Spaces:
Sleeping
Sleeping
File size: 62,710 Bytes
f0cf803 870b6c3 f0cf803 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 | #!/usr/bin/env python
# coding: utf-8
from IPython.core.debugger import set_trace
import random
import matplotlib.pyplot as plt
import numpy as np
import math
import copy
import sys
from scipy.stats import pearsonr # for computing correlation
from functools import reduce #for flattening lists
from operator import concat #for flattening lists
from scipy.stats import trim_mean # for ensemble evaluation
from scipy.stats import differential_entropy
import warnings
import time
import dill
import os
from sklearn.cluster import KMeans #for clustering in ensemble definition
from scipy.optimize import minimize #for uncertainty maximization
from sympy import symbols, simplify, expand
import sympy as sym
try:
from IPython.display import display, clear_output
except:
pass
import signal #for timing out functions
from contextlib import contextmanager #for timing out functions
warnings.filterwarnings('ignore', '.*invalid value.*' )
warnings.filterwarnings('ignore', '.*overflow.*' )
warnings.filterwarnings('ignore', '.*divide by.*' )
warnings.filterwarnings('ignore', '.*is constant.*' )
warnings.filterwarnings('ignore', '.*nearly constant.*' )
warnings.filterwarnings('ignore', '.*Polyfit may be.*' )
warnings.filterwarnings('ignore', '.*Number of.*')
## Get the arrays for the models
model_distributions = []
def protectDiv(a,b):
if (type(b)==int or type(b)==float or type(b)==np.float64) and b==0:
return a/math.nan
if (type(b)==np.ndarray) and (0 in b):
return a/np.where(b==0,math.nan,b)
return a/b
def add(a,b):
return a+b
def sub(a,b):
return a-b
def mult(a,b):
return a*b
def exp(a):
safe_a = np.clip(a, -90.0, 90.0)
return np.exp(safe_a)
# def sine(a,b):
# return np.sin(a)
def power(a,b):
if (type(a)==int or type(a)==float or type(a)==np.float64) and a==0:
return a/math.nan
if (type(a)==np.ndarray) and (0 in a):
return a/np.where(a==0,math.nan,a)
return a**b
def sqrt(a):
return np.sqrt(abs(a))
def sqrd(a):
return a**2
def inv(a):
return np.array(a).astype(float)**(-1)
def neg(a):
return -a
def sin(a):
return np.sin(a)
def cos(a):
return np.cos(a)
def tan(a):
return np.tan(a)
def arccos(a):
return np.arccos(a)
def arcsin(a):
return np.arcsin(a)
def arctan(a):
return np.arctan(a)
def tanh(a):
return np.tanh(a)
def log(a):
a = np.maximum(np.abs(a), 1e-9)
return np.log(a)
def log10(a):
return np.log10(a)
def log2(a):
return np.log2(a)
def abs1(a):
return np.abs(a)
def and1(a,b):
return np.logical_and(a,b)
def or1(a,b):
return np.logical_or(a,b)
def xor1(a,b):
return np.logical_xor(a,b)
def nand1(a,b):
return np.logical_not(np.logical_and(a,b))
def nor1(a,b):
return np.logical_not(np.logical_or(a,b))
def xnor1(a,b):
return np.logical_not(np.logical_xor(a,b))
def not1(a):
return np.logical_not(a)
def defaultOps():
return [protectDiv,add,sub,mult,exp,sqrd,sqrt,inv,neg,"pop","pop","pop","pop","pop","pop"]
def allOps():
return [protectDiv,add,sub,mult,exp,sqrd,sqrt,inv,neg,cos,sin,tan,arccos,arcsin,arctan,tanh,log,"pop","pop","pop","pop","pop","pop","pop","pop","pop","pop"]
def customOps():
return [protectDiv,add,sub,mult,exp,sqrd,sqrt,inv,neg,cos,sin,log,"pop","pop","pop","pop","pop","pop","pop","pop"]
def booleanOps():
return [and1,or1,xor1,nand1,nor1,xnor1,not1,"pop","pop","pop","pop","pop","pop","pop"]
def randomInt(a=-3,b=3):
return random.randint(a,b)
def defaultConst():
return [np.pi, np.e, randomInt,ranReal ]
def booleanConst():
return [1,0]
def ranReal(a=20,b=-10):
return random.random()*a-b
##### MODEL DISTRIBUTION CODE #####
import sympy as sym
import sympy as sym
from sympy import preorder_traversal
from collections import defaultdict
def extract_genotype(population_buffer):
"""
Extracts in terms of the geneotypes instead of using pheno
"""
stats = {
"operators": [],
"terminals": [],
}
for prog in population_buffer:
print(prog[0], prog[1])
ops = [op.__name__ if hasattr(op, '__name__') else str(op) for op in prog[0]]
stats["operators"].extend(ops)
#(Features and Constants)
stats["terminals"].extend([str(t) for t in prog[1]])
return stats
def extract_model_stats(expr):
stats = {
"constants": [],
"operators": [],
"features": [],
"powers": []
}
for node in preorder_traversal(expr):
if isinstance(node, sym.Number):
stats["constants"].append(float(node))
elif isinstance(node, sym.Symbol):
stats["features"].append(str(node))
# POWERS
elif isinstance(node, sym.Pow):
base, exp = node.args
stats["operators"].append("pow")
if exp.is_number:
stats["powers"].append(float(exp))
elif getattr(node, 'is_Function', False):
op_name = node.func.__name__
stats["operators"].append(op_name)
elif isinstance(node, sym.Add):
has_sub = any(arg.could_extract_minus_sign() for arg in node.args)
stats["operators"].append("add")
if has_sub:
stats["operators"].append("sub")
# MUL → detect * vs /
elif isinstance(node, sym.Mul):
has_div = False
for arg in node.args:
if isinstance(arg, sym.Pow):
base, exp = arg.args
if exp.is_number and exp < 0:
has_div = True
break
if has_div:
stats["operators"].append("div")
else:
stats["operators"].append("mul")
return stats
def get_model_distributions(models):
'''
Returns list of dictionaries, where each dict is breakdown of each model from list of models
'''
all_models_dist, gene_dist = [], []
for i, mod in enumerate(models):
expr = printGPModel(mod)
if isinstance(expr, sym.Expr):
expr = sym.simplify(expr)
#expr = sym.expand(expr)
stats = extract_model_stats(expr)
stats["DEBUG_INDEX"] = i
all_models_dist.append(stats)
print(f"--- MODEL INDEX {i} ---")
print(stats)
gene_dist.append(extract_genotype(stats))
return all_models_dist, gene_dist
# def get_model_distribution(models):
# # We will store the distributions in a dictionary where the key is the feature
# # and the value is a list of terms across all models from pareto front
# feature_distributions = {}
# all_models_expanded = []
# for mod in models:
# expr = printGPModel(mod)
# print(expr)
# if isinstance(expr, sym.Expr):
# expanded_expr = sym.expand(expr)
# else:
# expanded_expr = expr
# terms = expanded_expr.args if hasattr(expanded_expr, 'args') and expanded_expr.func == sym.Add else (expanded_expr,)
# print("WE ARE PRINTING OUT THE TERMS FOR ONE MODEL")
# print(terms)
# all_models_expanded.append(terms)
# for term in terms:
# print("EACH INDIVIDUAL term FOR ONE MODEL", term)
# if hasattr(term, 'free_symbols'):
# print("HAS FROZEN SYMBOLS")
# sig = frozenset(term.free_symbols)
# print("sig", sig)
# if sig not in feature_distributions:
# #THIS REPRESENTS THE MODEL REPRESENTATION
# feature_distributions[sig] = []
# for terms in all_models_expanded:
# current_model_contributions = {sig: 0 for sig in feature_distributions.keys()}
# for term in terms:
# if hasattr(term, 'free_symbols'):
# sig = frozenset(term.free_symbols)
# current_model_contributions[sig] += term
# else:
# sig = frozenset()
# current_model_contributions[sig] += term
# # Append this model's contributions to the global buckets
# for sig in feature_distributions.keys():
# feature_distributions[sig].append(current_model_contributions[sig])
# return feature_distributions
############################
#Data Subsampling Methods
############################
def randomSubsample(x,y, *args, **kwargs):
n=max(int(np.ceil(len(y)**(3/5))),3)
idx=np.random.choice(range(x.shape[1]),n,replace=False)
return np.array([i[idx] for i in x]),y[idx]
def generationProportionalSample(x,y,generation=100,generations=100):
n=max(int(np.ceil(len(y)*(generation/generations)**(3/5))),3)
idx=np.random.choice(range(x.shape[1]),n,replace=False)
return np.array([i[idx] for i in x]),y[idx]
def ordinalSample(x,y,generation=100,generations=100):
n=max(int(len(y)*generation/generations),3)
sortedIdx=np.argsort(y)
step=len(y)/(n-1)
idx=[sortedIdx[max(int(i*step)-1,0)] for i in range(n)]
