Spaces:
Sleeping
Sleeping
File size: 130,151 Bytes
8ea65ad 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 7055fde 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 7055fde 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 7055fde 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 7055fde 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 7de1c20 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 2ddf652 6fdc3b9 8ea65ad |
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 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 |
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "DS4E1PagbDgL"
},
"source": [
"# Setup"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "tQ1r1bbb0yBv"
},
"source": [
"\n",
"## Instructions\n",
"1. Work on a copy of this notebook: _File_ > _Save a copy in Drive_ (you will need a Google account).\n",
"2. (Optional) If you would like to do the deep learning component of this tutorial, turn on the GPU with Edit->Notebook settings->Hardware accelerator->GPU\n",
"3. Execute the following cell (click on it and press Ctrl+Enter) to install Julia. This may take a minute or so.\n",
"4. Continue to the next section.\n",
"\n",
"_Notes_:\n",
"* If your Colab Runtime gets reset (e.g., due to inactivity), repeat steps 3, 4.\n",
"* After installation, if you want to change the Julia version or activate/deactivate the GPU, you will need to reset the Runtime: _Runtime_ > _Delete and disconnect runtime_ and repeat steps 2-4."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "COndi88gbDgO"
},
"source": [
"**Run the following code to install Julia**"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "GIeFXS0F0zww",
"outputId": "5399ed75-f77f-47c5-e53b-4b2f231f2839"
},
"outputs": [],
"source": [
"!curl -fsSL https://install.julialang.org | sh -s -- -y --default-channel 1.10"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "Iu9X-Y-YNmwM",
"outputId": "ee14af65-043a-4ad6-efa0-3cdcc48a4eb8"
},
"outputs": [],
"source": [
"# Make julia available on PATH:\n",
"!ln -s $HOME/.juliaup/bin/julia /usr/local/bin/julia\n",
"\n",
"# Test it works:\n",
"!julia --version"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "ORv1c6xvbDgV"
},
"source": [
"Install PySR"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "EhMRSZEYFPLz",
"outputId": "e3aad3cb-d921-473e-b77b-8fa6a3a9e2e8"
},
"outputs": [],
"source": [
"!pip install pysr && python -m pysr install"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "etTMEV0wDqld"
},
"source": [
"Colab's printing is non-standard, so we need to manually initialize Julia and redirect its printing. Normally, however, this is not required, and PySR will automatically start Julia during the first call to `.fit`:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "j666aOI8xWF_"
},
"outputs": [],
"source": [
"def init_colab_printing():\n",
" from pysr.julia_helpers import init_julia\n",
" from julia.tools import redirect_output_streams\n",
"\n",
" julia_kwargs = dict(optimize=3, threads=\"auto\", compiled_modules=False)\n",
" init_julia(julia_kwargs=julia_kwargs)\n",
" redirect_output_streams()\n",
"\n",
"\n",
"init_colab_printing()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "qeCPKd9wldEK"
},
"source": [
"Now, let's import all of our libraries:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "vFpyRxmhFqeH"
},
"outputs": [],
"source": [
"import sympy\n",
"import numpy as np\n",
"from matplotlib import pyplot as plt\n",
"from pysr import PySRRegressor\n",
"from sklearn.model_selection import train_test_split"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "gsRMQ7grbDga"
},
"source": [
"# Simple PySR example:\n"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "myTEwdiUFiGL"
},
"source": [
"First, let's learn a simple function\n",
"\n",
"$$2.5382 \\cos(x3) + x0^2 - 2$$"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "Cb1eb2XuFQh8"
},
"outputs": [],
"source": [
"# Dataset\n",
"np.random.seed(0)\n",
"X = 2 * np.random.randn(100, 5)\n",
"y = 2.5382 * np.cos(X[:, 3]) + X[:, 0] ** 2 - 2"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "cturCkaVjzLs"
},
"source": [
"By default, we will set up 30 populations of expressions (which evolve independently except for migrations), use 4 threads, and use `\"best\"` for our model selection strategy:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "4nDAAnisdhTc"
},
"outputs": [],
"source": [
