_file.write('\\\\n') out_file.write('\",\n \"the anchor point of the previous body to the current body i #\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]),\",\n \"waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') ,\",\n \"sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3]\",\n \"flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if\",\n \"out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}]\",\n \"'{}{}'.format(new_string, cur_split[i]) else: new_string = cur_string cur_split = filter(None, re.split(\\\"([A-Z][^A-Z]*)\\\", new_string)) new_string =\",\n \"for k in range(0, nb_bodies): if xgj[k][i][j] != 0: if flag_first: out_file.write(' +\",\n \"writing outputs out_file.write(' // right wrist absolute position\\\\n') for i in range(0,3): out_file.write('\",\n \"{\\\\n') flag_first = 0 for i in range(0, nb_contacts): for j in range\",\n \"the derivative of an element (for jacobian) def der_elem(elem_str, Rj, xj, xgj, der_var):\",\n \"out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write('\",\n \"wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n')\",\n \"nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for j in range(0, nb_bodies-1): Rj[i][j] = sp.zeros(3, 3)\",\n \"for j in range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j\",\n \"in range(0, 3): out_file.write(' IMU{}{} = in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU -\",\n \"= get_tilde(om[0]) for i in range(1, nb_bodies): parent_id = parent_body_index[i] Om[i] = get_vector_axis(rot_axis[i],\",\n \"xp = nb_bodies*[None] x[0] = Rt[0] * Dpt[0] xp[0] = om_tilde[0] * (Rt[0]\",\n \"for each joint before body i (1:x, 2:y, 3:z) rot_axis = np.array([0, #\",\n \"point of the previous body to the current body i # (previous body\",\n \"# trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15]\",\n \"'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id' # left arm ]) out_file_name =\",\n \"= 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' //\",\n \"remove temporary file os.remove(in_temp) # main script # rotation axis for each joint\",\n \"//\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line) out_file.close() # remove\",\n \"c = int(value.split('_')[1][1]) return True, a, b, c except: return False, -1, -1,\",\n \"the file def write_file_end(out_file): out_file.write('}\\\\n') # print matrix components declaration def write_matrix_declaration(out_file, prefix):\",\n \"sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for j in range(0,3):\",\n \"sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] =\",\n \"b, c] = isRot(pos_str) [vec_flag, d, e] = isVec(pos_str) # rotation matrix if\",\n \"of body i\\\\n') out_file.write(' * xpi : derivative of xi\\\\n') out_file.write(' * xgpi\",\n \"in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // right foot contact points\",\n \"file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp, 'w') # beginning of the file\",\n \"result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result += xj[d-1][der_q-1][e-1] # apply negative\",\n \"# Omi : rotational vector from the previous body to the current body\",\n \"com_compute(out_file, nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for i\",\n \"R[l_foot_id] Dpt_r_foot = sp.zeros(3, 1) Dpt_l_foot = sp.zeros(3, 1) Dpt_r_foot[2] = sp.Symbol('DPT_3_16') Dpt_l_foot[2]\",\n \"= symbolic_jacob_der(Rj, xj, xgj, R_matrix[j], 'q{}'.format(i+1)) [Rj[index-1][i], Rj_print[index-1][i]] = print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) #\",\n \"variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization -- //\\\\n')\",\n \"sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')])\",\n \"{}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >= nb_max_char:\",\n \"Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'),\",\n \"= write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian\",\n \"xp_r_elb + om_tilde_r_elb * (R_r_elb.T * Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb *\",\n \"'x%a_%b' (indexes %a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b\",\n \"Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0,\",\n \"min, max): out_file.write(' double ') for i in range(min, max+1): out_file.write('{}{}'.format(prefix, i)) if\",\n \"frame: located at the origin (waist), but aligned with the ground (info from\",\n \"x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T *\",\n \"lines for line in f: cut_line = line.split(elem) if len(cut_line) == 2: count\",\n \"== 2: return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0],\",\n \"it is a float def isInt(value): try: int(value) return True except: return False\",\n \"# DGi : position vector from the anchor point of body i to\",\n \"position, velocity and orientation\\\\n') out_file.write(' * waist and torso orientaion angles and derivatives\\\\n')\",\n \"the waist, middle point between the two pitch hip rotations\\\\n') out_file.write(' * inertial\",\n \"before body i\\\\n') out_file.write(' * ci : cosine of the relative angle before\",\n \"= cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write(' c_z_Rfoot = cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot =\",\n \"# loop on all the matrix elements for j in range(0, 9): new_matrix[j]\",\n \"print_save_symb_matrix(new_matrix, 'R{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of an anchor point def write_symb_xj(nb_bodies,\",\n \"symbolic vector and replace symbolic variable by its name def write_symb_vector(out_file, vector, start_name,\",\n \"= sp.zeros(3, 3) flag_print = 0 for i in range(0,3): for j in\",\n \"for i in range(0, 9): out_file.write(' in_out.Rfoot_or[{}] = {};\\\\n'.format(i, R_r_foot[i])) out_file.write('\\\\n') for i\",\n \"# get tilde matrix def get_tilde(v): return np.array([[0.0, -v[2], v[1]], [v[2], 0.0, -v[0]],\",\n \"# left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0])\",\n \"R_r_foot.T * Dpt_r_foot_cont[i] x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs\",\n \"k, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot contact points absolute position\\\\n') for\",\n \"10 //\\\\n') out_file.write(' * // 05 11 //\\\\n') out_file.write(' * // 06 12\",\n \"anchor point of the previous body to the current body i # (previous\",\n \"wrists position, velocity and orientation def wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj,\",\n \"xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '') Rt[0] =\",\n \"import numpy as np import sympy as sp import re import os ######################\",\n \"# 04 10 # # 05 11 # # 06 12 # #\",\n \"return result # write the beginning of the file def write_file_beginning(out_file, joint_id_names): out_file.write('/*!\",\n \"the anchor point of body i to its COM (center of mass) G_i,\",\n \"joint_id_names, Dpt, Dg, M): # temporary file in_temp = './{}_temp.cc'.format(out_file_name) file_temp = open(in_temp,\",\n \"Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0,\",\n \"xgj, x_r[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"of an expression (for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list\",\n \"range(0, 3): out_file.write(' in_out.r_COM_der[{}][{}] = '.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in\",\n \"relative # frame of the previous body # Rdi : rotational matrix between\",\n \"in range(0, 3): out_file.write(' in_out.r_Lfoot_cont[{}][{}] = {};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left\",\n \"omega_out, body_part): out_file.write(' in_out.{}[0] = inv_c_y_{} * (c_z_{}*{} + s_z_{}*{});\\\\n'.format(omega_out, body_part, body_part, omega_in[0],\",\n \"= 0 for k in range(0, 3): cur_jac = symbolic_jacob_der(Rj, xj, xgj, x_l_cont[i][k],\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21]\",\n \"out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs and\",\n \"{}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count >= nb_max_line: out_write.write(';\\\\n') count\",\n \"else: out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from an\",\n \"else: out_file.write(' }\\\\n\\\\n') # from an orientation matrix, compute the roll, pitch, yaw\",\n \"left foot absolute velocity\\\\n') for i in range(0,3): out_file.write(' in_out.rp_Lfoot[{}] = {};\\\\n'.format(i, xp_l[i]))\",\n \"range(0,3): out_file.write('{}{}{}'.format(prefix, i+1, j+1)) if i == 2 and j == 2: out_file.write(';\\\\n')\",\n \"i in range(0, 9): out_file.write(' in_out.Lwrist_or[{}] = {};\\\\n'.format(i, R_l_elb[i])) out_file.write('\\\\n') out_file.write(' // right\",\n \"inputs and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation of:\\\\n') out_file.write(' * COM\",\n \"flag_first = 0 for j in range(0, nb_bodies): if flag_first: out_file.write(' + {}*xgp{}_{}'.format(M[j],\",\n \"matrix and replace symbolic variable by its name def write_symb_matrix(out_file, matrix, start_name, end_name):\",\n \"out_file.write('\\\\n // com absolute positions and velocities\\\\n') xg = nb_bodies*[None] xgp = nb_bodies*[None]\",\n \"sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0, sp.Symbol('DPT_2_6'), 0.0]) Dpt[3] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_8')])\",\n \"'M_26', 'M_27', 'M_28', 'M_29' # left arm ]) # joint names joint_id_names =\",\n \"# # 05 11 # # 06 12 # # 07 13 #\",\n \"count += 1 if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count\",\n \"// waist orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write('\",\n \"nb_contacts * [None] for i in range(0, nb_contacts): Dpt_r_foot_cont[i] = sp.zeros(3, 1) Dpt_l_foot_cont[i]\",\n \"angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_Rfoot', R[r_foot_id], 1) out_file.write('\\\\n') out_file.write(' // left foot orientation matrix\",\n \"file_temp = open(in_temp, 'w') # beginning of the file write_file_beginning(file_temp, joint_id_names) # variables\",\n \"wrist absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"xj, xgj, 6, 12, -0.06, 0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp,\",\n \"xgi : absolute position vector of the COM G_i of body i\\\\n') out_file.write('\",\n \"with the ground (info from IMU)\\\\n') out_file.write(' *\\\\n') out_file.write(' * Di : position\",\n \": transpose matrix of Ri\\\\n') out_file.write(' * xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write('\",\n \"len(elem_split) if cur_len == 1: # positive neg_flag = 0 pos_str = elem_split[0]\",\n \"else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1)) if j == nb_bodies-1: if flag_first:\",\n \"out_file.write(' // left foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"{\\\\n') flag_first = 0 for i in range(0, nb_bodies): for j in range(1,\",\n \"0: out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if\",\n \"= '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the file def\",\n \"count >= 6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*!\",\n \"[-v[1], v[0], 0.0]]) # get rotation matrix def get_rotation_matrix(axis, direct, cosine, sine): if\",\n \"# left arm ]) # parent index parent_body_index = np.array([ -1, # waist\",\n \"{};\\\\n'.format(i, j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot contact points jacobian\\\\n') out_file.write(' if\",\n \"for i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint\",\n \"3: return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return\",\n \"i in range(0, 3): out_file.write(' omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n')\",\n \"Dpt[15] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_32')]) # right arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')])\",\n \"on all the lines for line in f: cut_line_1 = line.split(elem) cut_line_2 =\",\n \"'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic jacobian of a\",\n \"all the lines for line in f: cut_line_1 = line.split(elem) cut_line_2 = line.split('\",\n \"Dpt[8] = sp.Matrix([0.0, sp.Symbol('DPT_2_18'), 0.0]) Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0,\",\n \"sp.Matrix([sp.Symbol('L_1_11'), sp.Symbol('L_2_11'), sp.Symbol('L_3_11')]) Dg[6] = sp.Matrix([sp.Symbol('L_1_12'), 0.0 , sp.Symbol('L_3_12')]) # left leg Dg[7]\",\n \"angles (and derivative) def yaw_pitch_roll_angles(out_file, angle_name, R_matrix, epsilon): if epsilon > 0: #\",\n \"Dg[21] = sp.Matrix([sp.Symbol('L_1_27'), sp.Symbol('L_2_27'), sp.Symbol('L_3_27')]) Dg[22] = sp.Matrix([sp.Symbol('L_1_28'), sp.Symbol('L_2_28'), sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'),\",\n \"arm Dg[16] = sp.Matrix([sp.Symbol('L_1_22'), sp.Symbol('L_2_22'), sp.Symbol('L_3_22')]) Dg[17] = sp.Matrix([sp.Symbol('L_1_23'), sp.Symbol('L_2_23'), sp.Symbol('L_3_23')]) Dg[18] =\",\n \"[0.0], [elem]]) else: return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0],\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Lfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot absolute position\\\\n') for i in\",\n \"sp.Symbol('DPT_3_16') Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts *\",\n \"R_matrix, index): # loop on all the joints for i in range (1,\",\n \"// joint sines\\\\n') for i in range(1, nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i]))\",\n \"cos(in_out.theta_Rfoot[2]);\\\\n') out_file.write(' c_z_Lfoot = cos(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' s_y_Rfoot = sin(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' s_y_Lfoot = sin(in_out.theta_Lfoot[1]);\\\\n')\",\n \"= len(elem_split) if cur_len == 1: # positive neg_flag = 0 pos_str =\",\n \"nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n') out_file.write('\",\n \"new_matrix # save the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector\",\n \"= line.split(' = ') if len(cut_line) == 2: if len(cut_line[0].split('[')) == 1: if\",\n \"range(0, 3): for j in range(0, 3): R[0][i,j] = sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj,\",\n \"angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso orientation angle\",\n \"Rj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1 elif\",\n \"'M_27', 'M_28', 'M_29' # left arm ]) # joint names joint_id_names = np.array(['0',\",\n \"omega_in[1])) out_file.write(' in_out.{}[1] = c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2]\",\n \"sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')])\",\n \"= 1 elem_name = '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*',''))\",\n \"= sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'), sp.Symbol('L_3_19')]) Dg[14] = sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'),\",\n \"Dpt_l_foot xp_r = xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot\",\n \"11 //\\\\n') out_file.write(' * // 06 12 //\\\\n') out_file.write(' * // 07 13\",\n \"R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0]\",\n \"[None] xgj_print[i][j] = 3 * [None] # rotation matrices out_file.write('\\\\n // rotation matrices\\\\n')\",\n \"new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xgj[index-1][i], xgj_print[index-1][i]] = print_save_symb_vector(new_vector, 'xg{}_'.format(index), '_d{}'.format(i+1))\",\n \"'M_19', 'M_20', 'M_21', # trunk 'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26',\",\n \"loop on all the matrix elements for j in range(0, 9): new_matrix[j] =\",\n \"'M_22', 'M_23', 'M_24', 'M_25', # right arm 'M_26', 'M_27', 'M_28', 'M_29' # left\",\n \"1) x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '')\",\n \"# xji : jacobian of 'xi' # xgji : jacobian of 'xgi' #\",\n \"# # xi : absolute position vector (from origin, expressed in the inertial\",\n \", sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') , sp.Symbol('L_2_9') , sp.Symbol('L_3_9')]) Dg[4] = sp.Matrix([sp.Symbol('L_1_10'), sp.Symbol('L_2_10'),\",\n \"i\\\\n') out_file.write(' * ci : cosine of the relative angle before body i\\\\n')\",\n \"symbolic jacobian of a com point def write_symb_xgj(nb_bodies, Rj, xj, xgj, xgj_print, x_vector,\",\n \"for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1, sp.Symbol('c{}'.format(i+1)), sp.Symbol('s{}'.format(i+1))) R[i] =\",\n \"the symbolic vector for print def print_save_symb_vector(vector, start_name, end_name): new_vector = sp.zeros(3, 1)\",\n \"out_file.write(' // left wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"== nb_bodies-1: if flag_first: out_file.write(')/m_tot;\\\\n') else: out_file.write('0.0;\\\\n') if i != nb_bodies-1: out_file.write('\\\\n') else:\",\n \"out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n')\",\n \"1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left\",\n \"anchor point def write_symb_xj(nb_bodies, Rj, xj, xgj, xj_print, x_vector, index): # loop on\",\n \"for i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0\",\n \"2: return np.array([[0.0], [elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [elem]])\",\n \"0 for i in range(0,3): for j in range(0,3): if matrix[i,j] == 0\",\n \"= sp.Symbol('IMU{}{}'.format(i+1, j+1)) write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file,\",\n \"Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3): om[0][i]\",\n \"xj, xgj, x_vector[j], 'q{}'.format(i+1)) [xj[index-1][i], xj_print[index-1][i]] = print_save_symb_vector(new_vector, 'x{}_'.format(index), '_d{}'.format(i+1)) # write symbolic\",\n \"expressed in the relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write(' *\",\n \"# print all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r')\",\n \"inv_c_y_Lfoot = 1.0 / c_y_Lfoot;\\\\n\\\\n') out_file.write(' // right foot orientation angle derivatives [rad/s]\\\\n')\",\n \"nb_bodies): out_file.write(' s{} = sin(in_out.q_mot[{}]);\\\\n'.format(i+1, joint_id_names[i])) out_file.write('\\\\n // joint relative velocities\\\\n') for i\",\n \"arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] =\",\n \"'RightFootPitch_id', # right leg 'LeftHipPitch_id' , 'LeftHipRoll_id' , 'LeftHipYaw_id' , 'LeftKneePitch_id' , 'LeftFootRoll_id'\",\n \"# 06 12 # # 07 13 # # # ###################### # #\",\n \"*\\\\n') out_file.write(' * Di : position vector from the anchor point of the\",\n \"= sin(in_out.theta_Lfoot[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_Rfoot) || (!c_y_Lfoot))\\\\n {\\\\n') out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot\",\n \"left wrist absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lwrist[{}] = {};\\\\n'.format(i, x_l[i]))\",\n \"'') Rt[0] = R[0].T for i in range(1, nb_bodies): Rd[i] = get_rotation_matrix(rot_axis[i], 1,\",\n \"body\\\\n') out_file.write(' * Rdi : rotational matrix between body i and its predecessor\\\\n')\",\n \"-1, -1 # return true if it has a shape 'x%a_%b' (indexes %a,\",\n \"0.08, -0.045, 0.045) wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19,\",\n \"= np.array(['0', # waist 'RightHipPitch_id', 'RightHipRoll_id', 'RightHipYaw_id', 'RightKneePitch_id', 'RightFootRoll_id', 'RightFootPitch_id', # right leg\",\n \"flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rwrist_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write('\",\n \"'': term_list.pop(0) result = 0 # loop on all terms for cur_term in\",\n \"x[i], i+1) x[i] = write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1),\",\n \"11 # # 06 12 # # 07 13 # # # ######################\",\n \"of the previous body\\\\n') out_file.write(' * DGi : position vector from the anchor\",\n \"for print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9\",\n \"Rt[i] = R[i].T # jacobian rotation matrices out_file.write('\\\\n // jacobian rotation matrices\\\\n') out_file.write('\",\n \"all declarations def print_all_declaration(in_file, out_write, nb_max_char): count = 0 with open(in_file,'r') as f:\",\n \"Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') ,\",\n \"# Rdi : rotational matrix between body i and its predecessor # si\",\n \"new_vector = sp.zeros(3, 1) flag_print = 0 for i in range(0,3): if vector[i]\",\n \"* feet position, velocity and orientation\\\\n') out_file.write(' * waist and torso orientaion angles\",\n \", sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8') , sp.Symbol('L_3_8')]) Dg[3] = sp.Matrix([sp.Symbol('L_1_9') ,\",\n \"count >= nb_max_char: out_write.write(';\\\\n') count = 0 if count != 0: out_write.write(';\\\\n') #\",\n \"R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1])) out_file.write(' in_out.