return np.array([i[idx] for i in x]),y[idx]
def orderedSample(x,y,generation=100,generations=100):
n=max(int(len(y)*generation/generations),3)
idx=[i for i in range(n)]
return np.array([i[idx] for i in x]),y[idx]
def ordinalBalancedSample(x,y,generation=100,generations=100):
n=max(int(len(y)*generation/generations),3)
numBins=int(max(np.ceil(np.sqrt(n)),3))
bins=np.linspace(min(y),max(y),numBins+1)
binIdx=np.digitize(y,bins)-1
samplesPerBin=max(int(n/numBins),1)
idx=[]
for i in range(numBins):
binMembers=[j for j in range(len(y)) if binIdx[j]==i]
if len(binMembers)>0:
chosen=np.random.choice(binMembers,min(samplesPerBin,len(binMembers)),replace=False)
idx=idx+chosen.tolist()
return np.array([i[idx] for i in x]),y[idx]
def balancedSample(x,y, *args, **kwargs):
n=int(np.ceil(len(y)**(3/5)))
numBins=max(round(n**(2/5)),3)
bins=np.linspace(min(y),max(y),numBins+1)
binIdx=np.digitize(y,bins)-1
samplesPerBin=max(int(n/numBins),1)
idx=[]
for i in range(numBins):
binMembers=[j for j in range(len(y)) if binIdx[j]==i]
if len(binMembers)>0:
chosen=np.random.choice(binMembers,min(samplesPerBin,len(binMembers)),replace=False)
idx=idx+chosen.tolist()
return np.array([i[idx] for i in x]),y[idx]
import inspect
def getArity(func): #Returns the arity of a function: used for model evaluations
if func=="pop":
return 1
return len(inspect.signature(func).parameters)
getArity.__doc__ = "getArity(func) takes a function and returns the function arity"
def modelArity(model): #Returns the total arity of a model
return 1+sum([getArity(i)-1 for i in model[0]])
modelArity.__doc__ = "modelArity(model) returns the total arity of a model"
def listArity(data): #Returns arity of evaluating a list of operators
if len(data)==0:
return 0
return 1+sum([getArity(i)-1 for i in data])
listArity.__doc__ = "listArity(list) returns the arity of evaluating a list of operators"
def buildEmptyModel(): # Generates an empty model
return [[],[],[]]
buildEmptyModel.__doc__ = "buildEmptyModel() takes no inputs and generates an empty GP model"
def variableSelect(num): #Function that creates a function to select a specific variable
return lambda variables: variables[num]
variableSelect.__doc__ = "variableSelect(n) is a function that creates a function to select the nth variable"
def modelToListForm(model):
model[0]=model[0].tolist()
def modelRestoreForm(model):
model[0]=np.array(model[0],dtype=object)
def generateRandomModel(variables,ops,const,maxLength): #Generates a random GP model
prog = buildEmptyModel() #Generate an empty model with correct structure
varChoices=[variableSelect(i) for i in range(variables)]+const #All variable and constants choices
prog[0]=np.array(np.random.choice(ops,random.randint(1,maxLength)),dtype=object) #Choose random operators
countVars=modelArity(prog) #Count how many variables/constants are needed
prog[1]=np.random.choice(varChoices,countVars) #Choose random variables/constants
# if all vars are constants then replace one random term
if all(t in const for t in prog[1]):
replace_idx = random.randrange(countVars)
prog[1][replace_idx] = random.choice(varChoices[:variables]) #Replace with a variable
prog[1]=[i() if (callable(i) and i.__name__!='<lambda>' )else i for i in prog[1]] #If function then evaluate
return prog
generateRandomModel.__doc__ = "generateRandomModel() takes as input the variables, operators, constants, and max program length and returns a random program"
def initializeGPModels(variables,ops=defaultOps(),const=defaultConst(),numberOfModels=100,maxLength=10): # generate random linear program
prog=[[],[],[]]
# prog stores [Operators, VarConst, QualityMetrics]
models=[generateRandomModel(variables,ops,const,maxLength) for i in range(numberOfModels)] #Generate models
return models
initializeGPModels.__doc__ = "initializeGPModels(countOfVariables, operators, constants, numberOfModels=100, maxLength=10) returns a set of randomly generated models"
def reverseList(data): #Returns a list reversed
return [i for i in reversed(data)]
reverseList.__doc__ = "reverseList(data) returns the data list reversed"
def varReplace(data,variables): #Replaces variable references with data during model evaluation
return [i(variables) if callable(i) else i for i in data]
varReplace.__doc__ = "varReplace(data,variables) replaces references to variables in data with actual values"
def inputLen(data): #Returns the number of data records in a data set
el1=data[0]
if type(el1)==list or type(el1)==np.ndarray:
return len(el1)
else:
return 1
inputLen.__doc__ = "inputLen(data) determines the number of data records in a data set"
def varCount(data): #Returns the number of variables in a data set
return len(data)
varCount.__doc__ = "varCount(data) determines the number of variables in a data set"
def evaluateGPModel(model,inputData): #Evaluates a model numerically
response=evModHelper(model[1],model[0],[],np.array(inputData).astype(float))[2][0]
if not type(response)==np.ndarray and inputLen(inputData)>1:
response=np.array([response for i in range(inputLen(inputData))])
return response
evaluateGPModel.__doc__ = "evaluateGPModel(model,data) numerically evaluates a model using the data stored in inputData"
def evModHelper(varStack,opStack,tempStack,data): #Recursive helper function for evaluateGPModel
stack1=varStack
stack2=opStack
stack3=tempStack
if len(stack2)==0:
return [stack3,stack2,stack1]
op=stack2[0]
stack2=stack2[1:]
if callable(op):
patt=getArity(op)
while patt>len(stack3):
stack3=[stack1[0]]+stack3
stack1=stack1[1:]
try:
temp=op(*varReplace(reverseList(stack3[:patt]),data))
except TypeError:
print("stack3: ", stack3, " patt: ", patt, " data: ", data)
temp=np.nan
except OverflowError:
temp=np.nan
stack3=stack3[patt:]
stack3=[temp]+stack3
else:
if len(stack1)>0:
stack3=varReplace([stack1[0]],data)+stack3
stack1=stack1[1:]
if len(stack2)>0:
stack1,stack2,stack3=evModHelper(stack1,stack2,stack3,data)
return [stack1,stack2,stack3]
evModHelper.__doc__ = "evModHelper(varStack,opStack,tempStack,data) is a helper function for evaluateGPModel"
def rmse(model, inputData, response):
predictions = evaluateGPModel(model, inputData)
if not all(np.isfinite(predictions)) or any(np.iscomplex(predictions)):
return np.nan
return np.sqrt(np.mean((predictions - response) ** 2))
rmse.__doc__ = "rmse(model, input, response) is a fitness objective that evaluates the root mean squared error"
def binaryError(model, input, response):
prediction=evaluateGPModel(model,input)
error=np.mean(np.abs(prediction-response))
if np.isnan(error) or np.isinf(error) or error > 1 or error < 0:
return 0.5
return min(error,1 - error)
def fitness(prog,data,response): # Fitness function using correlation
predicted=evaluateGPModel(prog,np.array(data))
if type(predicted)!=list and type(predicted)!=np.ndarray:
predicted=np.array([predicted for i in range(inputLen(data))])
try:
if np.isnan(predicted).any() or np.isinf(predicted).any():
return np.nan
except TypeError:
#print(predicted)
return np.nan
except OverflowError:
return np.nan
if (not all(np.isfinite(np.array(predicted,dtype=np.float32)))) or np.all(predicted==predicted[0]):
return np.nan
try:
fit=1-pearsonr(predicted,np.array(response))[0]**2 # 1-R^2
except ValueError:
return 1
if math.isnan(fit):
return 1 # If nan return 1 as fitness
return fit # Else return actual fitness 1-R^2
fitness.__doc__ = "fitness(program,data,response) returns the 1-R^2 value of a model"
def stackGPModelComplexity(model,*args):
return len(model[0])+len(model[1])-model[0].tolist().count("pop")
stackGPModelComplexity.__doc__ = "stackGPModelComplexity(model) returns the complexity of the model"
###################### Timeout function for model complexity ######################
class TimeoutException(Exception): pass
@contextmanager
def time_limit(seconds):
def signal_handler(signum, frame):
raise TimeoutException("Timed out!")