"default_pysr_params = dict(\n",
" populations=30,\n",
" model_selection=\"best\",\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "N4gANfkaj8ie"
},
"source": [
"PySR can run for arbitrarily long, and continue to find more and more accurate expressions. You can set the total number of cycles of evolution with `niterations`, although there are also a [few more ways](https://github.com/MilesCranmer/PySR/pull/134) to stop execution.\n",
"\n",
"**This first execution will take a bit longer to startup, as the library is JIT-compiled. The next execution will be much faster.**"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 1000
},
"id": "p4PSrO-NK1Wa",
"outputId": "55910ab3-895d-400b-e9ce-c75aef639c68"
},
"outputs": [],
"source": [
"# Learn equations\n",
"model = PySRRegressor(\n",
" niterations=30,\n",
" binary_operators=[\"+\", \"*\"],\n",
" unary_operators=[\"cos\", \"exp\", \"sin\"],\n",
" **default_pysr_params\n",
")\n",
"\n",
"model.fit(X, y)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "-bsAECbdkQsQ"
},
"source": [
"We can print the model, which will print out all the discovered expressions:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 252
},
"id": "4HR8gknlZz4W",
"outputId": "496283bd-a743-4cc6-a2f9-9619ba91d870"
},
"outputs": [],
"source": [
"model"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "ME3ddPxXkWQg"
},
"source": [
"We can also view the SymPy format of the best expression:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 38
},
"id": "IQKOohdpztS7",
"outputId": "0e7d058a-cce1-45ae-db94-6625f7e53a06"
},
"outputs": [],
"source": [
"model.sympy()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "EHIIPlmClltn"
},
"source": [
"We can also view the SymPy of any other expression in the list, using the index of it in `model.equations_`."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 39
},
"id": "GRcxq-TTlpRX",
"outputId": "50bda367-1ed1-4860-8fcf-c940f2e4d935"
},
"outputs": [],
"source": [
"model.sympy(2)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "YMugcGX4tbqj"
},
"source": [
"## Output"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "gIWt5wz5cjXE"
},
"source": [
"`model.equations_` is a Pandas DataFrame. We can export the results in various ways:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 35
},
"id": "HFGaNL6tbDgi",
"outputId": "0f364da5-e18d-4e31-cadf-087d641a3aed"
},
"outputs": [],
"source": [
"model.latex()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "4hS8kqutcmPQ"
},
"source": [
"These is also `model.sympy(), model.jax(), model.pytorch()`. All of these can take an index as input, to get the result for an arbitrary equation in the list.\n",
"\n",
"We can also use `model.predict` for arbitrary equations, with the default equation being the one chosen by `model_selection`:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "Vbz4IMsk2NYH",
"outputId": "361d4b6e-ac23-479d-b511-5001af05ca43"
},
"outputs": [],
"source": [
"ypredict = model.predict(X)\n",
"ypredict_simpler = model.predict(X, 2)\n",
"\n",
"print(\"Default selection MSE:\", np.power(ypredict - y, 2).mean())\n",
"print(\"Manual selection MSE for index 2:\", np.power(ypredict_simpler - y, 2).mean())"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "SQDUScGebDgr"
},
"source": [
"# Custom operators"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "qvgVbOoSFtQY"
},
"source": [
"A full list of operators is given here: https://astroautomata.com/PySR/operators,\n",
"but we can also use any binary or unary operator in `julia`, or define our own as arbitrary functions.\n",
"\n",
"Say that we want a command to do quartic powers:\n",
"\n",
"$$ y = x_0^4 - 2 $$"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "JvXOVqSyFsdr"
},
"outputs": [],
"source": [
"y = X[:, 0] ** 4 - 2"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "-zoqaL8KGSK5"
},
"source": [
"We can do this by passing a string in Julia syntax.\n",
"\n",
"We also define the operator in sympy, with `extra_sympy_mappings`, to enable its use in `predict`, and other export functions."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 339
},
"id": "PoEkpvYuGUdy",
"outputId": "02834373-a054-400b-8247-2bf33a5c5beb"
},
"outputs": [],
"source": [
"model = PySRRegressor(\n",
" niterations=5,\n",
" populations=40,\n",
" binary_operators=[\"+\", \"*\"],\n",
" unary_operators=[\"cos\", \"exp\", \"sin\", \"quart(x) = x^4\"],\n",