{}[2] = atan2({}, {});\\\\n'.format(angle_name, R_matrix[1], R_matrix[0])) else: #\",\n \"s_z_waist = sin(in_out.theta_waist[2]);\\\\n') out_file.write(' s_z_torso = sin(in_out.theta_torso[2]);\\\\n\\\\n') out_file.write(' if ((!c_y_waist) || (!c_y_torso))\\\\n {\\\\n')\",\n \"xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1),\",\n \"R_l_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"sp.Symbol('DPT_3_50')]) ## COM positions Dg = nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'),\",\n \"3): Dpt_l_foot_cont[i][j] = Dpt_r_foot_cont[i][j] x_r_cont = nb_contacts * [None] x_l_cont = nb_contacts *\",\n \"out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write(' }\\\\n') # omega out_file.write('\\\\n // joint absolute velocities\\\\n') Om =\",\n \"else: out_file.write('0.0;\\\\n') out_file.write('\\\\n') out_file.write(' // global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for\",\n \"om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\\\n // anchor\",\n \"foot orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[r_foot_id], 'omega_Rfoot', 'Rfoot') out_file.write('\\\\n') out_file.write(' // left\",\n \"x_min Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min\",\n \"xgj, x_r_cont[i][k], 'q{}'.format(j+1)) if cur_jac != 0: if not flag_first: flag_first = 1\",\n \"om[parent_id] + Rt[parent_id] * Om[i] om[i] = write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] =\",\n \"Dpt_l_foot[2] = sp.Symbol('DPT_3_29') Dpt_r_foot_cont = nb_contacts * [None] Dpt_l_foot_cont = nb_contacts * [None]\",\n \"derivative of xgi\\\\n') out_file.write(' * omi : absolute rotational vector of body i\\\\n')\",\n \"= y_max Dpt_r_foot_cont[2][1] = y_min Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts):\",\n \"== 0: out_write.write(' double {}'.format(cut_line_2[0].strip())) else: out_write.write(', {}'.format(cut_line_2[0].strip())) count += 1 if count\",\n \"= sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1] = r_wrist_y\",\n \"xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts =\",\n \"in range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xgj_print[i][j][k]\",\n \"neg_flag = 0 pos_str = elem_split[0] elif cur_len == 2: # negative neg_flag\",\n \"Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0])\",\n \"1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write('{}'.format(Rj_print[i][j][k])) out_file.write('\",\n \"of an element def print_declaration_elem(in_file, out_write, elem, nb_max_line): if count_elem(in_file, '{}'.format(elem)) >= 1:\",\n \"orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id,\",\n \"wrists_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj, 19, 23, -0.02, -0.005,\",\n \"declaration if len(line.split('// -- variables initialization -- //')) != 1: out_file.write(' // --\",\n \"new_vector = sp.zeros(3, 1) save_vector = 3 * [None] for i in range(0,3):\",\n \"# 07 13 # # # ###################### # # origin: in the waist,\",\n \"# xgji : jacobian of 'xgi' # Rji : jacobian of 'Ri' #\",\n \"nb_bodies, joint_id_names, M, xg, xgp, xgj): out_file.write(' m_tot = ') for i in\",\n \"in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU - angles velocity\\\\n') for i in range(0,\",\n \"def isRot(value): try: a = int(value.split('_')[0].split('R')[1]) b = int(value.split('_')[1][0]) c = int(value.split('_')[1][1]) return\",\n \"get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l =\",\n \"def symbolic_computation(out_file, nb_bodies, joint_id_names, rot_axis, parent_body_index, Dpt, Dg, M): out_file.write('\\\\n\\\\n // -- symbolic\",\n \"count += 1 return count # print the declaration of an element def\",\n \"// global com jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') for i in range(1, nb_bodies):\",\n \"write_symb_vector(out_file, x[i], 'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x\",\n \"in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute velocity\\\\n') for i\",\n \": rotational matrix between body i and its predecessor\\\\n') out_file.write(' * si :\",\n \"new_string)) new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper())\",\n \"velocity and orientation def feet_compute(out_file, joint_id_names, R, x, xp, om, Rj, xj, xgj,\",\n \"angle before body i\\\\n') out_file.write(' * ci : cosine of the relative angle\",\n \"out_file.write(' // left foot absolute position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] =\",\n \"out_file.write('\\\\n') out_file.write(' // right foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"return np.array([[0.0], [-elem], [0.0]]) elif axis == 3: return np.array([[0.0], [0.0], [-elem]]) else:\",\n \"xj, xgj, der_var) # one product elif cur_len == 2: result += der_elem(cur_term_split[0],\",\n \"(indexes %a, %b also returned) def isVec(value): try: a = int(value.split('_')[0].split('x')[1]) b =\",\n \"- s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{} *\",\n \"06 12 //\\\\n') out_file.write(' * // 07 13 //\\\\n') out_file.write(' * // //\\\\n')\",\n \"= {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute velocity\\\\n') for i in\",\n \"Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations\",\n \"Dpt[9] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_20')]) Dpt[10] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_22')]) Dpt[11] = sp.Matrix([0.0,\",\n \"angle before body i\\\\n') out_file.write(' *\\\\n') out_file.write(' * xi : absolute position vector\",\n \"cut_line = line.split(elem) if len(cut_line) == 2: count += 1 return count #\",\n \"= get_vector_axis(rot_axis[i], 1, sp.Symbol('Om{}'.format(i+1))) om[i] = om[parent_id] + Rt[parent_id] * Om[i] om[i] =\",\n \"absolute position vector of the COM G_i of body i\\\\n') out_file.write(' * xpi\",\n \"R_r_elb[j], 'q{}'.format(i+1)) if cur_jac != 0: if not flag_first: flag_first = 1 flag_print\",\n \"{} instead of 1 or 2 in negative detection !'.format(cur_len)) exit() # compute\",\n \"new_matrix, save_matrix # write symbolic jacobian of a rotation matrix def write_symb_Rj(nb_bodies, Rj,\",\n \"leg Dg[1] = sp.Matrix([sp.Symbol('L_1_7') , sp.Symbol('L_2_7') , sp.Symbol('L_3_7')]) Dg[2] = sp.Matrix([sp.Symbol('L_1_8') , sp.Symbol('L_2_8')\",\n \"if flag_first: out_file.write(' + {}*{}'.format(M[j], xg[j][i])) else: flag_first = 1 out_file.write('({}*xg{}_{}'.format(M[j], j+1, i+1))\",\n \"computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out] in_out inputs and outputs class\\\\n') out_file.write(' *\\\\n')\",\n \"c_z_{}*{} - s_z_{}*{};\\\\n'.format(omega_out, body_part, omega_in[1], body_part, omega_in[0])) out_file.write(' in_out.{}[2] = inv_c_y_{} * s_y_{}\",\n \"y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i in range(0,\",\n \"j, x_l_cont[i][j])) out_file.write('\\\\n') out_file.write(' // left foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"matrices out_file.write('\\\\n // rotation matrices\\\\n') R = nb_bodies*[None] Rt = nb_bodies*[None] Rd =\",\n \"= ' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','') new_vector[i] = sp.Symbol(elem_name) return new_vector, save_vector #\",\n \"range(1, nb_bodies): flag_print = 0 for k in range(0, 3): if xj_print[i][j][k] !=\",\n \"sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')])\",\n \"## anchor point positions Dpt = nb_bodies*[None] # waist Dpt[0] = sp.Matrix([0.0, 0.0,\",\n \"R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1], R_matrix[1]))\",\n \"derivative of xi # xgpi : derivative of xgi # omi : absolute\",\n \"= sp.zeros(3, 3) for i in range(0, 3): for j in range(0, 3):\",\n \"= in_out.IMU_Orientation[{}];\\\\n'.format(i+1, j+1, 3*i+j)) out_file.write('\\\\n // IMU - angles velocity\\\\n') for i in\",\n \"def torso_waist_angles(out_file, R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles\",\n \"write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n // jacobian anchor point positions\\\\n')\",\n \"variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100) out_file.write('\\\\n\\\\n') # copy temporary file out_file.write(line)\",\n \"arm Dpt[16] = sp.Matrix([sp.Symbol('DPT_1_36'), sp.Symbol('DPT_2_36'), sp.Symbol('DPT_3_36')]) Dpt[17] = sp.Matrix([0.0, sp.Symbol('DPT_2_39'), 0.0]) Dpt[18] =\",\n \"waist orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[waist_id], 'omega_waist', 'waist') out_file.write('\\\\n') out_file.write(' // torso\",\n \"# xg & xgp out_file.write('\\\\n // com absolute positions and velocities\\\\n') xg =\",\n \"'LeftFootRoll_id' , 'LeftFootPitch_id' , # left leg 'TorsoRoll_id' , 'TorsoPitch_id' , 'TorsoYaw_id' ,\",\n \"om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables\",\n \"transpose matrix of Ri\\\\n') out_file.write(' * xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' *\",\n \"= write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n // jacobian anchor point\",\n \"== max: out_file.write(';\\\\n') else: out_file.write(', ') # variables initialization def write_intialization(out_file, nb_bodies, joint_id_names):\",\n \"// 02 08 //\\\\n') out_file.write(' * // 03 09 //\\\\n') out_file.write(' * //\",\n \"in the relative\\\\n') out_file.write(' * frame of the previous body\\\\n') out_file.write(' * DGi\",\n \"om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb\",\n \"'') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for i in range(1, nb_bodies): parent_id\",\n \"Dg[7] = sp.Matrix([sp.Symbol('L_1_13'), sp.Symbol('L_2_13'), sp.Symbol('L_3_13')]) Dg[8] = sp.Matrix([sp.Symbol('L_1_14'), sp.Symbol('L_2_14'), sp.Symbol('L_3_14')]) Dg[9] = sp.Matrix([sp.Symbol('L_1_15'),\",\n \"x, xp, om, Rj, xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): #\",\n \"variables initialization -- //')) != 1: out_file.write(' // -- variables declaration -- //\\\\n\\\\n')\",\n \"'theta_waist', R[waist_id], 1) out_file.write('\\\\n') out_file.write(' // torso orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file,\",\n \"'RightShRoll_id' , 'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id'\",\n \"nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None]\",\n \"= write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n // jacobian com absolute\",\n \"relative frame of the current body i # Omi : rotational vector from\",\n \"apply negative if neg_flag: result = -result return result # compute the derivative\",\n \"* s_y_{} * (s_z_{}*{} + c_z_{}*{}) + {};\\\\n'.format(omega_out, body_part, body_part, body_part, omega_in[1], body_part,\",\n \"# computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T\",\n \"double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count >=\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Rwrist_or_der[{}][{}] =\",\n \"range(1, nb_bodies): parent_id = parent_body_index[i] x[i] = x[parent_id] + Rt[parent_id] * Dpt[i] xp[i]\",\n \"= nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies):\",\n \"<filename>symbolicR/python/forward_kin.py import numpy as np import sympy as sp import re import os\",\n \"rotation matrix if rot_flag: result += Rj[a-1][der_q-1][(b-1)*3+(c-1)] # vector elif vec_flag: result +=\",\n \"symbolic output file def gen_symbolic_out(out_file_name, nb_bodies, rot_axis, parent_body_index, joint_id_names, Dpt, Dg, M): #\",\n \"position\\\\n') for i in range(0,3): out_file.write(' in_out.r_Lfoot[{}] = {};\\\\n'.format(i, x_l[i])) out_file.write('\\\\n') out_file.write(' //\",\n \"anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in\",\n \"= write_symb_vector(out_file, om[i], 'om{}_'.format(i+1), '') om_tilde[i] = get_tilde(om[i]) # x & xp out_file.write('\\\\n\",\n \"R, om, waist_id, torso_id): out_file.write(' // waist orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file,\",\n \"# print matrix components declaration def write_matrix_declaration(out_file, prefix): out_file.write(' double ') for i\",\n \"out_file.write(' inv_c_y_waist = 1.0 / c_y_waist;\\\\n') out_file.write(' inv_c_y_torso = 1.0 / c_y_torso;\\\\n\\\\n') out_file.write('\",\n \"R[i] = Rd[i] * R[parent_body_index[i]] write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[i], i+1) R[i]\",\n \"sp.Symbol('DPT_3_24')]) Dpt[12] = sp.Matrix([0.0, 0.0, 0.0]) # trunk Dpt[13] = sp.Matrix([sp.Symbol('DPT_1_4'), 0.0, sp.Symbol('DPT_3_4')])\",\n \"for 0 or 1 !'.format(cur_len-1)) exit() return result # write the beginning of\",\n \"sp.Symbol('L_3_28')]) Dg[23] = sp.Matrix([sp.Symbol('L_1_29'), sp.Symbol('L_2_29'), sp.Symbol('L_3_29')]) # masses M = np.array([ 'M_6', #\",\n \"open('./{}.cc'.format(out_file_name), 'w') with open(in_temp, 'r') as f: # loop on all the lines\",\n \"08 # # 03 09 # # 04 10 # # 05 11\",\n \"is not always body i-1), expressed in the relative # frame of the\",\n \"range(0,3): out_file.write(' in_out.r_Rwrist[{}] = {};\\\\n'.format(i, x_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist absolute velocity\\\\n')\",\n \"nb_bodies*[None] Om[0] = sp.zeros(3, 1) om[0] = sp.zeros(3, 1) for i in range(0,3):\",\n \"out_file.write('\\\\n') out_file.write(' // torso orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1)\",\n \"in_out.{}[0] = atan2(-{}, -{});\\\\n'.format(angle_name, R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, -sqrt({}*{} + {}*{}));\\\\n'.format(angle_name,\",\n \"Dpt[5] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_12')]) Dpt[6] = sp.Matrix([0.0, 0.0, 0.0]) # left leg\",\n \"nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write(' + ') out_file.write(' //\",\n \"0 if count != 0: out_write.write(';\\\\n') # print all declarations def print_all_declaration(in_file, out_write,\",\n \"sp.Symbol('c{}'.format(der_q)) # other else: [rot_flag, a, b, c] = isRot(pos_str) [vec_flag, d, e]\",\n \"= 0 for i in range(0, nb_contacts): for j in range (1, nb_bodies):\",\n \"i in range(0, 9): out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right\",\n \"middle point between the two pitch hip rotations\\\\n') out_file.write(' * inertial frame: located\",\n \"derivative of xgi # omi : absolute rotational vector of body i #\",\n \"rot_axis, parent_body_index, Dpt, Dg, M) # end of the file write_file_end(file_temp) file_temp.close() #\",\n \": absolute rotational matrix\\\\n') out_file.write(' * Rti : transpose matrix of Ri\\\\n') out_file.write('\",\n \"Ri # xji : jacobian of 'xi' # xgji : jacobian of 'xgi'\",\n \"if xgj_print[i][j][k] != None: if not flag_first: flag_first = 1 flag_print = 1\",\n \"rotational matrix between body i and its predecessor # si : sine of\",\n \"out_file.write(' * // 17 16 21 //\\\\n') out_file.write(' * // 18 15 22\",\n \"2: new_string = cur_split[0] for i in range(1, len(cur_split)-1): new_string = '{}{}'.format(new_string, cur_split[i])\",\n \"om_tilde_r_foot = get_tilde(om_r_foot) om_tilde_l_foot = get_tilde(om_l_foot) x_r = x_r_foot + R_r_foot.T * Dpt_r_foot\",\n \"not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write('\",\n \"6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main\",\n \"= get_tilde(om[i]) # x & xp out_file.write('\\\\n // anchor point absolute positions and\",\n \"xj, xgj, r_foot_id, l_foot_id, x_min, x_max, y_min, y_max): # symbolic variables declarations nb_contacts\",\n \"== 3: return np.array([[0.0], [0.0], [-elem]]) else: return np.array([]) # compute the derivative\",\n \"if i == 2 and j == 2: out_file.write(';\\\\n') else: out_file.write(', ') #\",\n \"(for jacobian) def symbolic_jacob_der(Rj, xj, xgj, symb_var, der_var): # list of all terms\",\n \"+= xj[d-1][der_q-1][e-1] # apply negative if neg_flag: result = -result return result #\",\n \"import os ###################### # # # 17 16 21 # # 18 15\",\n \"* // 17 16 21 //\\\\n') out_file.write(' * // 18 15 22 //\\\\n')\",\n \"= 1 elem_name = '{}{}{}'.format(start_name, i+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, vector[i]).replace('1.0*','')) new_vector[i]\",\n \"print def print_save_symb_matrix(matrix, start_name, end_name): new_matrix = sp.zeros(3, 3) save_matrix = 9 *\",\n \"DGi : position vector from the anchor point of body i to its\",\n \"xj, xgj and xgj (jacobian) Rj = nb_bodies*[None] xj = nb_bodies*[None] xgj =\",\n \"sp.Symbol('L_3_17')]) Dg[12] = sp.Matrix([sp.Symbol('L_1_18'), 0.0 , sp.Symbol('L_3_18')]) # trunk Dg[13] = sp.Matrix([sp.Symbol('L_1_19'), sp.Symbol('L_2_19'),\",\n \"body i and its predecessor\\\\n') out_file.write(' * si : sine of the relative\",\n \"'theta_Lfoot', R[l_foot_id], 1) out_file.write('\\\\n') out_file.write(' c_y_Rfoot = cos(in_out.theta_Rfoot[1]);\\\\n') out_file.write(' c_y_Lfoot = cos(in_out.theta_Lfoot[1]);\\\\n') out_file.write('\",\n \"{\\\\n') flag_first = 0 for i in range (1, nb_bodies): flag_print = 0\",\n \"in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get a string for the\",\n \"right foot contact points jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for\",\n \"[None] for i in