signal.signal(signal.SIGALRM, signal_handler)
signal.alarm(seconds)
try:
yield
finally:
signal.alarm(0)
####################################################################################
# Compute Hess
def ComputeSymbolicHess(model,vars):
printedModel=sym.simplify(printGPModel(model))
if type(printedModel)==float:
return sym.matrices.dense.MutableDenseMatrix(np.zeros((vars,vars)))
hess=sym.hessian(printedModel, [symbols('x'+str(i)) for i in range(vars)])
return hess
def EvaluateHess(hess,vars,values):
numHess=hess.subs({symbols('x'+str(j)):values[j] for j in range(vars)})
hessN = np.array(numHess).astype(float)
rankN=np.linalg.matrix_rank(hessN,tol=0.0001*0.0001*10)
return rankN
def Approx2Deriv(model,values,diff1,diff2,positions): #maybe diff should be relative to the variation of each feature
term1=[values[i]+diff1 if i == positions[0] else values[i] for i in range(len(values))]
term1=[term1[i]+diff2 if i == positions[1] else term1[i] for i in range(len(term1))]
term2=[values[i]-diff1 if i == positions[0] else values[i] for i in range(len(values))]
term2=[term2[i]+diff2 if i == positions[1] else term2[i] for i in range(len(term2))]
term3=[values[i]+diff1 if i == positions[0] else values[i] for i in range(len(values))]
term3=[term3[i]-diff2 if i == positions[1] else term3[i] for i in range(len(term3))]
term4=[values[i]-diff1 if i == positions[0] else values[i] for i in range(len(values))]
term4=[term4[i]-diff2 if i == positions[1] else term4[i] for i in range(len(term4))]
return ((evaluateGPModel(model,term1)-evaluateGPModel(model,term2))/((2*diff1))
-(evaluateGPModel(model,term3)-evaluateGPModel(model,term4))/((2*diff1)))/(2*diff2)
def ApproxHessRank(model,vars,values,diff1=0.001,diff2=0.001):
hess=[[Approx2Deriv(model,values,diff1,diff2,[i,j]) for i in range(vars)] for j in range(vars)]
hessN = np.array(hess).astype(float)
rankN=np.linalg.matrix_rank(hessN,tol=0.0001*0.0001*10)
return rankN
#def HessRank(model,vars,values):
# try:
# with time_limit(.01):
# hess=ComputeSymbolicHess(model,vars)
# hess = EvaluateHess(hess,vars,values)
# #print(hess)
# return hess
# except TimeoutException as e:
# hess=ApproxHessRank(model,vars,values)
#print(hess)
# return hess
def HessRank(model,vars,values):
hess=ApproxHessRank(model,vars,values)
return hess
# Counts basis terms in a model
def count_basis_terms(equation, expand=False):
try:
with time_limit(2):
if expand:
# Simplify the equation to standardize the expression
simplified_eq = simplify(equation)
# Expand the expression to identify additive terms clearly
expanded_eq = expand(simplified_eq)
# Separate the terms of the expression
terms = expanded_eq.as_ordered_terms()
else:
terms = equation.as_ordered_terms()
#print(terms)
except TimeoutException as e:
return 1000
return len(terms)
# Determines the number of basis functions in a model by counting +s and -s
def basisFunctionComplexity(model,vars, values,*args):
try: # values should be max, min, and median with respect to response variable
return HessRank(model,vars,values)#count_basis_terms(printGPModel(model))
except:
return 1000
# Creates a lambda function to be used as a complexity metric when given a target dimensionality and deviation
def basisFunctionComplexityDiff(target, deviation, vars, low, mid, high):
return lambda model,*args: max(np.mean([abs(basisFunctionComplexity(model,vars,low)-target),abs(basisFunctionComplexity(model,vars,mid)-target) ,abs(basisFunctionComplexity(model,vars,high)-target)] ),(deviation))-deviation
def setModelQuality(model,inputData,response,modelEvaluationMetrics=[fitness,stackGPModelComplexity]):
model[2]=[i(model,inputData,response) for i in modelEvaluationMetrics]
setModelQuality.__doc__ = "setModelQuality(model, inputdata, response, metrics=[r2,size]) is an inplace operator that sets a models quality"
def stackPass(model,pt):
i=0
t=0
p=0
s=model[0]
if i <pt:
t+=1
while i<pt:
if s[i]=="pop":
t+=1
p+=1
else:
p+=max(0,getArity(s[i])-t)
t=max(1,t-getArity(s[i])+1)
i+=1
stack1=model[1][p:]
stack2=reverseList(model[1][:p])[:t+1]
return [stack1,stack2]
def stackGrab(stack1, stack2, num):
tStack1=copy.deepcopy(stack1)
tStack2=copy.deepcopy(stack2)
newStack=[]
if len(stack2)<num:
newStack=stack2+stack1[:(num-len(stack2))]
tStack1=tStack1[num-len(tStack2):]
tStack2=[]
else:
newStack=stack2[:num]
tStack2=tStack2[num:]
return [newStack,tStack1,tStack2]
def fragmentVariables(model,pts):
stack1,stack2=stackPass(model,pts[0])
opStack=model[0]
newStack=[]
i=pts[0]
while i<=pts[1]:
if opStack[i]=="pop" and len(stack1)>0:
stack2=[stack1[0]]+stack2
stack1=stack1[1:]
else:
if len(newStack)==0 and pts[0]==0:
tStack,stack1,stack2=stackGrab(stack1,stack2,getArity(opStack[i]))
else:
tStack,stack1,stack2=stackGrab(stack1,stack2,getArity(opStack[i])-1)
newStack=newStack+tStack
i+=1
return newStack
def recombination2pt(model1,model2): #2 point recombination
pts1=np.sort(random.sample(range(0,len(model1[0])+1),2))
pts2=np.sort(random.sample(range(0,len(model2[0])+1),2))
#pts1=[4,5]
#pts2=[2,4]
#pts1=[0,3]
#pts2=[1,3]
#print(pts1,pts2)
child1=buildEmptyModel()
child2=buildEmptyModel()
parent1=copy.deepcopy(model1)
parent2=copy.deepcopy(model2)
parent1[0]=np.array(parent1[0],dtype=object).tolist()
parent2[0]=np.array(parent2[0],dtype=object).tolist()
child1[0]=np.array(parent1[0][0:pts1[0]]+parent2[0][pts2[0]:pts2[1]]+parent1[0][pts1[1]:],dtype=object)
child2[0]=np.array(parent2[0][0:pts2[0]]+parent1[0][pts1[0]:pts1[1]]+parent2[0][pts2[1]:],dtype=object)
varPts1=[listArity(parent1[0][:(pts1[0])])+0,listArity(parent2[0][:(pts2[0])])+0,listArity(parent2[0][pts2[0]:pts2[1]]),listArity(parent1[0][pts1[0]:pts1[1]])]
if pts1[0]==0:
varPts1[0]+=1
if pts2[0]==0:
varPts1[1]+=1
child1[1]=parent1[1][:varPts1[0]]+parent2[1][varPts1[1]:(varPts1[1]+varPts1[2]-1)]+parent1[1][(varPts1[0]+varPts1[3]-1):]
varPts2=[listArity(parent2[0][:(pts2[0])])+0,listArity(parent1[0][:(pts1[0])])+0,listArity(parent1[0][pts1[0]:pts1[1]]),listArity(parent2[0][pts2[0]:pts2[1]])]
if pts1[0]==0:
varPts2[1]+=1