" extra_sympy_mappings={\"quart\": lambda x: x**4},\n",
")\n",
"model.fit(X, y)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 38
},
"id": "emn2IajKbDgy",
"outputId": "11d5d3cf-de43-4f2b-f653-30016e09bdd0"
},
"outputs": [],
"source": [
"model.sympy()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "wbWHyOjl2_kX"
},
"source": [
"Since `quart` is arguably more complex than the other operators, you can also give it a different complexity, using, e.g., `complexity_of_operators={\"quart\": 2}` to give it a complexity of 2 (instead of the default 2). You can also define custom complexities for variables and constants (`complexity_of_variables` and `complexity_of_constants`, respectively - both take a single number).\n",
"\n",
"\n",
"One can also add a binary operator, with, e.g., `\"myoperator(x, y) = x^2 * y\"`. All Julia operators that work on scalar 32-bit floating point values are available.\n",
"\n",
"Make sure that any operator you add is valid over the real line. So, e.g., you will need to define `\"mysqrt(x) = sqrt(abs(x))\"` to enable it for negative numbers,\n",
"or, simply have it return a very large number for bad inputs (to prevent negative input in a soft way):\n",
"`\"mysqrt(x::T) where {T} = (x >= 0) ? x : T(-1e9)\"` (Julia syntax for a template function of input type `T`), which will make `mysqrt(x)` return -10^9 for negative x–hurting the loss of the equation."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "pEXT4xskbDg0"
},
"source": [
"## Scoring"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "IyeYbVVOG60w"
},
"source": [
"Using `model_selection=\"best\"`selects the equation with the max score and prints it. But in practice it is best to look through all the equations manually, select an equation above some MSE threshold, and then use the score to select among that loss threshold.\n",
"\n",
"Here, \"score\" is defined by:\n",
"$$ \\text{score} = - \\log(\\text{loss}_i/\\text{loss}_{i-1})/\n",
"(\\text{complexity}_i - \\text{complexity}_{i-1})$$"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "I3IxmvSQrhfw"
},
"source": [
"This scoring is motivated by the common strategy of looking for drops in the loss-complexity curve.\n",
"\n",
"From Schmidt & Lipson (2009) -"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "eUeXyoLxrd8o"
},
"source": [
""
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "gDZyxsA7bDg9"
},
"source": [
"# Noise example"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "cJCHdDt6IOou"
},
"source": [
"Here is an example with noise. Known Gaussian noise with $\\sigma$ between 0.1 and 5.0. We record samples of $y$:\n",
"\n",
"$$ \\sigma \\sim U(0.1, 5.0) $$\n",
"$$ \\epsilon \\sim \\mathcal{N}(0, \\sigma^2)$$\n",
"$$ y = 5\\;\\cos(3.5 x_0) - 1.3 + \\epsilon.$$\n",
"We have 5 features, say. The weights change the loss function to be:\n",
"$$MSE = \\sum [(y - f(x))^2*w],$$\n",
"\n",
"so in this example, we can set:\n",
"$$w = 1/\\sigma^2.$$"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "up1RvmwyOdal"
},
"outputs": [],
"source": [
"np.random.seed(0)\n",
"N = 3000\n",
"upper_sigma = 5\n",
"X = 2 * np.random.rand(N, 5)\n",
"sigma = np.random.rand(N) * (5 - 0.1) + 0.1\n",
"eps = sigma * np.random.randn(N)\n",
"y = 5 * np.cos(3.5 * X[:, 0]) - 1.3 + eps"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "-EJPDZbP5YEZ"
},
"source": [
"Let's look at this dataset:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/",
"height": 467
},
"id": "sqMqb4nJ5ZR5",
"outputId": "aa24922b-2395-4e00-dce3-268fc8e603dc"
},
"outputs": [],
"source": [
"plt.scatter(X[:, 0], y, alpha=0.2)\n",
"plt.xlabel(\"$x_0$\")\n",
"plt.ylabel(\"$y$\")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "kaddasbBuDDv"
},
"source": [
"Define some weights to use:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "3wqz9_sIbDhA"
},
"outputs": [],
"source": [
"weights = 1 / sigma**2"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "v8WBYtcZbDhC",
"outputId": "37d4002f-e9d6-40c0-9a24-c671d9c384e6"
},
"outputs": [],
"source": [
"weights[:5]"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "NXWdQSCFuAzV"
},
"source": [
"Let's run PySR again:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"colab": {
"base_uri": "https://localhost:8080/"
},
"id": "a07K3KUjOxcp",
"outputId": "41d11915-78b7-4446-c153-b92a5e2abd4c"
},
"outputs": [],
"source": [
"model = PySRRegressor(\n",