range(0,3): for j in range(0,3): if matrix[i,j] == 0\",\n \"{};\\\\n'.format(i, xp_r[i])) out_file.write('\\\\n') out_file.write(' // right wrist jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first\",\n \"xj = nb_bodies*[None] xgj = nb_bodies*[None] Rj_print = nb_bodies*[None] xj_print = nb_bodies*[None] xgj_print\",\n \"for i in range(0, nb_bodies): out_file.write('{}'.format(M[i])) if i == nb_bodies-1: out_file.write(';\\\\n\\\\n') else: out_file.write('\",\n \"'x{}_'.format(i+1), '') xp[i] = write_symb_vector(out_file, xp[i], 'xp{}_'.format(i+1), '') # jacobian x out_file.write('\\\\n //\",\n \"R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] = r_wrist_x Dpt_r_wrist[1]\",\n \"= sp.Matrix([sp.Symbol('L_1_20'), sp.Symbol('L_2_20'), sp.Symbol('L_3_20')]) Dg[15] = sp.Matrix([sp.Symbol('L_1_21'), sp.Symbol('L_2_21'), sp.Symbol('L_3_21')]) # right arm Dg[16]\",\n \"Dpt_r_foot_cont[3][1] = y_max for i in range(0, nb_contacts): Dpt_r_foot_cont[i][2] = sp.Symbol('DPT_3_16') for i\",\n \"absolute position\\\\n') for i in range(0, nb_contacts): for j in range(0, 3): out_file.write('\",\n \"xi : absolute position vector (from origin, expressed in the inertial frame)\\\\n') out_file.write('\",\n \"out_file.write('{}'.format(xj_print[i][j][k])) out_file.write(' }\\\\n') # xg & xgp out_file.write('\\\\n // com absolute positions and\",\n \"[elem]]) else: return np.array([]) else: if axis == 1: return np.array([[-elem], [0.0], [0.0]])\",\n \"origin, expressed in the inertial frame) # of the anchor point of body\",\n \"foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range\",\n \"write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R[0], 1) R[0] = write_symb_matrix(out_file, R[0], 'R1_', '')\",\n \"2, 1, 3, 2, 1, 2, # left leg 1, 2, 3, #\",\n \", 'TorsoPitch_id' , 'TorsoYaw_id' , # trunk 'RightShPitch_id' , 'RightShRoll_id' , 'RightShYaw_id' ,\",\n \"out_file.write(' * ////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * // 17 16 21\",\n \"= sp.Matrix([0.0, 0.0, 0.0]) # left leg Dpt[7] = sp.Matrix([0.0, sp.Symbol('DPT_2_3'), 0.0]) Dpt[8]\",\n \"1: return np.array([[-elem], [0.0], [0.0]]) elif axis == 2: return np.array([[0.0], [-elem], [0.0]])\",\n \"= xp_r_foot + om_tilde_r_foot * (R_r_foot.T * Dpt_r_foot) xp_l = xp_l_foot + om_tilde_l_foot\",\n \"1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_cont_der[{}][{}][{}] = {};\\\\n'.format(i, get_string_enum(joint_id_names[j]),\",\n \"Dpt[21] = sp.Matrix([0.0, sp.Symbol('DPT_2_46'), 0.0]) Dpt[22] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_48')]) Dpt[23] = sp.Matrix([0.0,\",\n \"= '{}{}{}{}'.format(start_name, i+1, j+1, end_name) out_file.write(' {} = {};\\\\n'.format(elem_name, matrix[i,j]).replace('1.0*','')) new_matrix[i,j] = sp.Symbol(elem_name)\",\n \"= xp[r_foot_id] xp_l_foot = xp[l_foot_id] om_r_foot = om[r_foot_id] om_l_foot = om[l_foot_id] R_r_foot =\",\n \"current body i \\\\n') out_file.write(' * (previous body is not always body i-1),\",\n \"new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return new_string # write the end of the file\",\n \"// joint relative velocities\\\\n') for i in range(1, nb_bodies): out_file.write(' Om{} = in_out.qd_mot[{}];\\\\n'.format(i+1,\",\n \"all the vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj,\",\n \"vector elements for j in range(0, 3): new_vector[j] = symbolic_jacob_der(Rj, xj, xgj, x_vector[j],\",\n \"cur_jac)) out_file.write(' }\\\\n\\\\n') # get a string for the enumeration of joints def\",\n \"= line.split(' = ') if len(cut_line_1) == 2 and len(cut_line_2) == 2: if\",\n \"out_file.write(' in_out.Lfoot_or[{}] = {};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute orientation jacobian\\\\n')\",\n \"out_file.write(' return;\\\\n }\\\\n\\\\n') out_file.write(' inv_c_y_Rfoot = 1.0 / c_y_Rfoot;\\\\n') out_file.write(' inv_c_y_Lfoot = 1.0\",\n \"negative if neg_flag: result = -result return result # compute the derivative of\",\n \"0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) else: if axis == 1: return\",\n \"0.0], [sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([]) # get vector\",\n \"-r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation om_tilde_r_elb = get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r\",\n \"xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None] for\",\n \"sp.Matrix([sp.Symbol('L_1_15'), sp.Symbol('L_2_15'), sp.Symbol('L_3_15')]) Dg[10] = sp.Matrix([sp.Symbol('L_1_16'), sp.Symbol('L_2_16'), sp.Symbol('L_3_16')]) Dg[11] = sp.Matrix([sp.Symbol('L_1_17'), sp.Symbol('L_2_17'), sp.Symbol('L_3_17')])\",\n \"new_string = cur_split[0].upper() for i in range(1, len(cur_split)): new_string = '{}_{}'.format(new_string, cur_split[i].upper()) return\",\n \"orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i in range\",\n \"= x[r_elb_id] x_l_elb = x[l_elb_id] xp_r_elb = xp[r_elb_id] xp_l_elb = xp[l_elb_id] om_r_elb =\",\n \"Rdi : rotational matrix between body i and its predecessor # si :\",\n \"= in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n') for i in range(1, nb_bodies): out_file.write('\",\n \"xj, xgj, symb_var, der_var): # list of all terms term_list = str(symb_var).replace('- ','-').replace('-','+-').split('+')\",\n \"and derivatives\\\\n') out_file.write(' *\\\\n') out_file.write(' * ////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write(' *\",\n \"[rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write(' c_y_waist = cos(in_out.theta_waist[1]);\\\\n') out_file.write(' c_y_torso =\",\n \"out_file.write('\\\\n') out_file.write(' // torso orientation angle derivatives [rad/s]\\\\n') theta_dot_compute(out_file, om[torso_id], 'omega_torso', 'torso') #\",\n \"xgj_print, xg[i], i+1) xg[i] = write_symb_vector(out_file, xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i],\",\n \"'RightShYaw_id' , 'RightElbPitch_id', # right arm 'LeftShPitch_id' , 'LeftShRoll_id' , 'LeftShYaw_id' , 'LeftElbPitch_id'\",\n \"= R[r_elb_id] R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1)\",\n \"-1, -1 # count the number of 'elem' in the file def count_elem(in_file,\",\n \"out_write.write(' double {}'.format(cut_line[0].strip())) else: out_write.write(', {}'.format(cut_line[0].strip())) count += len(cut_line[0].strip()) + 2 if count\",\n \"if i == 1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\\\n\",\n \"15, 20, 21, 22 # left arm ]) nb_bodies = len(parent_body_index) ## anchor\",\n \"out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') # get a string for\",\n \"flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] =\",\n \"Dpt_r_foot_cont[1][0] = x_min Dpt_r_foot_cont[2][0] = x_max Dpt_r_foot_cont[3][0] = x_max Dpt_r_foot_cont[0][1] = y_min Dpt_r_foot_cont[1][1]\",\n \"in range(0, nb_contacts): for j in range(0, 3): out_file.write(' in_out.r_Rfoot_cont[{}][{}] = {};\\\\n'.format(i, j,\",\n \"return new_string # write the end of the file def write_file_end(out_file): out_file.write('}\\\\n') #\",\n \"Ri\\\\n') out_file.write(' * xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian\",\n \"cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_r_foot[j], 'q{}'.format(i+1)) if cur_jac != 0: if not\",\n \"out_file.write('\\\\n // joint cosines\\\\n') for i in range(1, nb_bodies): out_file.write(' c{} = cos(in_out.q_mot[{}]);\\\\n'.format(i+1,\",\n \"else: print('Error: {} * counted , only implemented for 0 or 1 !'.format(cur_len-1))\",\n \"// right foot jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first = 0 for i\",\n \"* xji : jacobian of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian of \\\\'xgi\\\\'\\\\n')\",\n \"for j in range(0,3): if matrix[i,j] == 0 or matrix[i,j] == 1: new_matrix[i,j]\",\n \"xg out_file.write('\\\\n // jacobian com absolute positions\\\\n') out_file.write(' if (flag_jacob)\\\\n {\\\\n') flag_first =\",\n \"G_i,\\\\n') out_file.write(' * expressed in the relative frame of the current body i\\\\n')\",\n \"out_file.write('// joints enumeration\\\\n') out_file.write('enum {') count = 0 for i in range(1, len(joint_id_names)):\",\n \"xj_print = nb_bodies*[None] xgj_print = nb_bodies*[None] for i in range(0, nb_bodies): Rj[i] =\",\n \"flag_first = 1 flag_print = 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n')\",\n \"0.0, 0.0]) # right leg Dpt[1] = sp.Matrix([0.0, sp.Symbol('DPT_2_2'), 0.0]) Dpt[2] = sp.Matrix([0.0,\",\n \"x_r_elb + R_r_elb.T * Dpt_r_wrist x_l = x_l_elb + R_l_elb.T * Dpt_l_wrist xp_r\",\n \"der_elem(elem_str, Rj, xj, xgj, der_var): # element to derive (string) elem_str = elem_str.replace('-\",\n \"nb_bodies*[None] xgj[i] = nb_bodies*[None] Rj_print[i] = nb_bodies*[None] xj_print[i] = nb_bodies*[None] xgj_print[i] = nb_bodies*[None]\",\n \"'') # jacobian x out_file.write('\\\\n // jacobian anchor point positions\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"vector of the COM G_i of body i # xpi : derivative of\",\n \"# expressed in the relative frame of the current body i # Omi\",\n \"out_file.write(', {}'.format(get_string_enum(joint_id_names[i]))) out_file.write('};\\\\n\\\\n') out_file.write('/*! \\\\\\\\brief main kinematics computation\\\\n') out_file.write(' *\\\\n') out_file.write(' * \\\\\\\\param[in,out]\",\n \"{};\\\\n'.format(i, R_l_foot[i])) out_file.write('\\\\n') out_file.write(' // right foot absolute orientation jacobian\\\\n') out_file.write(' if (flag_jacob)\\\\n\",\n \"matrix def write_symb_Rj(nb_bodies, Rj, xj, xgj, Rj_print, R_matrix, index): # loop on all\",\n \"return np.array([[cosine, sine, 0.0], [-sine, cosine, 0.0], [0.0, 0.0, 1.0]]) else: return np.array([])\",\n \"x_l_cont[i] = x_l_foot + R_l_foot.T * Dpt_l_foot_cont[i] # writing outputs out_file.write(' // right\",\n \"for j in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_elb[j], 'q{}'.format(i+1)) if cur_jac\",\n \"// 19 14 23 //\\\\n') out_file.write(' * // 20 01 24 //\\\\n') out_file.write('\",\n \"# variables initialization def write_intialization(out_file, nb_bodies, joint_id_names): out_file.write(' // -- variables initialization --\",\n \"2: if len(cut_line[0].split('[')) == 1: if count == 0: out_write.write(' double {}'.format(cut_line[0].strip())) else:\",\n \"0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20] = sp.Matrix([sp.Symbol('DPT_1_37'), sp.Symbol('DPT_2_37'), sp.Symbol('DPT_3_37')]) Dpt[21] = sp.Matrix([0.0,\",\n \"nb_bodies): new_vector = sp.zeros(3, 1) # loop on all the vector elements for\",\n \"if count_elem(in_file, '{}'.format(elem)) >= 1: count = 0 with open(in_file,'r') as f: #\",\n \"i in range(0, nb_bodies): for j in range(1, nb_bodies): flag_print = 0 for\",\n \"x[0] = write_symb_vector(out_file, x[0], 'x1_', '') xp[0] = write_symb_vector(out_file, xp[0], 'xp1_', '') for\",\n \"G_i of body i\\\\n') out_file.write(' * xpi : derivative of xi\\\\n') out_file.write(' *\",\n \"matrix def get_rotation_matrix(axis, direct, cosine, sine): if direct: if axis == 1: return\",\n \"nb_bodies-1: out_file.write('\\\\n') else: out_file.write(' }\\\\n\\\\n') # from an orientation matrix, compute the roll,\",\n \"'.format(get_string_enum(joint_id_names[i]), j)) flag_first = 0 for k in range(0, nb_bodies): if xgj[k][i][j] !=\",\n \"= 1 elif not flag_print: flag_print = 1 out_file.write('\\\\n') out_file.write(' in_out.Lwrist_or_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]),\",\n \"of \\\\'xi\\\\'\\\\n') out_file.write(' * xgji : jacobian of \\\\'xgi\\\\'\\\\n') out_file.write(' * Rji :\",\n \"// torso orientation matrix as angles [rad]\\\\n') yaw_pitch_roll_angles(out_file, 'theta_torso', R[torso_id], 1) out_file.write('\\\\n') out_file.write('\",\n \"////////////////////////\\\\n') out_file.write(' * // //\\\\n') out_file.write(' * // 17 16 21 //\\\\n') out_file.write('\",\n \"omega_{} = in_out.IMU_Angular_Rate[{}];\\\\n'.format(i+1, i)) out_file.write('\\\\n // joint cosines\\\\n') for i in range(1, nb_bodies):\",\n \"xg[i], 'xg{}_'.format(i+1), '') xgp[i] = write_symb_vector(out_file, xgp[i], 'xgp{}_'.format(i+1), '') # jacobian xg out_file.write('\\\\n\",\n \"1: out_file.write('{}'.format(get_string_enum(joint_id_names[i]))) elif count >= 6: count = 0 out_file.write(',\\\\n {}'.format(get_string_enum(joint_id_names[i]))) else: out_file.write(',\",\n \"arm 2, 1, 3, 2 # left arm ]) # parent index parent_body_index\",\n \"R_l_elb = R[l_elb_id] Dpt_r_wrist = sp.zeros(3, 1) Dpt_l_wrist = sp.zeros(3, 1) Dpt_r_wrist[0] =\",\n \"= sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_41')]) Dpt[19] = sp.Matrix([0.0, 0.0, sp.Symbol('DPT_3_43')]) # left arm Dpt[20]\",\n \"R_matrix[5], R_matrix[8])) out_file.write(' in_out.{}[1] = atan2(-{}, sqrt({}*{} + {}*{}));\\\\n'.format(angle_name, R_matrix[2], R_matrix[0], R_matrix[0], R_matrix[1],\",\n \"= r_wrist_z Dpt_l_wrist[0] = r_wrist_x Dpt_l_wrist[1] = -r_wrist_y Dpt_l_wrist[2] = r_wrist_z # computation\",\n \"= get_tilde(om_r_elb) om_tilde_l_elb = get_tilde(om_l_elb) x_r = x_r_elb + R_r_elb.T * Dpt_r_wrist x_l\",\n \"i in range(0, nb_bodies): Rj[i] = nb_bodies*[None] xj[i] = nb_bodies*[None] xgj[i] = nb_bodies*[None]\",\n \"out_file.write('\\\\n') out_file.write(' in_out.r_Rfoot_der[{}][{}] = {};\\\\n'.format(get_string_enum(joint_id_names[i]), j, cur_jac)) out_file.write(' }\\\\n\\\\n') out_file.write(' // left foot\",\n \"out_file.write(';\\\\n') else: out_file.write(', ') # print variables declaration def write_variables_declaration(out_file, prefix, min, max):\",\n \"//')) != 1: out_file.write(' // -- variables declaration -- //\\\\n\\\\n') print_all_declaration(in_temp, out_file, 100)\",\n \"out_file.write(' * \\\\\\\\param[in,out] in_out inputs and outputs class\\\\n') out_file.write(' *\\\\n') out_file.write(' * computation\",\n \"Dpt_r_wrist) xp_l = xp_l_elb + om_tilde_l_elb * (R_l_elb.T * Dpt_l_wrist) # writing outputs\",\n \"out_file.write(' in_out.rp_COM[{}] = '.format(i)) flag_first = 0 for j in range(0, nb_bodies): if\",\n \"in range(0,9): cur_jac = symbolic_jacob_der(Rj, xj, xgj, R_l_foot[j], 'q{}'.format(i+1)) if cur_jac != 0:\",\n \"== 2: return np.array([[cosine, 0.0, -sine], [0.0, 1.0, 0.0], [sine, 0.0, cosine]]) elif\",\n \"vector of the COM G_i of body i\\\\n') out_file.write(' * xpi : derivative\"\n]","isConversational":false}}},{"rowIdx":2697,"cells":{"ngram":{"kind":"list like","value":["required=True, help=\"can be defined in days, hours, minutes (for example 5d or 12h","minutes (for example 5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args","misp_key, misp_verifycert import argparse import os import json # Usage for pipe masters:","'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events","events from a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined in days,","out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that","<reponame>0xiso/PyMISP #!/usr/bin/env python # -*- coding: utf-8 -*- from pymisp import PyMISP from","instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined in days, hours, minutes (for example","if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.')","| jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m,","PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if out","from a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined in days, hours,","masters: ./last.py -l 5h | jq . def init(url, key): return PyMISP(url, key,","= argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be","from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import","None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that time period')","result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else: with open(out,","\"--last\", required=True, help=\"can be defined in days, hours, minutes (for example 5d or","else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser =","if args.output is not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0)","json # Usage for pipe masters: ./last.py -l 5h | jq . def","keys import misp_url, misp_key, misp_verifycert import argparse import os import json # Usage","None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key)","with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download","if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0)","that time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__","Usage for pipe masters: ./last.py -l 5h | jq . def init(url, key):","args = parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file already","file\") args = parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file","args.output is not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp","python # -*- coding: utf-8 -*- from pymisp import PyMISP from keys import","os import json # Usage for pipe masters: ./last.py -l 5h | jq","hours, minutes (for example 5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\")","help=\"Output file\") args = parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output","download_last(m, last, out=None): result = m.download_last(last) if out is None: if 'response' in","as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from","def download_last(m, last, out=None): result = m.download_last(last) if out is None: if 'response'","misp_url, misp_key, misp_verifycert import argparse import os import json # Usage for pipe","out=None): result = m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response']))","key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if out is","print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key) download_last(misp, args.last, args.output)","import os import json # Usage for pipe masters: ./last.py -l 5h |","return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if","print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else: with open(out, 'w')","'__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True,","-*- from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse","not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url,","and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key) download_last(misp,","# -*- coding: utf-8 -*- from pymisp import PyMISP from keys import misp_url,","import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import os import","\"--output\", help=\"Output file\") args = parser.parse_args() if args.output is not None and os.path.exists(args.output):",". def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None):","key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last)","defined in days, hours, minutes (for example 5d or 12h or 30m).