if pts2[0]==0:
varPts2[0]+=1
child2[1]=parent2[1][:varPts2[0]]+parent1[1][varPts2[1]:(varPts2[1]+varPts2[2]-1)]+parent2[1][(varPts2[0]+varPts2[3]-1):]
#print(varPts1,varPts2)
return [child1,child2]
recombination2pt.__doc__ = "recombination2pt(model1,model2) does 2 point crossover and returns two children models"
def get_numeric_indices(l): #Returns indices of list that are numeric
return [i for i in range(len(l)) if type(l[i]) in [int,float]]
def mutate(model,variables,ops=defaultOps(),const=defaultConst(),maxLength=10):
newModel=copy.deepcopy(model)
newModel[0]=np.array(newModel[0],dtype=object).tolist()
mutationType=random.randint(0,7)
varChoices=[variableSelect(i) for i in range(variables)]+const
opChoice=0
varChoice=0
tmp=0
if mutationType==0: #single operator mutation
opChoice=random.randint(0,len(newModel[0])-1)
if len(newModel[0])>0:
newModel[0][opChoice]=np.random.choice([i for i in ops] )
elif mutationType==1: #single variable mutation
varChoice=np.random.choice(varChoices)
if callable(varChoice) and varChoice.__name__!='<lambda>':
varChoice=varChoice()
newModel[1][random.randint(0,len(newModel[1])-1)]=varChoice
elif mutationType==2: #insertion mutation to top of stack
opChoice=np.random.choice(ops)
newModel[0]=[opChoice]+newModel[0]
while modelArity(newModel)>len(newModel[1]):
varChoice=np.random.choice(varChoices)
if callable(varChoice) and varChoice.__name__!='<lambda>':
varChoice=varChoice()
newModel[1]=[varChoice]+newModel[1]
elif mutationType==3: #deletion mutation from top of stack
if len(newModel[0])>1:
opChoice=random.randint(1,len(newModel[0])-1)
newModel[0]=newModel[0][-opChoice:]
newModel[1]=newModel[1][-listArity(newModel[0]):]
elif mutationType==4: #insertion mutation to bottom of stack
opChoice=np.random.choice([i for i in ops])
newModel[0].append(opChoice)
elif mutationType==5: #mutation via crossover with random model
newModel=recombination2pt(newModel,generateRandomModel(variables,ops,const,maxLength))[0]
elif mutationType==6: #single operator insertion mutation
singleOps=[op for op in ops if getArity(op)==1 and op!='pop']
singleOps.append('pop')
pos=random.randint(0,len(newModel[0])-1)
newModel[0].insert(pos,np.random.choice(singleOps))
elif mutationType==7: #nudge numeric constant
pos=get_numeric_indices(newModel[1])
if(len(pos)>0): #If there are numeric constants
pos=random.choice(pos)
newModel[1][pos]=newModel[1][pos]+np.random.normal(-1,1)
if modelArity(newModel)<len(newModel[1]):
newModel[1]=newModel[1][:modelArity(newModel)]
elif modelArity(newModel)>len(newModel[1]):
newModel[1]=newModel[1]+[np.random.choice(varChoices) for i in range(modelArity(newModel)-len(newModel[1]))]
newModel[1]=[varChoice() if callable(varChoice) and varChoice.__name__!='<lambda>' else varChoice for varChoice in newModel[1]]
newModel[0]=np.array(newModel[0],dtype=object)
return newModel
mutate.__doc__ = "mutate(model,variableCount,ops,constants,maxLength) mutates a model"
def paretoFront(fitValues): #Returns Boolean list of Pareto front elements
onFront = np.ones(fitValues.shape[0], dtype = bool)
for i, j in enumerate(fitValues):
if onFront[i]:
onFront[onFront] = np.any(fitValues[onFront]<j, axis=1)
onFront[i] = True
return onFront
def paretoTournament(pop): # selects the Pareto front of a model set
fitnessValues=np.array([mod[2] for mod in pop])
return (np.array(pop,dtype=object)[paretoFront(fitnessValues)]).tolist()
def tournamentModelSelection(models, popSize=100,tourneySize=5):
selectedModels=[]
selectionSize=popSize
while len(selectedModels)<popSize:
tournament=random.sample(models,tourneySize)
winners=paretoTournament(tournament)
selectedModels=selectedModels+winners
return selectedModels
paretoTournament.__doc__ = "paretoTournament(models, inputData, responseData) returns the Pareto front of a model set"
def modelSameQ(model1,model2): #Checks if two models are the same
return len(model1[0])==len(model2[0]) and len(model1[1]) == len(model2[1]) and all(model1[0]==model2[0]) and model1[1]==model2[1]
modelSameQ.__doc__ = "modelSameQ(model1,model2) checks if model1 and model2 are the same and returns True if so, else False"
def deleteDuplicateModels(models): #Removes any models that are the same, does not consider simplified form
uniqueMods = [models[0]]
for mod in models:
test=False
for checkMod in uniqueMods:
if modelSameQ(mod,checkMod):
test=True
if not test:
uniqueMods.append(mod)
return uniqueMods
deleteDuplicateModels.__doc__ = "deleteDuplicateModels(models) deletes models that have the same form without simplifying"
def deleteDuplicateModelsPhenotype(models): #Removes any models that are the same regarding phenotype, does not consider simplified form
uniqueMods = [printGPModel(models[0])]
remainingMods=[printGPModel(mod) for mod in models[1:]]
uniquePos = [0]
currPos=1
for mod in remainingMods:
test=False
for checkMod in uniqueMods:
if mod==checkMod:
test=True
if not test:
uniqueMods.append(mod)
uniquePos.append(currPos)
currPos+=1
return [models[i] for i in uniquePos]
def removeIndeterminateModels(models): #Removes models from the population that evaluate to nonreal values
return [i for i in models if (not any(np.isnan(i[2]))) and all(np.isfinite(np.isnan(i[2])))]
removeIndeterminateModels.__doc__ = "removeIndeterminateModels(models) removes models that have a fitness that results from inf or nan values"
def sortModels(models):
return sorted(models, key=lambda m:m[2])
sortModels.__doc__ = "sortModels(models) sorts a model population by the models' accuracies"
def selectModels(models, selectionSize=0.5, thresholds=None):
tMods=copy.deepcopy(models)
[modelToListForm(mod) for mod in tMods]
if thresholds is not None:
tMods=[mod for mod in tMods if all([mod[2][i]<=thresholds[i] for i in range(len(thresholds))])]
paretoModels=[]
if selectionSize<=1:
selection=selectionSize*len(models)
else:
selection=selectionSize
while len(paretoModels)<selection and len(tMods)>0:
front=paretoTournament(tMods)
paretoModels=paretoModels+front
for i in front:
tMods.remove(i)