" loss=\"myloss(x, y, w) = w * abs(x - y)\", # Custom loss function with weights.\n",
" niterations=20,\n",
" populations=20, # Use more populations\n",
" binary_operators=[\"+\", \"*\"],\n",
" unary_operators=[\"cos\"],\n",
")\n",
"model.fit(X, y, weights=weights)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "CHCMO9CouFLP"
},
"source": [
"Let's see if we get similar results to the true equation"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "oHyUbcg6ggmx"
},
"outputs": [],
"source": [
"model"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "OchlZZQP8Ums"
},
"source": [
"We can also filter all equations up to 2x the most accurate equation, then select the best score from that list:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "PB67POLr8b_L"
},
"outputs": [],
"source": [
"best_idx = model.equations_.query(\n",
" f\"loss < {2 * model.equations_.loss.min()}\"\n",
").score.idxmax()\n",
"model.sympy(best_idx)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "SRHTP4x55roh"
},
"source": [
"We can also use `denoise=True`, which will run the input through a Gaussian process to denoise the dataset, before fitting on it."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "eTGQ4NA78yAw"
},
"source": [
"Let's look at the fit:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "ezCC0IkS8zFf"
},
"outputs": [],
"source": [
"plt.scatter(X[:, 0], y, alpha=0.1)\n",
"y_prediction = model.predict(X, index=best_idx)\n",
"plt.scatter(X[:, 0], y_prediction)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "2x-8M8W4G-KM"
},
"source": [
"# Multiple outputs"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "LIJcWqBQG-KM"
},
"source": [
"For multiple outputs, multiple equations are returned:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "_Aar1ZJwG-KM"
},
"outputs": [],
"source": [
"X = 2 * np.random.randn(100, 5)\n",
"y = 1 / X[:, [0, 1, 2]]"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "9Znwq40PG-KM"
},
"outputs": [],
"source": [
"model = PySRRegressor(\n",
" binary_operators=[\"+\", \"*\"],\n",
" unary_operators=[\"inv(x) = 1/x\"],\n",
" extra_sympy_mappings={\"inv\": lambda x: 1 / x},\n",
")\n",
"model.fit(X, y)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "0Y_vy0sqG-KM"
},
"outputs": [],
"source": [
"model"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "-UP49CsGG-KN"
},
"source": [
"# Julia packages and types"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "tOdNHheUG-KN"
},
"source": [
"PySR uses [SymbolicRegression.jl](https://github.com/MilesCranmer/SymbolicRegression.jl)\n",
"as its search backend. This is a pure Julia package, and so can interface easily with any other\n",
"Julia package.\n",
"For some tasks, it may be necessary to load such a package.\n",
"\n",
"For example, let's say we wish to discovery the following relationship:\n",
"\n",
"$$ y = p_{3x + 1} - 5, $$\n",
"\n",
"where $p_i$ is the $i$th prime number, and $x$ is the input feature.\n",
"\n",
"Let's see if we can discover this using\n",
"the [Primes.jl](https://github.com/JuliaMath/Primes.jl) package.\n",
"\n",
"First, let's get the Julia backend\n",
"Here, we might choose to manually specify unlimited threads, `-O3`,\n",
"and `compile_modules=False`, although this will only propagate if Julia has not yet started:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "yUC4BMuHG-KN"
},
"outputs": [],
"source": [
"import pysr\n",
"\n",
"jl = pysr.julia_helpers.init_julia(\n",
" julia_kwargs=dict(optimize=3, threads=\"auto\", compiled_modules=False)\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "af07m4uBG-KN"
},
"source": [
"\n",
"\n",
"`jl` stores the Julia runtime.\n",
"\n",
"Now, let's run some Julia code to add the Primes.jl\n",
"package to the PySR environment:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "xBlMY-s4G-KN"
},
"outputs": [],
"source": [
"jl.eval(\n",
" \"\"\"\n",
"import Pkg\n",
"Pkg.add(\"Primes\")\n",
"\"\"\"\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "1rJFukD6G-KN"
},
"source": [
"This imports the Julia package manager, and uses it to install\n",
"`Primes.jl`. Now let's import `Primes.jl`:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "1PQl1rIaG-KN"
},
"outputs": [],
"source": [
"jl.eval(\"import Primes\")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "edGdMxKnG-KN"
},
"source": [
"\n",
"Now, we define a custom operator:\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "9Ut3HcW3G-KN"