\") parser.add_argument(\"-o\",","-*- coding: utf-8 -*- from pymisp import PyMISP from keys import misp_url, misp_key,","import argparse import os import json # Usage for pipe masters: ./last.py -l","'json') def download_last(m, last, out=None): result = m.download_last(last) if out is None: if","./last.py -l 5h | jq . def init(url, key): return PyMISP(url, key, misp_verifycert,","#!/usr/bin/env python # -*- coding: utf-8 -*- from pymisp import PyMISP from keys","parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can","PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import os import json","utf-8 -*- from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import","== '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\"-l\", \"--last\",","be defined in days, hours, minutes (for example 5d or 12h or 30m).\")","init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result =","pipe masters: ./last.py -l 5h | jq . def init(url, key): return PyMISP(url,","example 5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args = parser.parse_args()","# Usage for pipe masters: ./last.py -l 5h | jq . def init(url,","= m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No","time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ ==","in result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else: with","help=\"can be defined in days, hours, minutes (for example 5d or 12h or","else: print('No results for that time period') exit(0) else: with open(out, 'w') as","30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args = parser.parse_args() if args.output is not None","period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__':","results for that time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response']))","m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results","if out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for","a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined in days, hours, minutes","misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if out is None:","print('No results for that time period') exit(0) else: with open(out, 'w') as f:","last, out=None): result = m.download_last(last) if out is None: if 'response' in result:","is not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp =","misp_verifycert import argparse import os import json # Usage for pipe masters: ./last.py","'response' in result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else:","result = m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response'])) else:","jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last,","in days, hours, minutes (for example 5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\",","argparse import os import json # Usage for pipe masters: ./last.py -l 5h","import json # Usage for pipe masters: ./last.py -l 5h | jq .","5h | jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def","import misp_url, misp_key, misp_verifycert import argparse import os import json # Usage for","MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined in days, hours, minutes (for","coding: utf-8 -*- from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert","for pipe masters: ./last.py -l 5h | jq . def init(url, key): return","for that time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if","-l 5h | jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json')","from keys import misp_url, misp_key, misp_verifycert import argparse import os import json #","days, hours, minutes (for example 5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output","parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined in days, hours, minutes (for example 5d","__name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\"-l\",","pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import os","parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args = parser.parse_args() if args.output is not None and","12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args = parser.parse_args() if args.output is","def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result","is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that time","f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP","latest events from a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined in","os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key) download_last(misp, args.last,","or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args = parser.parse_args() if args.output","f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a","argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\"-l\", \"--last\", required=True, help=\"can be defined","or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args = parser.parse_args() if args.output is not","5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args = parser.parse_args() if","= parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file already exists,","exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser","open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest","parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file already exists, abord.')","(for example 5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args ="],"string":"[\n \"required=True, help=\\\"can be defined in days, hours, minutes (for example 5d or 12h\",\n \"minutes (for example 5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args\",\n \"misp_key, misp_verifycert import argparse import os import json # Usage for pipe masters:\",\n \"'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events\",\n \"events from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined in days,\",\n \"out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that\",\n \"<reponame>0xiso/PyMISP #!/usr/bin/env python # -*- coding: utf-8 -*- from pymisp import PyMISP from\",\n \"instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined in days, hours, minutes (for example\",\n \"if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.')\",\n \"| jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m,\",\n \"PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if out\",\n \"from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined in days, hours,\",\n \"masters: ./last.py -l 5h | jq . def init(url, key): return PyMISP(url, key,\",\n \"= argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be\",\n \"from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import\",\n \"None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that time period')\",\n \"result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else: with open(out,\",\n \"\\\"--last\\\", required=True, help=\\\"can be defined in days, hours, minutes (for example 5d or\",\n \"else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser =\",\n \"if args.output is not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0)\",\n \"json # Usage for pipe masters: ./last.py -l 5h | jq . def\",\n \"keys import misp_url, misp_key, misp_verifycert import argparse import os import json # Usage\",\n \"None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key)\",\n \"with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download\",\n \"if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0)\",\n \"that time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__\",\n \"Usage for pipe masters: ./last.py -l 5h | jq . def init(url, key):\",\n \"args = parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file already\",\n \"file\\\") args = parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file\",\n \"args.output is not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp\",\n \"python # -*- coding: utf-8 -*- from pymisp import PyMISP from keys import\",\n \"os import json # Usage for pipe masters: ./last.py -l 5h | jq\",\n \"hours, minutes (for example 5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\")\",\n \"help=\\\"Output file\\\") args = parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output\",\n \"download_last(m, last, out=None): result = m.download_last(last) if out is None: if 'response' in\",\n \"as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from\",\n \"def download_last(m, last, out=None): result = m.download_last(last) if out is None: if 'response'\",\n \"misp_url, misp_key, misp_verifycert import argparse import os import json # Usage for pipe\",\n \"out=None): result = m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response']))\",\n \"key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if out is\",\n \"print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key) download_last(misp, args.last, args.output)\",\n \"import os import json # Usage for pipe masters: ./last.py -l 5h |\",\n \"return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if\",\n \"print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else: with open(out, 'w')\",\n \"'__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True,\",\n \"-*- from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse\",\n \"not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url,\",\n \"and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key) download_last(misp,\",\n \"# -*- coding: utf-8 -*- from pymisp import PyMISP from keys import misp_url,\",\n \"import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import os import\",\n \"\\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args() if args.output is not None and os.path.exists(args.output):\",\n \". def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None):\",\n \"key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last)\",\n \"defined in days, hours, minutes (for example 5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\",\",\n \"-*- coding: utf-8 -*- from pymisp import PyMISP from keys import misp_url, misp_key,\",\n \"import argparse import os import json # Usage for pipe masters: ./last.py -l\",\n \"'json') def download_last(m, last, out=None): result = m.download_last(last) if out is None: if\",\n \"./last.py -l 5h | jq . def init(url, key): return PyMISP(url, key, misp_verifycert,\",\n \"#!/usr/bin/env python # -*- coding: utf-8 -*- from pymisp import PyMISP from keys\",\n \"parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can\",\n \"PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import os import json\",\n \"utf-8 -*- from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import\",\n \"== '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\",\",\n \"be defined in days, hours, minutes (for example 5d or 12h or 30m).\\\")\",\n \"init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result =\",\n \"pipe masters: ./last.py -l 5h | jq . def init(url, key): return PyMISP(url,\",\n \"example 5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args()\",\n \"# Usage for pipe masters: ./last.py -l 5h | jq . def init(url,\",\n \"= m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No\",\n \"time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ ==\",\n \"in result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else: with\",\n \"help=\\\"can be defined in days, hours, minutes (for example 5d or 12h or\",\n \"else: print('No results for that time period') exit(0) else: with open(out, 'w') as\",\n \"30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args() if args.output is not None\",\n \"period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__':\",\n \"results for that time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response']))\",\n \"m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results\",\n \"if out is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for\",\n \"a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined in days, hours, minutes\",\n \"misp_verifycert, 'json') def download_last(m, last, out=None): result = m.download_last(last) if out is None:\",\n \"print('No results for that time period') exit(0) else: with open(out, 'w') as f:\",\n \"last, out=None): result = m.download_last(last) if out is None: if 'response' in result:\",\n \"is not None and os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp =\",\n \"misp_verifycert import argparse import os import json # Usage for pipe masters: ./last.py\",\n \"'response' in result: print(json.dumps(result['response'])) else: print('No results for that time period') exit(0) else:\",\n \"result = m.download_last(last) if out is None: if 'response' in result: print(json.dumps(result['response'])) else:\",\n \"jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last,\",\n \"in days, hours, minutes (for example 5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\",\",\n \"argparse import os import json # Usage for pipe masters: ./last.py -l 5h\",\n \"import json # Usage for pipe masters: ./last.py -l 5h | jq .\",\n \"5h | jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def\",\n \"import misp_url, misp_key, misp_verifycert import argparse import os import json # Usage for\",\n \"MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined in days, hours, minutes (for\",\n \"coding: utf-8 -*- from pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert\",\n \"for pipe masters: ./last.py -l 5h | jq . def init(url, key): return\",\n \"for that time period') exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if\",\n \"-l 5h | jq . def init(url, key): return PyMISP(url, key, misp_verifycert, 'json')\",\n \"from keys import misp_url, misp_key, misp_verifycert import argparse import os import json #\",\n \"days, hours, minutes (for example 5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output\",\n \"parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined in days, hours, minutes (for example 5d\",\n \"__name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\\\"-l\\\",\",\n \"pymisp import PyMISP from keys import misp_url, misp_key, misp_verifycert import argparse import os\",\n \"parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args() if args.output is not None and\",\n \"12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args() if args.output is\",\n \"def init(url, key): return PyMISP(url, key, misp_verifycert, 'json') def download_last(m, last, out=None): result\",\n \"is None: if 'response' in result: print(json.dumps(result['response'])) else: print('No results for that time\",\n \"f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a MISP\",\n \"latest events from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined in\",\n \"os.path.exists(args.output): print('Output file already exists, abord.') exit(0) misp = init(misp_url, misp_key) download_last(misp, args.last,\",\n \"or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args() if args.output\",\n \"f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest events from a\",\n \"argparse.ArgumentParser(description='Download latest events from a MISP instance.') parser.add_argument(\\\"-l\\\", \\\"--last\\\", required=True, help=\\\"can be defined\",\n \"or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args() if args.output is not\",\n \"5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args = parser.parse_args() if\",\n \"= parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file already exists,\",\n \"exit(0) else: with open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser\",\n \"open(out, 'w') as f: f.write(json.dumps(result['response'])) if __name__ == '__main__': parser = argparse.ArgumentParser(description='Download latest\",\n \"parser.parse_args() if args.output is not None and os.path.exists(args.output): print('Output file already exists, abord.')\",\n \"(for example 5d or 12h or 30m).\\\") parser.add_argument(\\\"-o\\\", \\\"--output\\\", help=\\\"Output file\\\") args =\"\n]","isConversational":false}}},{"rowIdx":2698,"cells":{"ngram":{"kind":"list like","value":["+ bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'),","as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as","import include, url from django.views.generic.base import TemplateView from . import views as core_views","urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns","import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [","url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly Extensions","from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns","import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import",":: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/',","+ special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/',","urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns","urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly","BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)),","include, url from django.views.generic.base import TemplateView from . import views as core_views from","as page_urls from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as","import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import","Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/',","url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls","urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns","include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)),","as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as",".widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls","as core_views from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as","url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls","[ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/',","url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), #","urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns","include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN ::","as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as","core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/',","shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls","from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls #",".widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls","END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)),",".order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from .product.urls","import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import","from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from","import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import","include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly Extensions ------------------------------------------","social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)),",".page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from .search.urls","BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls",".staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN",".taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls","include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)),","url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/',","import views as core_views from .category.urls import urlpatterns as category_urls from .collection.urls import","include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), #","TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)),","urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly","urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns",".brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls","from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from","import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import",".widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls","include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls +","+ footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)),","from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from",".search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls","urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns","urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns","+ social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/',","import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import","include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls +",".customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls","import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import",".widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls","import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import","---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)),","site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls","store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls","url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly Extensions ------------------------------------------ ]","page_urls from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls","from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from","as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as","url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/',","as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as","as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as","name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ----------------------------------------","as order_urls from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as","as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as","urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns","special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide,","url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN","as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as","from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from","urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns","category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls","social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls","footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/',","include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)),","url from django.views.generic.base import TemplateView from . import views as core_views from .category.urls",":: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/',","import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN ::","site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)),","from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls #","from django.conf.urls import include, url from django.views.generic.base import TemplateView from . import views","from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from","urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns","urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns","import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END ::","# Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)),","include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/',","collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls","customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls","scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls","url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)),","urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns","django.views.generic.base import TemplateView from . import views as core_views from .category.urls import urlpatterns","from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from","as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions","import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import","from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from","order_urls from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls","as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as","from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from","as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as","import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from .product.urls import","product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls",".shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls","taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls",".store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls",". import views as core_views from .category.urls import urlpatterns as category_urls from .collection.urls","---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)),",".category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls","urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns","include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END","from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from","from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from","banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls","url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions","import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import","urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/',","include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls +","import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import","urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns",".store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls","as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions","-------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls","Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/',","import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import","import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import","urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns","as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index,","from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from","slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls","import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import","= [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)),","Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/',","import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from .search.urls import","include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), #","bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/',","url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/',","search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls","as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as","import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import","benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls","Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as","from django.views.generic.base import TemplateView from . import views as core_views from .category.urls import","url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)),","# END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/',","as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as",".store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls","as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as","url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/',","from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from","as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as","footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'),","# BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/',","import TemplateView from . import views as core_views from .category.urls import urlpatterns as","discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls",".store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END","url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')),","urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns","from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from","urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$',","include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)),","bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ----------------------------------------------",".sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls","name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)),","as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as","url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls","# BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from","brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls",".menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls",".store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns =","include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls +","import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import","import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import","url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END ::","views as core_views from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns","import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import","SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)),","as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as","urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns","from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from","from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from","url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls","url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly","from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from","as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as","from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from","urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns",".collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls","from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from","include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly","include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include(","urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns","url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/',","SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)),","django.conf.urls import include, url from django.views.generic.base import TemplateView from . import views as","from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from","TemplateView from . import views as core_views from .category.urls import urlpatterns as category_urls","staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions --------------------------------------------",".discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls","menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls","core_views from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls","url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/',",":: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import","as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as","core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions","special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls","urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns",".product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls","from . import views as core_views from .category.urls import urlpatterns as category_urls from","SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns"],"string":"[\n \"+ bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'),\",\n \"as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as\",\n \"import include, url from django.views.generic.base import TemplateView from . import views as core_views\",\n \"urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns\",\n \"import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [\",\n \"url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly Extensions\",\n \"from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns\",\n \"import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import\",\n \":: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/',\",\n \"+ special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/',\",\n \"urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns\",\n \"urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly\",\n \"BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)),\",\n \"include, url from django.views.generic.base import TemplateView from . import views as core_views from\",\n \"as page_urls from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as\",\n \"import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import\",\n \"Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/',\",\n \"url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls\",\n \"urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns\",\n \"include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)),\",\n \"as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as\",\n \".widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls\",\n \"as core_views from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as\",\n \"url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls\",\n \"[ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/',\",\n \"url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), #\",\n \"urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns\",\n \"include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN ::\",\n \"as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as\",\n \"core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/',\",\n \"shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls\",\n \"from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls #\",\n \".widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls\",\n \"END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)),\",\n \".order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from .product.urls\",\n \"import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import\",\n \"from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from\",\n \"import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import\",\n \"include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly Extensions ------------------------------------------\",\n \"social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)),\",\n \".page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from .search.urls\",\n \"BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls\",\n \".staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN\",\n \".taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls\",\n \"include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)),\",\n \"url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/',\",\n \"import views as core_views from .category.urls import urlpatterns as category_urls from .collection.urls import\",\n \"include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), #\",\n \"TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)),\",\n \"urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly\",\n \"urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns\",\n \".brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls\",\n \"from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from\",\n \"import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import\",\n \".widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls\",\n \"include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls +\",\n \"+ footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)),\",\n \"from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from\",\n \".search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls\",\n \"urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns\",\n \"urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns\",\n \"+ social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/',\",\n \"import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import\",\n \"include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls +\",\n \".customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls\",\n \"import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import\",\n \".widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls\",\n \"import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import\",\n \"---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)),\",\n \"site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls\",\n \"store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls\",\n \"url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly Extensions ------------------------------------------ ]\",\n \"page_urls from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls\",\n \"from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from\",\n \"as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as\",\n \"url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/',\",\n \"as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as\",\n \"as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as\",\n \"name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ----------------------------------------\",\n \"as order_urls from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as\",\n \"as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as\",\n \"urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns\",\n \"special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide,\",\n \"url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN\",\n \"as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as\",\n \"from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from\",\n \"urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns\",\n \"category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls\",\n \"social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls\",\n \"footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/',\",\n \"include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)),\",\n \"url from django.views.generic.base import TemplateView from . import views as core_views from .category.urls\",\n \":: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/',\",\n \"import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN ::\",\n \"site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)),\",\n \"from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns as taxes_urls #\",\n \"from django.conf.urls import include, url from django.views.generic.base import TemplateView from . import views\",\n \"from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from\",\n \"urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns\",\n \"urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import urlpatterns\",\n \"import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END ::\",\n \"# Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)),\",\n \"include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/',\",\n \"collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls\",\n \"customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls\",\n \"scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls\",\n \"url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)),\",\n \"urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns\",\n \"django.views.generic.base import TemplateView from . import views as core_views from .category.urls import urlpatterns\",\n \"from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls from\",\n \"as staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions\",\n \"import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls import\",\n \"from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from\",\n \"order_urls from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls\",\n \"as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as\",\n \"from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from\",\n \"as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as\",\n \"import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from .product.urls import\",\n \"product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as shipping_urls\",\n \".shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls\",\n \"taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls\",\n \".store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls\",\n \". import views as core_views from .category.urls import urlpatterns as category_urls from .collection.urls\",\n \"---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)),\",\n \".category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls from .customer.urls\",\n \"urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from .product.urls import urlpatterns\",\n \"include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END\",\n \"from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from\",\n \"from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from\",\n \"banner_urls from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls\",\n \"url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), # Extensions\",\n \"import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import\",\n \"urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/',\",\n \"include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls +\",\n \"import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import\",\n \"urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns\",\n \".store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls\",\n \"as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions\",\n \"-------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls\",\n \"Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/',\",\n \"import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import\",\n \"import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import\",\n \"urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns\",\n \"as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index,\",\n \"from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls from\",\n \"slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns as scene_urls\",\n \"import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import\",\n \"= [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)),\",\n \"Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/',\",\n \"import urlpatterns as page_urls from .product.urls import urlpatterns as product_urls from .search.urls import\",\n \"include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls + footer_item_urls)), #\",\n \"bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/',\",\n \"url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/',\",\n \"search_urls from .shipping.urls import urlpatterns as shipping_urls from .sites.urls import urlpatterns as site_urls\",\n \"as shipping_urls from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as\",\n \"import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import\",\n \"benefit_urls from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls\",\n \"Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns as\",\n \"from django.views.generic.base import TemplateView from . import views as core_views from .category.urls import\",\n \"url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)),\",\n \"# END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/',\",\n \"as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as\",\n \".store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls\",\n \"as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns as\",\n \"url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/',\",\n \"from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from\",\n \"as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as\",\n \"footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'),\",\n \"# BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/',\",\n \"import TemplateView from . import views as core_views from .category.urls import urlpatterns as\",\n \"discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls\",\n \".store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END\",\n \"url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')),\",\n \"urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns\",\n \"from .collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from\",\n \"urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$',\",\n \"include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)),\",\n \"bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ----------------------------------------------\",\n \".sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from .taxes.urls\",\n \"name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)),\",\n \"as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as\",\n \"url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls\",\n \"# BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from\",\n \"brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls\",\n \".menu.urls import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls\",\n \".store.footer_item.urls import urlpatterns as footer_item_urls # END :: SoftButterfly Extensions ---------------------------------------------- urlpatterns =\",\n \"include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls + bank_account_urls +\",\n \"import urlpatterns as customer_urls from .discount.urls import urlpatterns as discount_urls from .menu.urls import\",\n \"import urlpatterns as menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import\",\n \"url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END ::\",\n \"views as core_views from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns\",\n \"import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from .brand.urls import\",\n \"SoftButterfly Extensions ---------------------------------------------- urlpatterns = [ url(r'^$', core_views.index, name='index'), url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)),\",\n \"as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as\",\n \"urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls import urlpatterns\",\n \"from .product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from\",\n \"from .store.physical_store.urls import urlpatterns as store_urls from .store.social_network.urls import urlpatterns as social_network_urls from\",\n \"url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include( site_urls + social_network_urls + special_page_urls\",\n \"url(r'^style-guide/', core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly\",\n \"from .widget.scene.urls import urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from\",\n \"as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as\",\n \"from .sites.urls import urlpatterns as site_urls from .staff.urls import urlpatterns as staff_urls from\",\n \"urlpatterns as special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns\",\n \".collection.urls import urlpatterns as collection_urls from .customer.urls import urlpatterns as customer_urls from .discount.urls\",\n \"from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions -------------------------------------------- from\",\n \"include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)), url(r'^benefit/', include(benefit_urls)), # END :: SoftButterfly\",\n \"include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/', include(staff_urls)), url(r'^discounts/', include(discount_urls)), url(r'^settings/', include(\",\n \"urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as banner_urls from .widget.scene.urls import urlpatterns\",\n \"url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions ---------------------------------------- url(r'^brand/',\",\n \"SoftButterfly Extensions ---------------------------------------- url(r'^brand/', include(brand_urls)), url(r'^slider/', include(slider_urls)), url(r'^banner/', include(banner_urls)), url(r'^scene/', include(scene_urls)), url(r'^store/', include(store_urls)),\",\n \"django.conf.urls import include, url from django.views.generic.base import TemplateView from . import views as\",\n \"from .store.social_network.urls import urlpatterns as social_network_urls from .store.special_page.urls import urlpatterns as special_page_urls from\",\n \"TemplateView from . import views as core_views from .category.urls import urlpatterns as category_urls\",\n \"staff_urls from .taxes.urls import urlpatterns as taxes_urls # BEGIN :: SoftButterfly Extensions --------------------------------------------\",\n \".discount.urls import urlpatterns as discount_urls from .menu.urls import urlpatterns as menu_urls from .order.urls\",\n \"menu_urls from .order.urls import urlpatterns as order_urls from .page.urls import urlpatterns as page_urls\",\n \"core_views from .category.urls import urlpatterns as category_urls from .collection.urls import urlpatterns as collection_urls\",\n \"url(r'^categories/', include(category_urls)), url(r'^collections/', include(collection_urls)), url(r'^orders/', include(order_urls)), url(r'^page/', include(page_urls)), url(r'^products/', include(product_urls)), url(r'^customers/', include(customer_urls)), url(r'^staff/',\",\n \":: SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import\",\n \"as brand_urls from .widget.slider.urls import urlpatterns as slider_urls from .widget.banner.urls import urlpatterns as\",\n \"core_views.styleguide, name='styleguide'), url(r'^search/', include(search_urls)), url(r'^taxes/', include(taxes_urls)), url(r'^next/', TemplateView.as_view(template_name='dashboard/next.html')), # BEGIN :: SoftButterfly Extensions\",\n \"special_page_urls from .store.bank_account.urls import urlpatterns as bank_account_urls from .store.footer_item.urls import urlpatterns as footer_item_urls\",\n \"urlpatterns as scene_urls from .widget.benefit.urls import urlpatterns as benefit_urls from .store.physical_store.urls import urlpatterns\",\n \".product.urls import urlpatterns as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls\",\n \"from . import views as core_views from .category.urls import urlpatterns as category_urls from\",\n \"SoftButterfly Extensions -------------------------------------------- from .brand.urls import urlpatterns as brand_urls from .widget.slider.urls import urlpatterns\"\n]","isConversational":false}}},{"rowIdx":2699,"cells":{"ngram":{"kind":"list like","value":["FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path","trainer, frn_train from datasets import dataloaders from models.FRN import FRN args = trainer.train_parser()","os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot,","from datasets import dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml',","yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way","fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot]","from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp","= [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot,","sys import torch import yaml from functools import partial sys.path.append('../../../../') from trainers import","train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func","import trainer, frn_train from datasets import dataloaders from models.FRN import FRN args =","import yaml from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from","as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm =","= os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots =","shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm =","import torch import yaml from functools import partial sys.path.append('../../../../') from trainers import trainer,","import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders from","= os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader","transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm = trainer.Train_Manager(args,path_manager=pm,train_func=train_func)","dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f:","datasets import dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r')","with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path =","'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm","partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders from models.FRN","args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet)","models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp =","trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path","open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw')","args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path =","import sys import torch import yaml from functools import partial sys.path.append('../../../../') from trainers","import os import sys import torch import yaml from functools import partial sys.path.append('../../../../')","from trainers import trainer, frn_train from datasets import dataloaders from models.FRN import FRN","way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm","model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm = trainer.Train_Manager(args,path_manager=pm,train_func=train_func) tm.train(model)","torch import yaml from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train","os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader =","= trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way,","import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f)","import dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as","temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way","= yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way =","= FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm = trainer.Train_Manager(args,path_manager=pm,train_func=train_func) tm.train(model) tm.evaluate(model)","shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way,","f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args)","[args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot],","train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type)","= trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path'])","trainers import trainer, frn_train from datasets import dataloaders from models.FRN import FRN args","functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders","sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders from models.FRN import","data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots","from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import","os import sys import torch import yaml from functools import partial sys.path.append('../../../../') from","dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader)","trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots,","pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train,","= args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model","args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model =","= dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func =","yaml from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets","frn_train from datasets import dataloaders from models.FRN import FRN args = trainer.train_parser() with"],"string":"[\n \"FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path\",\n \"trainer, frn_train from datasets import dataloaders from models.FRN import FRN args = trainer.train_parser()\",\n \"os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot,\",\n \"from datasets import dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml',\",\n \"yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way\",\n \"fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot]\",\n \"from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp\",\n \"= [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot,\",\n \"sys import torch import yaml from functools import partial sys.path.append('../../../../') from trainers import\",\n \"train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func\",\n \"import