[modelRestoreForm(mod) for mod in paretoModels]
return paretoModels
selectModels.__doc__ = "selectModels(models, selectionSize=0.5) iteratively selects the Pareto front of a model population until n or n*popSize models are selected"
def stackVarUsage(opStack): #Counts how many variables are used by the operator stack
pos=getArity(opStack[0])
for j in range(1,len(opStack)):
pos+=getArity(opStack[j])-1
if opStack[j]=='pop':
pos+=1
return pos
stackVarUsage.__doc__ = "stackVarUsage(opStack) is a helper function that determines how many variables/constants are needed by the operator stack"
def trimModel(mod): #Removes extra pop operators that do nothing
model=copy.deepcopy(mod)
i=0
varStack=len(mod[1])
tempStack=0
varStack-=getArity(model[0][i])
tempStack+=1
i+=1
while varStack>0:
if model[0][i]=='pop':
varStack-=1
tempStack+=1
else:
take=getArity(model[0][i])-tempStack
if take>0:
varStack-=take
tempStack=1
else:
tempStack-=getArity(model[0][i])-1
i+=1
model[0]=np.array(model[0][:i].tolist()+[j for j in model[0][i:] if not j=='pop'],dtype=object)
return model
trimModel.__doc__ = "trimModel(model) trims extra pop operators off the operator stack so that further modifications such as a model alignment aren't altered by those pop operators"
def alignGPModel(model, data, response): #Aligns a model
prediction=evaluateGPModel(model,data)
if (not all(np.isfinite(np.array(prediction)))) or np.all(prediction==prediction[0]):
return model
if np.isnan(np.array(prediction)).any() or np.isnan(np.array(response)).any() or not np.isfinite(np.array(prediction,dtype=np.float32)).all():
return model
# Variance guards
if np.std(prediction) < 1e-12:
return model
if np.ptp(prediction) < 1e-12:
return model
try:
align=np.polyfit(prediction,response,1,rcond=1e-16)#np.round(np.polyfit(prediction,response,1,rcond=1e-16),decimals=14)
except np.linalg.LinAlgError:
#print("Alignment failed for: ", model, " with prediction: ", prediction, "and reference data: ", response)
return model
newModel=trimModel(model)
newModel[0]=np.array(newModel[0].tolist()+[mult,add],dtype=object)
newModel[1]=newModel[1]+align.tolist()
#setModelQuality(newModel,data,response)
return newModel
alignGPModel.__doc__ = "alignGPModel(model, input, response) aligns a model such that response-a*f(x)+b are minimized over a and b"
def evolve(inputData, responseData, generations=100, ops=defaultOps(), const=defaultConst(), variableNames=[], mutationRate=79, crossoverRate=11, spawnRate=10, extinction=False,extinctionRate=10,elitismRate=10,popSize=300,maxComplexity=100,align=True,initialPop=[],timeLimit=300,capTime=False,tourneySize=5,tracking=False,returnTracking=False,liveTracking=False,liveTrackingInterval=1,modelEvaluationMetrics=[fitness,stackGPModelComplexity],dataSubsample=False,samplingMethod=randomSubsample,alternateObjectives=[],alternateObjFrequency=10,allowEarlyTermination=False,earlyTerminationThreshold=0):
evolution_hisotry = []
alternatingFlag = False
if callable(modelEvaluationMetrics):
metrics=[modelEvaluationMetrics]
allMetrics=[modelEvaluationMetrics]+alternateObjectives
elif isinstance(modelEvaluationMetrics, list) and callable(modelEvaluationMetrics[0]):
metrics=modelEvaluationMetrics
allMetrics=modelEvaluationMetrics+alternateObjectives
elif isinstance(modelEvaluationMetrics, list) and isinstance(modelEvaluationMetrics[0], list):
metrics=modelEvaluationMetrics[0]
allMetrics=[item for sublist in modelEvaluationMetrics for item in sublist]+alternateObjectives
alternatingFlag = True
else:
raise ValueError("modelEvaluationMetrics must be a function, list of functions, or a list of lists of functions")
fullInput,fullResponse=copy.deepcopy(inputData),copy.deepcopy(responseData)
inData=copy.deepcopy(fullInput)
resData=copy.deepcopy(fullResponse)
variableCount=varCount(inData)
models=initializeGPModels(variableCount,ops,const,popSize)
models=models+initialPop
startTime=time.perf_counter()
bestFits=[]
gene_dists = []
if liveTracking:
fig, ax = plt.subplots(figsize=(20,10))
ckTime=time.perf_counter()
for i in range(generations):
if capTime and time.perf_counter()-startTime>timeLimit:
break
if len(alternateObjectives)>0 and (i+1)%alternateObjFrequency==0:
metrics=modelEvaluationMetrics[:1]+alternateObjectives
else:
if alternatingFlag:
metrics=modelEvaluationMetrics[i%len(modelEvaluationMetrics)]
else:
metrics=modelEvaluationMetrics
if dataSubsample:
inData,resData=samplingMethod(fullInput,fullResponse,generations=generations,generation=i)
for mods in models:
setModelQuality(mods,inData,resData,modelEvaluationMetrics=metrics)
models=removeIndeterminateModels(models)
if allowEarlyTermination and min([mods[2][0] for mods in models])<=earlyTerminationThreshold:
print("Early termination at generation ", i)
break
if tracking or liveTracking or returnTracking:
bestFits.append(min([mods[2][0] for mods in paretoTournament(models)]))
if liveTracking and time.perf_counter()-ckTime>liveTrackingInterval:
ax.clear()
ax.plot(bestFits)
ax.set_title(f"Best Model: {bestFits[-1]:.2f} at Generation {(i+1)}")
ax.set_xlabel("Generations")
ax.set_ylabel("Fitness")
clear_output(wait=True)
display(fig)
#plt.show()
plt.close(fig)
ckTime=time.perf_counter()
#get distribution of the models from the last generation and use the models from the Pareto Front
paretoModels=selectModels(models,elitismRate/100*popSize if elitismRate/100*popSize<len(models) else len(models))
if extinction and i%extinctionRate==0 and i>0:
models=initializeGPModels(variableCount,ops,const,popSize)
for mods in models:
setModelQuality(mods,inData,resData,modelEvaluationMetrics=metrics)
models=tournamentModelSelection(models,popSize,tourneySize)
crossoverPairs=random.sample(models,round(crossoverRate/100*popSize))
toMutate=random.sample(models,round(mutationRate/100*popSize))
childModels=paretoModels
for j in range(round(len(crossoverPairs)/2)-1):
childModels=childModels+recombination2pt(crossoverPairs[j],crossoverPairs[j+round(len(crossoverPairs)/2)])
for j in toMutate:
childModels=childModels+[mutate(j,variableCount,ops,const)]