},
"outputs": [],
"source": [
"jl.eval(\n",
" \"\"\"\n",
"function p(i::T) where T\n",
" if 0.5 < i < 1000\n",
" return T(Primes.prime(round(Int, i)))\n",
" else\n",
" return T(NaN)\n",
" end\n",
"end\n",
"\"\"\"\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "_wcV8889G-KN"
},
"source": [
"\n",
"We have created a function `p`, which takes a number `i` of type `T` (e.g., `T=Float64`).\n",
"`p` first checks whether the input is between 0.5 and 1000.\n",
"If out-of-bounds, it returns `NaN`.\n",
"If in-bounds, it rounds it to the nearest integer, computes the corresponding prime number, and then\n",
"converts it to the same type as input.\n",
"\n",
"The equivalent function in Python would be:\n",
"\n",
"```python\n",
"import sympy\n",
"\n",
"def p(i):\n",
" if 0.5 < i < 1000:\n",
" return float(sympy.prime(int(round(i))))\n",
" else:\n",
" return float(\"nan\")\n",
"```\n",
"\n",
"(However, note that this version assumes 64-bit float input, rather than any input type `T`)\n",
"\n",
"Next, let's generate a list of primes for our test dataset.\n",
"Since we are using PyJulia, we can just call `p` directly to do this:\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "giqwisEPG-KN"
},
"outputs": [],
"source": [
"primes = {i: jl.p(i * 1.0) for i in range(1, 999)}"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "MPAqARj6G-KO"
},
"source": [
"Next, let's use this list of primes to create a dataset of $x, y$ pairs:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "jab4tRRRG-KO"
},
"outputs": [],
"source": [
"import numpy as np\n",
"\n",
"X = np.random.randint(0, 100, 100)[:, None]\n",
"y = [primes[3 * X[i, 0] + 1] - 5 + np.random.randn() * 0.001 for i in range(100)]"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "3eFgWrjcG-KO"
},
"source": [
"Note that we have also added a tiny bit of noise to the dataset.\n",
"\n",
"Finally, let's create a PySR model, and pass the custom operator. We also need to define the sympy equivalent, which we can leave as a placeholder for now:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "pEYskM2_G-KO"
},
"outputs": [],
"source": [
"from pysr import PySRRegressor\n",
"import sympy\n",
"\n",
"\n",
"class sympy_p(sympy.Function):\n",
" pass\n",
"\n",
"\n",
"model = PySRRegressor(\n",
" binary_operators=[\"+\", \"-\", \"*\", \"/\"],\n",
" unary_operators=[\"p\"],\n",
" niterations=20,\n",
" extra_sympy_mappings={\"p\": sympy_p},\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "ee30bd41"
},
"source": [
"We are all set to go! Let's see if we can find the true relation:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "li-TB19iG-KO"
},
"outputs": [],
"source": [
"model.fit(X, y)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "jwhTWZryG-KO"
},
"source": [
"if all works out, you should be able to see the true relation (note that the constant offset might not be exactly 1, since it is allowed to round to the nearest integer).\n",
"\n",
"You can get the sympy version of the best equation with:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "bSlpX9xAG-KO"
},
"outputs": [],
"source": [
"model.sympy()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "cPc1EDvRbDhL"
},
"source": [
"# High-dimensional input: Neural Nets + Symbolic Regression"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "3hS2kTAbbDhL"
},
"source": [
"In this example, let's learn a high-dimensional problem. **This will use the method proposed in our NeurIPS paper: https://arxiv.org/abs/2006.11287.**\n",
"\n",
"Let's consider a time series problem:\n",
"\n",
"$$ z = y^2,\\quad y = \\frac{1}{10} \\sum(y_i),\\quad y_i = x_{i0}^2 + 6 \\cos(2*x_{i2})$$\n",
"\n",
"Imagine our time series is 10 timesteps. That is very hard for symbolic regression, even if we impose the inductive bias of $$z=f(\\sum g(x_i))$$ - it is the square of the number of possible equations!\n",
"\n",
"But, as in our paper, **we can break this problem down into parts with a neural network. Then approximate the neural network with the symbolic regression!**\n",
"\n",
"Then, instead of, say, $(10^9)^2=10^{18}$ equations, we only have to consider $2\\times 10^9$ equations."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "SXJGXySlbDhL"
},
"outputs": [],
"source": [
"import numpy as np\n",
"\n",
"rstate = np.random.RandomState(0)\n",
"\n",
"N = 100000\n",
"Nt = 10\n",
"X = 6 * rstate.rand(N, Nt, 5) - 3\n",