trainer, frn_train from datasets import dataloaders from models.FRN import FRN args =\",\n \"import yaml from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from\",\n \"as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm =\",\n \"= os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots =\",\n \"shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm =\",\n \"import torch import yaml from functools import partial sys.path.append('../../../../') from trainers import trainer,\",\n \"import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders from\",\n \"= os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader\",\n \"transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm = trainer.Train_Manager(args,path_manager=pm,train_func=train_func)\",\n \"dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f:\",\n \"datasets import dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r')\",\n \"with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path =\",\n \"'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm\",\n \"partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders from models.FRN\",\n \"args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet)\",\n \"models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp =\",\n \"trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path\",\n \"open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw')\",\n \"args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path =\",\n \"import sys import torch import yaml from functools import partial sys.path.append('../../../../') from trainers\",\n \"import os import sys import torch import yaml from functools import partial sys.path.append('../../../../')\",\n \"from trainers import trainer, frn_train from datasets import dataloaders from models.FRN import FRN\",\n \"way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm\",\n \"model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm = trainer.Train_Manager(args,path_manager=pm,train_func=train_func) tm.train(model)\",\n \"torch import yaml from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train\",\n \"os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader =\",\n \"= trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way,\",\n \"import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f)\",\n \"import dataloaders from models.FRN import FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as\",\n \"temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way\",\n \"= yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way =\",\n \"= FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader) tm = trainer.Train_Manager(args,path_manager=pm,train_func=train_func) tm.train(model) tm.evaluate(model)\",\n \"shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way,\",\n \"f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args)\",\n \"[args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot],\",\n \"train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type)\",\n \"= trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path = os.path.abspath(temp['data_path'])\",\n \"trainers import trainer, frn_train from datasets import dataloaders from models.FRN import FRN args\",\n \"functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders\",\n \"sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import dataloaders from models.FRN import\",\n \"data_path = os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fewshot_raw') pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots\",\n \"from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets import\",\n \"os import sys import torch import yaml from functools import partial sys.path.append('../../../../') from\",\n \"dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func = partial(frn_train.default_train,train_loader=train_loader)\",\n \"trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots,\",\n \"pm = trainer.Path_Manager(fewshot_path=fewshot_path,args=args) train_way = args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train,\",\n \"= args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model\",\n \"args.train_way shots = [args.train_shot, args.train_query_shot] train_loader = dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model =\",\n \"= dataloaders.meta_train_dataloader(data_path=pm.train, way=train_way, shots=shots, transform_type=args.train_transform_type) model = FRN(way=train_way, shots=[args.train_shot, args.train_query_shot], resnet=args.resnet) train_func =\",\n \"yaml from functools import partial sys.path.append('../../../../') from trainers import trainer, frn_train from datasets\",\n \"frn_train from datasets import dataloaders from models.FRN import FRN args = trainer.train_parser() with\"\n]","isConversational":false}}}],"truncated":false,"partial":false},"paginationData":{"pageIndex":26,"numItemsPerPage":100,"numTotalItems":12866649,"offset":2600,"length":100}},"jwt":"eyJhbGciOiJFZERTQSIsImtpZCI6Im1sR1pFYktIVHRVSGRhX1RRdEczQ0N1S3k5ME9Qd25DNHY5elRjM3FVYzgifQ.eyJyZWFkIjp0cnVlLCJwZXJtaXNzaW9ucyI6eyJyZXBvLmNvbnRlbnQucmVhZCI6dHJ1ZX0sImlhdCI6MTc4MzQ5NzQyNSwianRpIjoiZWJiYjAzNjktODY1OC00NmFhLWIxMDEtNDYzYmM3YjNiODQ0Iiwic3ViIjoiL2RhdGFzZXRzL3RlcnJ5eXovc3RhcmNvZGVyZGF0YV9uZ3JhbV8xMyIsImV4cCI6MTc4MzUwMTAyNSwiaXNzIjoiaHR0cHM6Ly9odWdnaW5nZmFjZS5jbyJ9.zMWYp_WdBTeXdssGeu9N07wfQUPTlUIwKMR31OhGL8aqpY2-2aelO-MvYMAp5sz0E8cd7WKLkgUFL-FB11f4DQ","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 |
|---|
[
"raise TypeError(\"Expected Tensor, but received type of x: {}\".format( type(x))) dtype = x.dtype",
"paddle.static import Variable from ..fluid.framework import _in_legacy_dygraph, in_dygraph_mode __all__ = [] def _complex_to_real_dtype(dtype):",
"x (Tensor): The input tensor. Returns: bool: True if the data t... |
[
"a 'date' key\"\"\" def __init__(self, key, message): self.key = key self.message = message",
"def __str__(self): return str(self.message) class NonObjectReleaseError(OCDSMergeError, TypeError): \"\"\"Raised when a release is not",
"class NullDateValueError(OCDSMergeError, TypeError): \"\"\"Raised when a releas... |
[
"is list: self.config[element[1]] = element[0][0] elif (\"SYN\" in str(element[0])) or (\"FF\" in str(element[0])):",
"in self.capaRAW[1]: self.state[self.config[code]] = 0 for code in self.capaRAW[3]: self.state[self.config[code]] = 0 print(\"waiting",
"buttons.append(int(i)) for i in holder2: buttons.append(i... |
[
"2000) sensor.set_auto_gain(False) # must turn this off to prevent image washout... sensor.set_auto_whitebal(False) #",
"Camera. import sensor, image, time, math sensor.reset() sensor.set_pixformat(sensor.GRAYSCALE) sensor.set_framesize(sensor.VGA) # High Res! sensor.set_windowing((640,",
"can run at the full 6... |
[
"<reponame>devaslooper/Code-Overflow n=int(input(\"Enter number \")) fact=1 for i in range(1,n+1): fact=fact*i print(\"Factorial is \",fact)"
] |
[
"<reponame>thiagofreitascarneiro/Curso-de-Python---Curso-em-Video import time # O * é para desempacotar o paramêtro. Permite atribuir",
"para desempacotar o paramêtro. Permite atribuir inumeros parametros. def maior(* num): contador =",
"in num: contador = contador + 1 print(f'{v} ', end='', flush=True) time.sl... |
[
"be 3-dimensional.\"): values = torch.rand((100,), device=device) first_idxs = torch.tensor([0, 80], dtype=torch.int64, device=device) padded_to_packed(values,",
"True values_packed = padded_to_packed( values, mesh_to_faces_packed_first_idx, num_faces_per_mesh.sum().item() ) values_packed_torch = TestPackedToPadd... |
[
"roi1] roi_wrong_1 = (123, 345) roi_wrong_2 = [123, 456] roi_wrong_3 = [[123, 345],",
"geotiff._prase_header_string(in_str) assert out_dict['tie_point'] == (368090.77975000005, 3956071.13823) def test_prase_header_string_nodata(): out_dict = geotiff._prase_header_string(\"* 42113 gdal_nodata",
"[np.asarray([97.... |
[
"\"\"\"Gets the address_line2 of this ItemLocation. # noqa: E501 The second line of",
"title=\\\"Limited Release\\\" alt=\\\"Limited Release\\\" />(Limited Release)</a> The Marketplace Insights API provides the ability",
"@address_line2.setter def address_line2(self, address_line2): \"\"\"Sets the address_line2... |
[
"100 data.sort(key=lambda x: x[1]/x[2], reverse=True) profit=0 ans=[] i=0 while i<n: if data[i][2]<=bag: bag-=data[i][2]",
"Knapsack wt = [40,50,30,10,10,40,30] pro = [30,20,20,25,5,35,15] n = len(wt) data = [",
"= 100 data.sort(key=lambda x: x[1]/x[2], reverse=True) profit=0 ans=[] i=0 while i<n: if data[i][2]... |
[
"# # Licensed under the Apache License, Version 2.0 (the \"License\"); # you",
"writing, software # distributed under the License is distributed on an \"AS IS\"",
"#!/usr/bin/env python # -*- python -*- #BEGIN_LEGAL # #Copyright (c) 2019 Intel Corporation",
"One can look for # those with SSE2MMX and SSSE3MMX ... |
[
"details.id == target_group.id, u'Expected the `details` response to match the requested target group.'",
"return type `TargetGroup`.' assert details.id == target_group.id, u'Expected the `details` response to match",
"not return type `TargetGroup`.' assert edited_group.id == target_group.id, u'Expected the edi... |
[
"``True``, the time index is the center of the STFT kernel. Default value",
"in the shape of (batch, freq_bins, timesteps)\" # Initializing Random Phase rand_phase =",
"to ``n_fft//4``. window : str The windowing function for STFT. It uses ``scipy.signal.get_window``,",
"be controlled by ``fmin`` and ``fmax``... |
[
"return val def test_dataloader(self): test = DataLoader(self.test, batch_size=self.batch_size, num_workers=self.num_workers, shuffle=False) return test class",
"test = DataLoader(self.test, batch_size=self.batch_size, num_workers=self.num_workers, shuffle=False) return test class Base(pl.LightningModule): def __... |
[
"== LAST_RESET_UPTIME and self.attribute_value is not None ): value = cast(float, self.attribute_value) if",
"\"rollerEnergy\"): BlockAttributeDescription( name=\"Energy\", unit=ENERGY_KILO_WATT_HOUR, value=lambda value: round(value / 60 / 1000, 2), device_class=sensor.DEVICE_CLASS_ENERGY,",
"), (\"sensor\", \"... |
[
"smaller size for testing QUERY_KWARGS['GET']['facet_size']['default'] = 3 QUERY_KWARGS['GET']['facet_size']['max'] = 5 QUERY_KWARGS['POST']['q']['jsoninput'] = True",
"# User Input Control # ***************************************************************************** # use a smaller size for testing",
"Query ... |
[
"except FileNotFoundError: logger.error(f\"ERROR: Image file '{img}' is missing\") except UnidentifiedImageError: logger.error(f\"ERROR: Image file",
"part in Part.objects.exclude(image=None): try: self.rebuild_thumbnail(part) except (OperationalError, ProgrammingError): logger.error(\"ERROR: Database read error.... |
[
"twice def hex_to_rgb(hex: str) -> tuple: \"\"\" Args: hex (str): \"\"\" return tuple(int(hex.lstrip(\"#\")[i",
"= requests.get(carbon_url) b = BytesIO(r.content) await ctx.send(file=discord.File(fp=b, filename=\"code.png\")) async def setup(bot): await bot.add_cog(Carbon(bot))",
"themes, controls, languages, f... |
[
"or URL import spotipy_twisted import sys import pprint if len(sys.argv) > 1: urn",
"import pprint if len(sys.argv) > 1: urn = sys.argv[1] else: urn = 'spotify:artist:3jOstUTkEu2JkjvRdBA5Gu'",
"1: urn = sys.argv[1] else: urn = 'spotify:artist:3jOstUTkEu2JkjvRdBA5Gu' sp = spotipy_twisted.Spotify() artist =",
"... |
[
"<filename>examples/mcp3xxx_mcp3002_single_ended_simpletest.py import busio import digitalio import board import adafruit_mcp3xxx.mcp3002 as MCP from adafruit_mcp3xxx.analog_in",
"AnalogIn(mcp, MCP.P0) print(\"Raw ADC Value: \", chan.value) print(\"ADC Voltage: \" + str(chan.voltage) +",
"import adafruit_mcp3xx... |
[
"' 'csv.gz txt.gz tbl.bz ' 'dat.gz'), priority=1) def tabular_data(path, **kwargs): from glue.core.data_factories.astropy_table import",
"astropy_tabular_data from glue.core.data_factories.pandas import pandas_read_table for fac in [astropy_tabular_data, pandas_read_table]: try: return fac(path,",
"'dat.gz'), p... |
[
"= 10 template_name = \"code_doc/authors/author_list.html\" context_object_name = \"authors\" model = Author def detail_author(request,",
"from django.http import Http404 from django.views.generic.edit import UpdateView from django.views.generic import ListView, View",
"ListView, View from django.contrib.auth.d... |
[
"soc_sloc += driver_sloc #print(\"{}: {}\".format(cmp_str, driver_sloc)) if arch not in soc_sloc_by_arch: soc_sloc_by_arch[arch] =",
"driver_sloc = avg_sloc_by_arch[arch] soc_sloc += driver_sloc #print(\"{}: {}\".format(cmp_str, driver_sloc)) if arch not in",
"DRIVER_NAME_TO_SLOC else: print(\"Error: no SLOC fi... |
[
"y = batch if self.lr_decay: self.tokens += (y >= 0).sum() # number of",
"trainer.optimizers[0] _, y = batch if self.lr_decay: self.tokens += (y >= 0).sum() #",
"max(0.1, 0.5 * (1.0 + math.cos(math.pi * progress))) lr = self.learning_rate * lr_mult",
"self.final_tokens = final_tokens self.lr_decay = lr_decay ... |
[
"if any were set results['tag'] = set(parse_list(result.group('tags'))) # Prepare our Asset Object results['asset']",
"in this configuration file to our list self._cached_servers.extend( cfg_plugin.servers(asset=asset)) # We no longer",
"files marked to be in STRICT mode are treated as being in ALWAYS",
"{}.'... |
[
"startupinfo def popen_ffmpeg(inner_args): cmd = [ 'ffmpeg', *inner_args, # logging '-loglevel', ffmpeg_loglevel, '-hide_banner',",
"'win32' in str(sys.platform).lower() SUBPROCESS_ARGS = {} if IS_WIN32: startupinfo = subprocess.STARTUPINFO() startupinfo.dwFlags =",
"AV_LOG_DEBUG = \"debug\" ffmpeg_loglevel = A... |
[] |
[
"if x_pos < min_x: continue if x_pos > max_x: break b_above_x, b_below_x =",
"= min(z_min, np.min(veh_surf[\"z\"])), max(z_max, np.max(veh_surf[\"z\"])) x_min, x_max = x_min - settings[\"voxel_size\"], x_max +",
"veh_surf['nodes'][:, 1] veh_surf[\"z\"] = veh_surf['nodes'][:, 2] return veh_surf def find_occupied... |
[
"''' # CODE ======================================================================== import zlib import random from hashlib import sha1 from",
"counter in range(len(pad)): operated=int(pad[counter])^locker[counter] holder.append(operated) return holder def encrypt(user_input,passphrase): compressed=zlib.compress(... |
[
"pyfire.errors ~~~~~~~~~~~~~~~~~~~~~~ Holds the global used base errors :copyright: 2011 by the pyfire",
":license: BSD, see LICENSE for more details. \"\"\" import xml.etree.ElementTree as ET class",
"via XMPP Protocol to peer \"\"\" def __init__(self, error_element, error_namespace, error_name=None): self.err... |
[
"my_position[1]['x'] and my_position[0]['y'] < my_position[1]['y']: return 'up' else: return 'down' def getHealth(gamedata): pass",
"'none': choice = random.choice(options) #remove opposite direction if current_direction == 'up': options.remove('down') if",
"Status.getMyDirection(gamedata) #first go if current_... |
[
"total < 30 or total >= 160 else: return total < 160 or",
"* 40 if master_only or mates: if bullet: return total < 30 or",
"= nps[-20:] # print(info) best = EngineMove(info[0][\"pv\"][0], info[0][\"score\"].pov(winner)) second = EngineMove(info[1][\"pv\"][0], info[1][\"score\"].pov(winner)) if",
"limit: chess... |
[
"= datetime.now() if n.hour == 6 and (n.minute-(n.minute%5)) == 15: return morning.main() time.sleep(300",
"if n.hour == 6 and (n.minute-(n.minute%5)) == 15: return morning.main() time.sleep(300 - (time.mktime(datetime.now().timetuple())-a))",
"a = time.mktime(datetime.now().timetuple()) n = datetime.now() if n... |
[
"samcli.commands.validate.lib.exceptions import InvalidSamDocumentException from samcli.commands.local.lib.exceptions import OverridesNotWellDefinedError class TestCli(TestCase): def setUp(self): self.template =",
"as context: self.call_cli() msg = str(context.exception) expected = \"bad env vars\" self.assertEqu... |
[
"\"\"\"Whether it returns correct df when input is correct\"\"\" # Returns correct df",
"True) self.assertEquals(Getter(table_name=IndicatorType, df_flag=True, param_list={\"name\": \"abcd\"}).get_data() .equals( Converter(obj_list=list(IndicatorType.objects.filter(**{\"name\": \"abcd\"}))).to_df() ), True) def t... |
[
"governing permissions and limitations under the License. \"\"\" import json import logging import",
"'SSDHead': return False if not cfg.model.bbox_head.anchor_generator.type == 'SSDAnchorGeneratorClustered': return False return True @staticmethod",
"in widths: n_clust += len(w) if isinstance(w, (list, tuple)) ... |
[
"set() files.update(complexity.keys()) files.update(duplicate_const_strings.keys()) result = [] for f in files: result.append({ 'filename': f,",
"in which to keep track of the complixities # for each file files",
"= {} # Build an array containing the number of potentially duplicated #",
"array of complexities... |
[
"'SyntaxError: invalid syntax' return [wanted], line_nr def test_default_except_error_postition(): # For this error the",
"+', [(1, 3), (2, 3)]), ('x + 2', []), ('[\\n', [(2, 0)]), ('[\\ndef",
"that are oddly enough not SyntaxErrors. wanted = 'SyntaxError: (value error) ' +",
"assert False, \"The piece of cod... |
[
"src, None, modifier=modifier) step += 1 elif line[0].isupper(): dst = Coords(int(line[2]), RANKNUM[line[3]]) yield",
"1 elif line[0].isupper(): dst = Coords(int(line[2]), RANKNUM[line[3]]) yield Move(color[step & 1], dst, None,",
"= PROMOTE else: modifier = None yield Move(color[step & 1], dst, src, None,",
... |
[
"tx_hash = utils.send_transaction(func, 0, private_key) return tx_hash @main.command() @click.option( '--private-key', callback=utils.check_account, help='The privat",
"oracles to work with Ether dollar\" pass @main.command() @click.option('--ether-price', type=float, help=\"The ether price",
"Ether dollar\" pa... |
[
"return local def with_machine_rule(cls): old_init = cls.__init__ def new_init(self, config): if \"machine\" not",
"if machine_type == \"local\": self.machine = get_local_machine() self.files_to_copy = None elif machine_type ==",
"ParamikoMachine(host, user=user, password=password) return rem @_once def get_rem... |
[
"<reponame>legna7/Python<gh_stars>0 salario = float(input('digite o seu salario: ')) aumento = (salario + (salario",
"o seu salario: ')) aumento = (salario + (salario * 15)/100 if salario",
"seu salario: ')) aumento = (salario + (salario * 15)/100 if salario <=",
"aumento = (salario + (salario * 15)/100 if sa... |
[