childModels=childModels+initializeGPModels(variableCount,ops,const,round(spawnRate/100*popSize))
childModels=deleteDuplicateModels(childModels)
childModels=[model for model in childModels if stackGPModelComplexity(model)<maxComplexity]
#for mods in childModels:
# setModelQuality(mods,inData,resData,modelEvaluationMetrics=modelEvaluationMetrics)
#childModels=removeIndeterminateModels(childModels)
if len(childModels)<popSize:
childModels=childModels+initializeGPModels(variableCount,ops,const,popSize-len(childModels))
models=copy.deepcopy(childModels)
if ( i + 5 >= generations):
evolution_hisotry.append(models)
yield models
for mods in models:
setModelQuality(mods,fullInput,fullResponse,modelEvaluationMetrics=allMetrics)
models=[trimModel(mod) for mod in models]
models=deleteDuplicateModels(models)
models=removeIndeterminateModels(models)
models=sortModels(models)
if align:
models=[alignGPModel(mods,fullInput,fullResponse) for mods in models]
for mods in models:
setModelQuality(mods,fullInput,fullResponse,modelEvaluationMetrics=allMetrics)
if tracking or returnTracking:
bestFits.append(min([mods[2][0] for mods in paretoTournament(models)]))
if returnTracking:
return models, bestFits
plt.figure()
plt.plot(bestFits)
plt.title("Fitness over Time")
plt.xlabel("Generations")
plt.ylabel("Fitness")
plt.show()
return models,evolution_hisotry
#model_distributions
def replaceFunc(stack,f1,f2):
return [i if i!=f1 else f2 for i in stack]
def printGPModel(mod,inputData=symbols(["x"+str(i) for i in range(100)])): #Evaluates a model algebraically
def inv1(a):
return a**(-1)
from sympy import tan as tan1, exp as exp1, sqrt as sqrt1, sin as sin1, cos as cos1, acos, asin, atan, tanh as tanh1, log as log1
def sqrt2(a):
return sqrt1(a)
def log2(a):
return log1(a)
model = copy.deepcopy(mod)
model[0] = replaceFunc(model[0],exp,exp1)
model[0] = replaceFunc(model[0],tan,tan1)
model[0] = replaceFunc(model[0],sqrt,sqrt2)
model[0] = replaceFunc(model[0],inv,inv1)
model[0] = replaceFunc(model[0],sin,sin1)
model[0] = replaceFunc(model[0],cos,cos1)
model[0] = replaceFunc(model[0],arccos,acos)
model[0] = replaceFunc(model[0],arcsin,asin)
model[0] = replaceFunc(model[0],arctan,atan)
model[0] = replaceFunc(model[0],tanh,tanh1)
model[0] = replaceFunc(model[0],log,log2)
try:
response=evModHelper(model[1],model[0],[],np.array(inputData))[2][0]
except:
return np.nan
return response
def ensembleSelect(models, inputData, responseData, numberOfClusters=10): #Generates a model ensemble using input data partitions
data=np.transpose(inputData)
if len(data)<numberOfClusters:
numberOfClusters=len(data)
clusters=KMeans(n_clusters=numberOfClusters).fit_predict(data)
if numberOfClusters>len(set(clusters)):
numberOfClusters=len(set(clusters))
clusters=KMeans(n_clusters=numberOfClusters).fit_predict(data)
dataParts=[]
partsResponse=[]
for i in range(numberOfClusters):
dataParts.append([])
partsResponse.append([])
for i in range(len(clusters)):
dataParts[clusters[i]].append(data[i])
partsResponse[clusters[i]].append(responseData[i])
modelResiduals=[]
for i in range(len(models)):
modelResiduals.append([])
for i in range(len(models)):
for j in range(numberOfClusters):
modelResiduals[i].append(fitness(models[i],np.transpose(dataParts[j]),partsResponse[j]))
best=[]
for i in range(numberOfClusters):
ordering=np.argsort(modelResiduals[i])
j=0
while ordering[j] in best:
j+=1
best.append(ordering[j])
ensemble=[models[best[i]] for i in range(numberOfClusters)]
return ensemble
def uncertainty(data,trim=0.3):
wl=None
if len(data)<=4:
wl=1
h=differential_entropy(data,window_length=wl)
if np.isfinite(h):
return h
else:
return 0
def evaluateModelEnsemble(ensemble, inputData):
responses=[evaluateGPModel(mod, inputData) for mod in ensemble]
if type(responses[0])==np.ndarray:
responses=np.transpose(responses)
predictions=[np.median(res) for res in responses]
else:
predictions=[np.median(responses)]
return predictions
def evaluateModelEnsembleUncertainty(ensemble, inputData):
responses=[evaluateGPModel(mod, inputData) for mod in ensemble]
if type(responses[0])==np.ndarray:
responses=np.transpose(responses)
uncertainties=[uncertainty(res,0) for res in responses]
else:
uncertainties=[uncertainty(responses,0)]
return uncertainties
def relativeEnsembleUncertainty(ensemble,inputData):
output=evaluateModelEnsembleUncertainty(ensemble,inputData)
return np.array(output)
def createUncertaintyFunc(ensemble):
return lambda x: -relativeEnsembleUncertainty(ensemble,x)
def maximizeUncertainty(ensemble,varCount,bounds=[]): #Used to select a new point of maximum uncertainty
func=createUncertaintyFunc(ensemble)
x0=[np.mean(bounds[i]) for i in range(varCount)]
if bounds==[]:
pt=minimize(func,x0).x
else:
pt=minimize(func,x0,bounds=bounds).x
return pt
def extendData(data,newPoint):
return np.concatenate((data.T,np.array([newPoint]))).T
def activeLearningCheckpoint(eqNum,version,i,inputData,response,testInput,testResponse,errors,models,minerr):
path=os.path.join(str(eqNum),str(version))
file=open(path,"wb+")
dill.dump([i,inputData,response,testInput,testResponse,errors,models,minerr],file)
file.close()
def activeLearningCheckpointLoad(eqNum,version,i,inputData,response,testInput,testResponse,errors,models,minerr):
path=os.path.join(str(eqNum),str(version))
try:
with open(path,'rb') as f:
i,inputData,response,testInput,testResponse,errors,models,minerr=dill.load(f)
except FileNotFoundError:
return i,inputData,response,testInput,testResponse,errors,models,minerr
return i,inputData,response,testInput,testResponse,errors,models,minerr
def subSampleSpace(space):
newSpace=copy.deepcopy(space)
newSpace=list(newSpace)
for i in range(len(newSpace)):
pts=sorted([np.random.uniform(newSpace[i][0],newSpace[i][1]),np.random.uniform(newSpace[i][0],newSpace[i][1])])
newSpace[i]=tuple(pts)
return tuple(newSpace)
def activeLearning(func, dims, ranges,rangesP,eqNum=1,version=1,iterations=100): #func should be a lamda function of form lambda data: f(data[0],data[1],...)
try:
with open(os.path.join(str(eqNum),str(version))+".txt",'rb') as f:
return -1
except FileNotFoundError:
pass
inputData=[]
testInput=[]
found=False
for i in range(dims):
inputData.append(np.random.uniform(ranges[i][0],ranges[i][1],3))
testInput.append(np.random.uniform(ranges[i][0],ranges[i][1],200))
inputData=np.array(inputData)
testInput=np.array(testInput)
response=func(inputData)
testResponse=func(testInput)
errors=[]
models=[]
minerr=1
for i in range(iterations):
print("input: ",inputData)
print("\n response: ",response)
i,inputData,response,testInput,testResponse,errors,models,minerr=activeLearningCheckpointLoad(eqNum,version,i,inputData,response,testInput,testResponse,errors,models,minerr)
if i>iterations-1:
break
i+=1
models1=evolve(inputData,response,initialPop=models,generations=1000,tracking=False,popSize=300,ops=allOps(),timeLimit=120,capTime=True,align=False,elitismRate=10)
models2=evolve(inputData,response,initialPop=models,generations=1000,tracking=False,popSize=300,ops=allOps(),timeLimit=120,capTime=True,align=False,elitismRate=10)
models3=evolve(inputData,response,initialPop=models,generations=1000,tracking=False,popSize=300,ops=allOps(),timeLimit=120,capTime=True,align=False,elitismRate=10)
models4=evolve(inputData,response,initialPop=models,generations=1000,tracking=False,popSize=300,ops=allOps(),timeLimit=120,capTime=True,align=False,elitismRate=10)
models=models1+models2+models3+models4
models=selectModels(models,20)
alignedModels=[alignGPModel(mods,inputData,response) for mods in models]
ensemble=ensembleSelect(alignedModels,inputData,response)
out=maximizeUncertainty(ensemble,dims,rangesP)
while out in inputData.T:
out=maximizeUncertainty(ensemble,dims,subSampleSpace(rangesP))
inputData=extendData(inputData,out)
response=func(inputData)
fitList=np.array([fitness(mod,testInput,testResponse) for mod in alignedModels])
errors.append(min(fitList[np.logical_not(np.isnan(fitList))]))
minerr=errors[-1]
if minerr<1e-14:
#print("Points needed in round", j,": ",3+i, " Time needed: ", time.perf_counter()-roundTime)
if not os.path.exists(str(eqNum)):
os.makedirs(str(eqNum))
path=os.path.join(str(eqNum),str(version))
file=open(path,"wb+")
dill.dump([i,inputData,response,testInput,testResponse,errors,models,minerr],file)
file.close()
file=open(path+'.txt','w+')
file.write(str(i+3)+'\n')
file.write(str(errors))
file.close()
return 3+i
found=True
ptsNeeded.append(3+i)
break
activeLearningCheckpoint(eqNum,version,i,inputData,response,testInput,testResponse,errors,models,minerr)
if found==False:
#print("Points needed in round",j,": NA (model not found)")
path=os.path.join(str(eqNum),str(version))
file=open(path,"wb")
dill.dump([-1,inputData,response,testInput,testResponse,errors,models,minerr],file)
file.close()
file=open(path+'.txt',"w+")
file.write(str(i+3)+"\n")
file.write(str(errors))
file.close()
return -1
def plotModels(models, modelExpression=False):
tMods=copy.deepcopy(models)
if len(tMods[0][2])<2:
# add complexity as second value
for mod in tMods:
mod[2]=[mod[2][0],stackGPModelComplexity(mod)]
[modelToListForm(mod) for mod in tMods]
paretoModels=paretoTournament(tMods)
for i in paretoModels:
tMods.remove(i)
[modelRestoreForm(mod) for mod in paretoModels]
[modelRestoreForm(mod) for mod in tMods]
pAccuracies=[mod[2][0] for mod in paretoModels]
pComplexities=[mod[2][1] for mod in paretoModels]
accuracies=[mod[2][0] for mod in tMods]+pAccuracies
complexities=[mod[2][1] for mod in tMods]+pComplexities
colors=['blue' for i in range(len(tMods))]+['red' for i in range(len(pAccuracies))]
fig,ax = plt.subplots()
sc=plt.scatter(complexities,accuracies,color=colors)
plt.xlabel("Complexity")
plt.ylabel("1-R**2")
if modelExpression:
names=[str(printGPModel(mod)) for mod in tMods]+[str(printGPModel(mod)) for mod in paretoModels]
else:
names = [str(mod) for mod in tMods]+[str(mod) for mod in paretoModels]
label = ax.annotate("", xy=(0,0), xytext=(np.min(complexities),np.mean([np.max(accuracies),np.min(accuracies)])),
bbox=dict(boxstyle="round", fc="w"),
arrowprops=dict(arrowstyle="->"))
label.set_visible(False)
def update_labels(ind):
pos = sc.get_offsets()[ind["ind"][0]]
label.xy = pos
text = "{}".format(" ".join([names[n] for n in [ind["ind"][0]]]))
label.set_text(text)
label.get_bbox_patch().set_facecolor('grey')
label.get_bbox_patch().set_alpha(0.9)
def hover(event):
vis = label.get_visible()
if event.inaxes == ax:
cont, ind = sc.contains(event)
if cont:
update_labels(ind)
label.set_visible(True)
fig.canvas.draw_idle()
else:
if vis:
label.set_visible(False)
fig.canvas.draw_idle()
fig.canvas.mpl_connect("motion_notify_event", hover)
plt.show()
def plotModelResponseComparison(model,inputData,response,sort=False):
plt.scatter(range(len(response)),response,label="True Response")
plt.scatter(range(len(response)),evaluateGPModel(model,inputData),label="Model Prediction")
plt.legend()
plt.xlabel("Data Index")
plt.ylabel("Response Value")
plt.show()
def plotPredictionResponseCorrelation(model,inputData,response):
plt.scatter(response,evaluateGPModel(model,inputData),label="Model")
plt.plot(response,response,label="Perfect Correlation",color='green')
plt.xlabel("True Response")
plt.ylabel("Predicted Response")
plt.legend()
plt.show()
#Plot model complexity distribution
def plotModelComplexityDistribution(models):
tMods=copy.deepcopy(models)
[modelToListForm(mod) for mod in tMods]
paretoModels=paretoTournament(tMods)
for i in paretoModels:
tMods.remove(i)
[modelRestoreForm(mod) for mod in paretoModels]
[modelRestoreForm(mod) for mod in tMods]
pComplexities=[mod[2][1] for mod in paretoModels]
tComplexities=[mod[2][1] for mod in tMods]
plt.hist(tComplexities,label="Non-Pareto Models")
plt.hist(pComplexities,label="Pareto Models")
plt.xlabel("Model Complexity")
plt.ylabel("Frequency")
plt.legend()
plt.show()
#Plot model accuracy distribution
def plotModelAccuracyDistribution(models):
tMods=copy.deepcopy(models)
[modelToListForm(mod) for mod in tMods]
paretoModels=paretoTournament(tMods)
for i in paretoModels:
tMods.remove(i)
[modelRestoreForm(mod) for mod in paretoModels]
[modelRestoreForm(mod) for mod in tMods]
pAccuracies=[mod[2][0] for mod in paretoModels]
tAccuracies=[mod[2][0] for mod in tMods]
plt.hist(tAccuracies,label="Non-Pareto Models")
plt.hist(pAccuracies,label="Pareto Models")
plt.xlabel("Model Accuracy")
plt.ylabel("Frequency")
plt.legend()
plt.show()
#Plot model residuals relative to response
def plotModelResiduals(model,input,response):
plt.scatter(response,evaluateGPModel(model,input)-response)
plt.xlabel("Response")
plt.ylabel("Residual")
plt.show()
#Plot model residual distribution
def plotModelResidualDistribution(model,input,response):
plt.hist(evaluateGPModel(model,input)-response)
plt.xlabel("Residual")
plt.ylabel("Frequency")
plt.show()
##CAN USE THESE FOR UNCERTAINITY FURTHERMORE
#Plot the presence of variables in a model population
def plotVariablePresence(models,variables=["x"+str(i) for i in range(100)],sort=False):
vars=[varReplace(model[1],variables) for model in models]
#Remove all numeric entries in vars
vars=[[i for i in var if type(i)!=int and type(i)!=float] for var in vars]
#Merge into one list
vars=[j for i in vars for j in i]
#Count frequency of each variable in vars
varFreqs=[vars.count(i) for i in variables]
#Keep only variables that appear at least once
variablesUsed=[variables[i] for i in range(len(varFreqs)) if varFreqs[i]>0]