"self.assertTrue(len(t) > 0) self.assertTrue('git' not in (x['id'] for x in t)) t =",
"x.get('id') == '1']) == 1) def test_git(self): with app.app_context(): t = get_tree(\"git/HEAD\") self.assertTrue(t",
"def test_tree(self): with app.app_context(): t = get_tree() self.assertTrue(len(t) > 0) self.assertTrue('g... |
[
"login endpoint. Will return a fresh access token and # a refresh token",
"not actually verify a password in this endpoint. @app.route(\"/refresh\", methods=[\"POST\"]) @jwt_refresh_token_required async def",
"# As we just verified their username and password, we are # going",
"login(): username = (await requ... |
[
"__all__ = 'AnyDict' AnyDict = Dict[str, Any] # pragma: no mutate datetime_or_str =",
"datetime from typing import Any, Dict, Union __all__ = 'AnyDict' AnyDict = Dict[str,",
"Union __all__ = 'AnyDict' AnyDict = Dict[str, Any] # pragma: no mutate datetime_or_str",
"import datetime from typing import Any, Dict,... |
[
"from keras.models import save_model, load_model from sklearn.preprocessing import StandardScaler, MinMaxScaler, Normalizer if __name__",
"keras.layers import Input, Dense from keras.models import Model from keras.models import save_model, load_model",
"model.compile(optimizer=sgd ,loss='mse') model.fit(x_,y_, ... |
[
"CheerLightsEffect(canvas) # Must check before and after in case it changes during the",
"effect = CheerLightsEffect(canvas) # Must check before and after in case it changes",
"# Mock the hardware layer to avoid errors. from ledshimdemo.canvas import Canvas from",
"canvas = Canvas(self.TEST_CANVAS_SIZE) effec... |
[
"np.zeros((num_images, *shape)) out_dict[k][i] = v return out_dict def generate_image(cells, shape, max_dist=5): thetas =",
"out_dict = {name: np.zeros(shape) for name, dclass in zip(data_list[0].names, data_list[0].dclasses) if dclass !=",
"num_images)).astype(int) nums = nums[nums > 0] assert sum(nums) < len(... |
[
"self._value = settings[\"value\"] @property def id(self): return self._id @property def value(self): return self._value",
"<filename>epiphancloud/models/settings.py class DeviceSettings: def __init__(self, settings): self._id = settings[\"id\"] self._title = settings[\"title\"] self._type",
"= settings[\"id\"]... |
[
"with one or two horns on its nose\") else: print(\"That word must not",
"on its nose\") else: print(\"That word must not exist. This dictionary is very",
"if word == \"a\": print(\"one; any\") elif word == \"apple\": print(\"familiar, round fleshy",
"\"a\": print(\"one; any\") elif word == \"apple\": print(\... |
[
"= list(map(int, input())) sum = 0 for i in range(len(num)): sum = sum",
"= int(input()) num = list(map(int, input())) sum = 0 for i in range(len(num)):",
"cnt = int(input()) num = list(map(int, input())) sum = 0 for i in",
"int(input()) num = list(map(int, input())) sum = 0 for i in range(len(num)): sum",
... |
[
"in a second keywords = ['testing', 'logging', 'example'], # arbitrary keywords classifiers =",
"above version = '0.1', description = 'Kubernetes resource discovery toolkit', author = '<NAME>',",
"author = '<NAME>', author_email = '<EMAIL>', url = 'https://github.com/sdnhub/kube-navi', # use the URL",
"url = ... |
[
"permissions and limitations # under the License. # import ast from typing import",
"distributed on an # \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY",
"if MetricTypeProto.DATASET == metric_meta_proto.metric_type: metric_type = MetricType.DATASET else: metric_type = MetricType.MODEL return MetricMet... |
[
"moduels.component.NormalValue import 常量 from moduels.component.SponsorDialog import SponsorDialog import os, webbrowser class Tab_Help(QWidget): 状态栏消息",
"self.tipButton.setMaximumHeight(100) def openHelpDocument(self): try: if 常量.系统平台 == 'Darwin': import shlex os.system(\"open \" +",
"Signal from moduels.compo... |
[
"except InvalidPasswordException as e: raise HTTPException( status_code=status.HTTP_400_BAD_REQUEST, detail={ \"code\": ErrorCode.REGISTER_INVALID_PASSWORD, \"reason\": e.reason, },",
"app.common.user import ErrorCode, run_handler from app.users.user import ( CreateUserProtocol, InvalidPasswordException, UserAlre... |
[
"HSV color model, instead of the saturation plane, as we do here. Args:",
"representation of the dense optical flow in RGB format Ref: https://github.com/philferriere/tfoptflow/blob/33e8a701e34c8ce061f17297d40619afbd459ade/tfoptflow/optflow.py \"\"\" hsv",
"0, 255, cv2.NORM_MINMAX) else: hsv[..., 1] = flow_magn... |
[
"1.66 ming = 10 ratr = 2 speed = 8 leafcount = 0",
"= 2 speed = 8 leafcount = 0 total = 0 for i",
"8 leafcount = 0 total = 0 for i in range(0, int(imWidth/speed)): for",
"= Image.open(args.fname) RGB = im.convert('RGB') imWidth, imHeight = im.size ratg = 1.2 ratgb",
"i in range(0, int(imWidth/speed)): for j... |
[
"endMsecs)) parseStats(cluster['name'], capacityStats['dataPointVec'][0], 'capacity') parseStats(cluster['name'], consumedStats['dataPointVec'][0], 'consumed') parseStats(cluster['name'], dataInStats['dataPointVec'][0], 'dataIn') parseStats(cluster['name'], dataWrittenStats['dataPointVec'][0], 'dataWritten')",
"=... |
[
"n_neighbors = 5; # better to have this at the start of the",
"# 10重 skf = StratifiedKFold(y,kfold) skfind = [None] * len(skf) cnt = 0",
"= ind[0] # print ind l = serial imgs = [] for rank",
"# roughly 2.5 secs # Testing y_predict = [] tic = time.time() y_predict",
"= train_index cnt += 1 list_fams = n cach... |
[
"\"XCLAIM\", \"key\": \"mystream\", \"group\": \"mygroup\", \"consumer\": \"Alice\", \"millisecond\": \"3600000\", \"stream_id\": \"1526569498055-0\", \"force\": \"FORCE\",",
"\"XTRIM\", \"key\": \"mystream\", \"maxlen\": \"MAXLEN\", \"count\": \"2\"}, ) judge_command( \" XTRIM mystream MAXLEN",
"\"XRANGE somes... |
[
"= fixture.read() response_mock = self.make_mock(json_response) self.make_test(response_mock) def test_determine_names(self): \"\"\"To run this test you'll",
"= determine_names('name-with-an-error') def test_find_forks(self): sent_args = { 'per_page': 99, 'start_page': 3 } url",
"'https://github.com/frost-nzcr4... |
[
"LOG.debug(\"gw_port_host missing from router: %s\", self.router['id']) return host def internal_network_added(self, port): super(DvrRouter, self).internal_network_added(port)",
"LOG.exception(_LE('DVR: removed snat failed')) def get_gw_port_host(self): host = self.router.get('gw_port_host') if not host: LOG.debu... |
[
"class PerformanceWarning(SiderWarning, RuntimeWarning): \"\"\"The category for warnings about performance worries. Operations that warn",
"this base class.\"\"\" class PerformanceWarning(SiderWarning, RuntimeWarning): \"\"\"The category for warnings about performance worries.",
"custom warning category classes... |
[
"($search: String) { User(search: $search) { id name html_about: about(asHtml: true) about avatar",
"avatar { large } bannerImage siteUrl stats { watchedTime chaptersRead } } }",
"} } \"\"\" MEDIA_BY_ID = \"\"\" query ($id: Int, $type: MediaType) { Media(id:",
"{ large } bannerImage siteUrl stats { watchedTim... |
[
"all_names: if name.endswith(suffix): suite.addTest(unittest.makeSuite(all_names[name], \"test\")) return suite if __name__ == '__main__': runner =",
"sandbox.chainop as chainop class BasicChainTest (unittest.TestCase): def testBasicChain(self): double = lambda x: x",
"= lambda x: x ** 2 self.assertEqual([2**31... |
[
"= \"data/local/trpo-pendulum\" np.random.seed(42) # Clean up existing logs os.system(\"rm -rf {}\".format(log_dir)) with logger.session(log_dir):",
"= GaussianMLPPolicy( observation_space=env.observation_space, action_space=env.action_space, env_spec=env.spec, hidden_sizes=(64, 64), hidden_nonlinearity=chainer.f... |
[
"root = self._root.find(tag) attrib = root.attrib if attrib.get('part', 'False').lower() == 'true': self._part_tag(root) return",
"= str_ self._root = et.parse(xml).getroot() self.re = '' self.data = [] def select(self,",
"class RegexEngine: def __init__(self, xml, str_): \"\"\"加载正则表。正则表为xml :param xml: 正则表的位置 ... |
[
"monitor=\"val_acc\")], \"verbose\": False, \"validation_split\": 0.33, }, A dictionary to hold epochs, callbacks, verbose,",
"default=\"categorical_crossentropy\" A loss function used when compiling the neural network. pretrained : bool,",
"more typechecking self.network.compile( loss=self.loss, optimizer=self... |
[
"cmap = plt.get_cmap(\"Oranges\") N = 12 width = (2 * np.pi) / N",
"Default colors equally spaced on [0,1]: if z is None: z = np.linspace(0.0,",
"segments = make_segments(x, y) lc = mcoll.LineCollection(segments, array=z, cmap=cmap, norm=norm, linewidth=linewidth, alpha=alpha, zorder=zorder)",
"norm=norm, lin... |
[
"LoginRequiredMixin, PermissionRequiredMixin): permission_required = 'accounts.view_account' raise_exception = True def post(self, request, pk, *args,",
"return redirect(\"home\") else: messages.warning(request, 'There is an issue with your login processes') return",
"login as auth_login from django.contrib.aut... |
[
"enc_param = [int(cv2.IMWRITE_JPEG_QUALITY), 40] _, im = await loop.run_in_executor(threadpool, lambda: cv2.imencode('.jpg', im, enc_param))",
"= await queue.get() im = np.frombuffer(im, dtype=np.uint8) im = await loop.run_in_executor(threadpool, lambda: cv2.imdecode(im,",
"except asyncio.CancelledError: await ... |
[
"typing import List from Giveme5W1H.extractor.root import path from Giveme5W1H.extractor.tools.util import bytes_2_human_readable class KeyValueCache(object):",
"+ str(value)) self.persist() self._lock.release() def get(self, key): \"\"\" Read cache entries :param key:",
"[value, str(time.time())] def _unpack(s... |
[
"len(word)] token_tgt = [nsst.tokenization_tgt[word] if word in nsst.tokenization_tgt else 0 for word in",
"= NSST.NSST() nsst.load(args.nsst_file) args.nsst = nsst # open files src_file = open(args.src_lang, 'r')",
"in args.nsst_file or k[0] == -1) ), key=lambda x: x[2], reverse=True)[0] n_res =",
"in token_... |
[
"Define functionts def mean(data): return sum(data) / len(data) def stddev(data, size): sum =",
"len(data) def stddev(data, size): sum = 0 for i in range(size): sum =",
"data size = int(input()) numbers = list(map(int, input().split())) # Get standard deviation print(round(stddev(numbers,",
"range(size): sum ... |
[
"<reponame>jmsevillam/Herramientas-Computacionales-UniAndes def decode(word1,word2,code): if len(word1)==1: code+=word1+word2 return code else: code+=word1[0]+word2[0] return decode(word1[1:],word2[1:],code) Alice='Ti",
"else: code+=word1[0]+word2[0] return decode(word1[1:],word2[1:],code) Alice='Ti rga eoe esg o... |
[
") from pychron.database.adapters.power_calibration_adapter import PowerCalibrationAdapter from pychron.paths import paths from pychron.graph.graph import Graph",
"KIND, either express or implied. # See the License for the specific language",
"self._get_calibration_path() self.info(\"saving calibration to {}\".... |
[
"fetch the list of candidate games from the game history # we need",
"to be submitted to maximin maximin_list.append(these_probs) #print(\"maximin_list:\", maximin_list) # send the list of",
"if candidate_games != False: # this is the list of games that match",
"for the next move recommended_move = maximin(ma... |
[
"interval, times): for cnt in range(times): time.sleep(interval) print(\" -- %s: %s\" % (thread_name,",
"2, 6)) except: print(\"Error: 无法启动线程\") # 主线程无限等待 while 1: pass @staticmethod def demo(args=[]):",
"6)) except: print(\"Error: 无法启动线程\") # 主线程无限等待 while 1: pass @staticmethod def demo(args=[]): RawThreadDemo... |
[
"= \"connection_density\" EVAL_NUM_EPOCHS = \"evaluate_every_number_of_epochs\" EVAL_NUM_EXAMPLES = \"evaluate_on_number_of_examples\" INTENT_CLASSIFICATION = \"intent_classification\" ENTITY_RECOGNITION =",
"= \"maximum_negative_similarity\" USE_MAX_NEG_SIM = \"use_maximum_negative_similarity\" SCALE_LOSS = \"sc... |
[
"self.adjust = adjust self.angle = None def calcAngle(self, distances): long_distance = distances[self.long_point_index] short_distance",
"self.points = {0: 0, 45: 0, 90: 0, 135: 0, 180: 0, 225:",
"lines[self.left_point_index] mline = lines[self.mid_point_index] rline = lines[self.right_point_index] dlong1 = di... |
[
"['EscalationPolicy'], 'spaceone.monitoring.interface.grpc.v1.event_rule': ['EventRule'], 'spaceone.monitoring.interface.grpc.v1.webhook': ['Webhook'], 'spaceone.monitoring.interface.grpc.v1.maintenance_window': ['MaintenanceWindow'], 'spaceone.monitoring.interface.grpc.v1.alert': ['Alert'], 'spaceone.monitoring.in... |
[
"toys = models.ManyToManyField(Toy, through='PlayedWith') class PlayedWith(models.Model): child = models.ForeignKey(Child) toy = models.ForeignKey(Toy) date",
"object_id = models.PositiveIntegerField() content_type = models.ForeignKey(ContentType) content_object = GenericForeignKey() class AwardNote(models.Model)... |
[
"= np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2) dst = cv2.perspectiveTransform(pts,M) img2 = cv2.polylines(img2,[np.int32(dst)],True,255,3, cv2.LINE_AA) else: print(\"Not",
"SIFT kp1, des1 = sift.detectAndCompute(img1,None) kp2, des2 = sift.detectAndCompute(img2,None) FLANN_INDEX_KDTREE = 0 ind... |
[
"relevant_data) / (ideal - nadir) distance = distance.max(axis=1) distance_order = np.argsort(distance) best =",
"12: \"12\", }, className=\"text-center mt-5\", ), ], className=\"mr-3 ml-3 mb-2 mt-2\", ), dbc.FormGroup(",
"preference\", \"value\": \"both\", }, ], id=\"os-choice\", value=\"both\", inline=True, #... |
[
"in class annotations by removing the need to write lambda functions. \"\"\" return",
"JSON data. Example: >>> @sprop.a # sprop annotations are needed to tell the",
"out. >>> class Baz(object): pass >>> def __init__(self, a, b, c): >>> self.a",
"sprops = pd.class_sprops.get(cls,{}) # [name] cprops = pd.class_... |
[
"to a resolution of {1} to create standard sized patches.\\n\".format(config[\"trainingset_patchsize\"], resolution_model), \"One of",
"path_model, overlap_value, config, resolution_model, acquired_resolution = None, verbosity_level=0): ''' Segment the image located",
"that resolution.\\n\".format(round(psm * m... |
[
"low d assert close_enough(approx_95CI_lower, bs_95CI_lower, count_error) assert close_enough(approx_95CI_upper, bs_95CI_upper, count_error) assert close_enough(h_testing.d, (m2",
"/ sig, h_testing.correction) print (\"lower: approx, bootstrap\", approx_95CI_lower, bs_95CI_lower) print (\"upper: approx, bootstrap... |
[
"path from . import views urlpatterns = [ path('', views.Records, name =\"fRec\"), ]",
"import path from . import views urlpatterns = [ path('', views.Records, name =\"fRec\"),",
"django.urls import path from . import views urlpatterns = [ path('', views.Records, name",
"from django.urls import path from . im... |
[
"make_tempdir from spacy import util from thinc.api import Config TRAIN_DATA = [ (\"I'm",
"@layers = \"reduce_mean.v1\" [components.transformer] factory = \"transformer\" \"\"\" # Xfail this until the",
"spacy.util import make_tempdir from spacy import util from thinc.api import Config TRAIN_DATA =",
"up. @py... |
[
"run(self): await asyncio.sleep(1) print(\"Insert command: \") action_creator = ActionCreator() while True: input_data =",
"\") action_creator = ActionCreator() while True: input_data = await input(\"~> \") if not",
"config, *args, **kwargs): self.action_queue = action_queue self.config = config async def run(s... |
[
"global_step, sample_rate=fs) except Exception as e: warn(str(e)) pass def save_states(global_step, writer, mel_outputs, linear_outputs,",
"hparams.masked_loss_weight > 0: # decoder output domain mask decoder_target_mask = sequence_mask( target_lengths /",
"torch.autograd import Variable from torch import nn fr... |
[
"screen.blit(write(\"Loading index... %fs\" % (time.time() - start)), (0,0)) pygame.display.flip() with open(os.path.join(imgdir, \"index.tsv\")) as",
"= pygame.font.Font(\"Futura.ttc\", font_size) return fonts[font_size] def write(text, colour = font_colour, font_size = 40):",
"(255, 0, 0), rect, 2) capture = ... |
[
"pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine in Django. \"\"\"",
"<reponame>plojyon/resolwe<gh_stars>10-100 \"\"\".. Ignore pydocstyle D400. ======= Resolwe ======= Open source enterprise dataflow engine",
"from resolwe.__about__ import ( # noqa: F401 __author__, __copyright_... |
[
"sample dates # %% df_tmp = pd.DataFrame(df['file_name'].resample('1D').count()) df_tmp['count'] = df_tmp['file_name'] del df_tmp['file_name'] df_tmp['rac']",
"x=df_tmp['count'], hue=df_tmp['rac'], orient='h') plt.draw() # %% df_melt = pd.melt(df, value_vars=['rac'], value_name='ractopamine') plt.figure(figsize=(... |
[
"code. \"\"\" # return Competition return Competition def update_competition(self, competition): \"\"\" Retrieve data",
"the most recent fixtures and results for the supplied competition code. \"\"\" return",
"results. Returns ------- Competition A Competition object with the most recent fixtures and",
"the s... |
[
"Exception(\"No files found to convert\") subprocess.call(\"mkdir -p %s\" % HTML_DIR, shell=True) pool.map(create_rst, candidate_files)",
"Pool(processes=4) candidate_files = [f for f in EXAMPLE_GLOB] if not candidate_files: raise Exception(\"No",
"you see once you click into a grid cell on the pydeck gallery",... |
[
"rotations # inertial frame: located at the origin (waist), but aligned with the",
"[None] for i in range(0,3): if vector[i] == 0 or vector[i] == 1:",
"= nb_bodies*[None] # waist Dg[0] = sp.Matrix([sp.Symbol('L_1_6'), sp.Symbol('L_2_6'), sp.Symbol('L_3_6')]) # right leg Dg[1]",
"{}'.format(get_string_enum(joi... |
[
"required=True, help=\"can be defined in days, hours, minutes (for example 5d or 12h",
"minutes (for example 5d or 12h or 30m).\") parser.add_argument(\"-o\", \"--output\", help=\"Output file\") args",
"misp_key, misp_verifycert import argparse import os import json # Usage for pipe masters:",
"'w') as f: f.w... |
[
"+ bank_account_urls + footer_item_urls)), # Extensions url(r'^menu/', include(menu_urls)), url(r'^shipping/', include(shipping_urls)), url(r'^style-guide/', core_views.styleguide, name='styleguide'),",
"as product_urls from .search.urls import urlpatterns as search_urls from .shipping.urls import urlpatterns as"... |
[
"FRN args = trainer.train_parser() with open('../../../../config.yml', 'r') as f: temp = yaml.safe_load(f) data_path",
"trainer, frn_train from datasets import dataloaders from models.FRN import FRN args = trainer.train_parser()",
"os.path.abspath(temp['data_path']) fewshot_path = os.path.join(data_path,'CUB_fe... |