varFreqs=[varFreqs[i] for i in range(len(varFreqs)) if varFreqs[i]>0]
if sort:
order=np.argsort(varFreqs)[::-1]
variablesUsed=[variablesUsed[i] for i in order]
varFreqs=[varFreqs[i] for i in order]
#Plot variable frequency
plt.bar(variablesUsed,varFreqs)
plt.xlabel("Variable")
plt.ylabel("Frequency")
plt.show()
def replaceOpsWithStrings(opStack):
model = copy.deepcopy(opStack)
model = replaceFunc(model,exp,str("exp"))
model = replaceFunc(model,tan,str("tan"))
model = replaceFunc(model,sqrt,str("sqrt"))
model = replaceFunc(model,inv,str("1/#"))
model = replaceFunc(model,sin,str("sin"))
model = replaceFunc(model,cos,str("cos"))
model = replaceFunc(model,arccos,str("acos"))
model = replaceFunc(model,arcsin,str("asin"))
model = replaceFunc(model,arctan,str("atan"))
model = replaceFunc(model,tanh,str("tanh"))
model = replaceFunc(model,log,str("log"))
model = replaceFunc(model,add,"+")
model = replaceFunc(model,mult,"*")
model = replaceFunc(model,sub,"-")
model = replaceFunc(model,protectDiv,"/")
model = replaceFunc(model,sqrd,"^2")
return model
#Plot the presence of operators in a model population
def plotOperatorPresence(models,sort=False,excludePop=True):
ops=[replaceOpsWithStrings(model[0]) for model in models]
#Merge into one list
ops=[j for i in ops for j in i]
#Remove duplicates in ops
uniqueOps=list(set(ops))
if excludePop:
#Remove pop operator
uniqueOps.remove('pop')
#Count frequency of each operator in ops
opFreqs=[ops.count(i) for i in uniqueOps]
#Keep only operators that appear at least once
opsUsed=[str(uniqueOps[i]) for i in range(len(opFreqs)) if opFreqs[i]>0]
opFreqs=[opFreqs[i] for i in range(len(opFreqs)) if opFreqs[i]>0]
if sort:
order=np.argsort(opFreqs)[::-1]
opsUsed=[opsUsed[i] for i in order]
opFreqs=[opFreqs[i] for i in order]
#Plot operator frequency
plt.bar(opsUsed,opFreqs)
#Rotate x axis labels
plt.xticks(rotation=0)
plt.xlabel("Operator")
plt.ylabel("Frequency")
plt.show()
def diverse_models(best_model, filtered_model_population, trainInput, trainResponse):
###
## Residuals allow us to capture different diverse models since we can fit the models in multiple different ways
###
predicted_data = evaluateGPModel(best_model, trainInput)
candidate_preds = []
for model in filtered_model_population:
candidate_preds.append(np.array(evaluateGPModel(model, trainInput)).flatten())
candidate_preds = np.array(candidate_preds)
trainResponse = np.array(trainResponse).flatten()
best_preds = np.array(evaluateGPModel(best_model, trainInput)).flatten()
selected_preds = [best_preds]
ensemble_cluster = [best_model]
picked_indices = set()
for r in range(10):
#choose up to 10 best models from this ensemble
if len(picked_indices) >= len(filtered_model_population):
break
current_ensemble_mean = np.median(selected_preds, axis=0)
ensemble_residuals = trainResponse - current_ensemble_mean
best_candidate_idx = -1
anti_correlation = float('inf')
for i, cur_model in enumerate(candidate_preds):
if i in picked_indices:continue
cand_residual = trainResponse - cur_model
corr_val = corr = pearsonr(cand_residuals, ensemble_residuals)[0]
if np.nan(corr_val):
corr_val = 1.0
if corr_val < anti_correlation:
anti_correlation = corr_val
best_candidate_idx = i
##add the choosen best model to our current ensemble
picked_indices.add(best_candidate_idx)
selected_preds.append(candidate_preds[best_candidate_idx])
ensemble_cluster.append(filtered_model_population[best_candidate_idx])
print(f"Added Model {best_candidate_idx} (Residual Correlation: {lowest_correlation:.4f})")
return ensemble_cluster
############################
#Sharpness Computations
############################
def sharpnessConstants(model,inputData,responseData,numPerturbations=10,percentPerturbation=0.2):
fits=[]
#For each model parameter, if numeric, randomly perturb by x% and see how much the model changes
for i in range(numPerturbations):
tempModel=copy.deepcopy(model)
newParameters=[param if callable(param) else param*(1+percentPerturbation*(np.random.uniform()-0.5)) for param in model[1]]
tempModel[1]=newParameters
fits.append(fitness(tempModel,inputData,responseData))
return np.std(fits)
def sharpnessData(model,inputData,responseData,numPerturbations=10,percentPerturbation=0.2,preserveSign=False):
fits=[]
#For each vector, randomly perturb by x% of the standard deviation and see how much the model fitness changes
for i in range(numPerturbations):
tempData=copy.deepcopy(inputData)
tempData=np.array([(vec+percentPerturbation*np.std(vec)*(np.random.uniform(size=len(vec))-0.5)) for vec in tempData])
if preserveSign:
signs=[np.unique(var) for var in np.sign(inputData)]
tempData=[signs[i]*abs(tempData[i]) if len(signs[i])==1 else tempData[i] for i in range(len(signs))]
fits.append(fitness(model,tempData,responseData))
return np.std(fits)
def totalSharpness(model,inputData,responseData,numPerturbations=10,percentPerturbation=0.2,preserveSign=False):
return sharpnessConstants(model,inputData,responseData,numPerturbations=numPerturbations,percentPerturbation=percentPerturbation)+sharpnessData(model,inputData,responseData,numPerturbations=numPerturbations,percentPerturbation=percentPerturbation,preserveSign=preserveSign)
############################
#Multiple Independent Searches
############################
def runEpochs(x,y,epochs=5,**kwargs):
models=[]
for i in range(epochs):
models+=evolve(x,y,**kwargs)
return sortModels(models)
############################
#Parallelization
############################
from joblib import Parallel, delayed
def parallelEvolve(*args,n_jobs=-1,avail_cores=-1, **kwargs):
if avail_cores==-1:
try:
avail_cores=len(os.sched_getaffinity(0))
except:
avail_cores=os.cpu_count()
if n_jobs==-1:
try:
n_jobs=len(os.sched_getaffinity(0))
except:
n_jobs=os.cpu_count()
if "tracking" in kwargs and kwargs["tracking"]:
kwargs["returnTracking"]=True
print(f"Running parallel evolution with {n_jobs} jobs.")
if "liveTracking" in kwargs and kwargs["liveTracking"]:
print("Live tracking is not supported in parallel evolution, disabling live tracking.")
kwargs["liveTracking"]=False
runs = Parallel(n_jobs=avail_cores, backend="loky")(delayed(evolve)(*args, **kwargs) for _ in range(n_jobs))
runs, tracking_histories = zip(*runs)
if ("tracking" in kwargs and kwargs["tracking"]):
runs, tracking = zip(*runs)
# plot tracking for each job
plt.figure(figsize=(12, 6))
for i, track in enumerate(tracking):
plt.plot(track, label=f'Job {i+1}')
plt.title('Best Fitness Over Generations for Each Parallel Run')
plt.xlabel('Generations')
plt.ylabel('Best Fitness')
if n_jobs <= 16: # Only show legend if there are a reasonable number of jobs
plt.legend()
plt.show()
flat = [model for sublist in runs for model in sublist]
return sortModels(flat), tracking_histories
############################
#Benchmarking
############################
def generateRandomBenchmark(numVars=5, numSamples=100, noiseLevel=0, opsChoices=defaultOps(), constChoices=defaultConst(), maxLength=10):
# Generate random input data
inputData = np.random.rand(numVars, numSamples)
# Generate a random target function
randomModel = generateRandomModel(numVars, opsChoices, constChoices, maxLength)
# Evaluate the model to get response data
responseData = evaluateGPModel(randomModel, inputData)
# Add noise if specified
if noiseLevel > 0:
noise = np.random.normal(0, noiseLevel, size=responseData.shape)
responseData += noise
return inputData, responseData, randomModel
|