"y_i = X[..., 0] ** 2 + 6 * np.cos(2 * X[..., 2])\n",
"y = np.sum(y_i, axis=1) / y_i.shape[1]\n",
"z = y**2\n",
"X.shape, y.shape"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "8ZqGupq_uSgp"
},
"source": [
"## Neural Network definition"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "r2NR0h8-bDhN"
},
"source": [
"So, as described above, let's first use a neural network with the sum inductive bias to solve this problem.\n",
"\n",
"Essentially, we will learn two neural networks:\n",
"- `f`\n",
"- `g`\n",
"\n",
"each defined as a multi-layer perceptron. We will sum over `g` the same way as in our equation, but we won't define the summed part beforehand.\n",
"\n",
"Then, we will fit `g` and `f` **separately** using symbolic regression."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "aca54ffa"
},
"source": [
"> **Warning**\n",
">\n",
"> We import torch *after* already starting PyJulia. This is required due to interference between their C bindings. If you use torch, and then run PyJulia, you will likely hit a segfault. So keep this in mind for mixed deep learning + PyJulia/PySR workflows."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "k-Od8b9DlkHK"
},
"outputs": [],
"source": [
"!pip install pytorch_lightning"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "nWVfkV_YbDhO"
},
"outputs": [],
"source": [
"import torch\n",
"from torch import nn, optim\n",
"from torch.nn import functional as F\n",
"from torch.utils.data import DataLoader, TensorDataset\n",
"import pytorch_lightning as pl\n",
"\n",
"hidden = 128\n",
"total_steps = 50_000\n",
"\n",
"\n",
"def mlp(size_in, size_out, act=nn.ReLU):\n",
" return nn.Sequential(\n",
" nn.Linear(size_in, hidden),\n",
" act(),\n",
" nn.Linear(hidden, hidden),\n",
" act(),\n",
" nn.Linear(hidden, hidden),\n",
" act(),\n",
" nn.Linear(hidden, size_out),\n",
" )\n",
"\n",
"\n",
"class SumNet(pl.LightningModule):\n",
" def __init__(self):\n",
" super().__init__()\n",
"\n",
" ########################################################\n",
" # The same inductive bias as above!\n",
" self.g = mlp(5, 1)\n",
" self.f = mlp(1, 1)\n",
"\n",
" def forward(self, x):\n",
" y_i = self.g(x)[:, :, 0]\n",
" y = torch.sum(y_i, dim=1, keepdim=True) / y_i.shape[1]\n",
" z = self.f(y)\n",
" return z[:, 0]\n",
"\n",
" ########################################################\n",
"\n",
" # PyTorch Lightning bookkeeping:\n",
" def training_step(self, batch, batch_idx):\n",
" x, z = batch\n",
" predicted_z = self(x)\n",
" loss = F.mse_loss(predicted_z, z)\n",
" return loss\n",
"\n",
" def validation_step(self, batch, batch_idx):\n",
" return self.training_step(batch, batch_idx)\n",
"\n",
" def configure_optimizers(self):\n",
" optimizer = torch.optim.Adam(self.parameters(), lr=self.max_lr)\n",
" scheduler = {\n",
" \"scheduler\": torch.optim.lr_scheduler.OneCycleLR(\n",
" optimizer,\n",
" max_lr=self.max_lr,\n",
" total_steps=self.trainer.estimated_stepping_batches,\n",
" final_div_factor=1e4,\n",
" ),\n",
" \"interval\": \"step\",\n",
" }\n",
" return [optimizer], [scheduler]"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "kK725aSEuUvG"
},
"source": [
"## Data bookkeeping"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "KdWVtWUcbDhQ"
},
"source": [
"Put everything into PyTorch and do a train/test split:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "0ym19abgbDhR"
},
"outputs": [],
"source": [
"from multiprocessing import cpu_count\n",
"\n",
"Xt = torch.tensor(X).float()\n",
"zt = torch.tensor(z).float()\n",
"X_train, X_test, z_train, z_test = train_test_split(Xt, zt, random_state=0)\n",
"train_set = TensorDataset(X_train, z_train)\n",
"train = DataLoader(\n",
" train_set, batch_size=128, num_workers=cpu_count(), shuffle=True, pin_memory=True\n",
")\n",
"test_set = TensorDataset(X_test, z_test)\n",
"test = DataLoader(test_set, batch_size=256, num_workers=cpu_count(), pin_memory=True)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "3dw_NefuudIq"
},
"source": [
"## Train the model with PyTorch Lightning on GPUs:"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "hhlhLQUBbDhT"
},
"source": [
"Start the model:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "1ldN0999bDhU"
},
"outputs": [],
"source": [
"pl.seed_everything(0)\n",
"model = SumNet()\n",
"model.total_steps = total_steps\n",
"model.max_lr = 1e-2"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "WWRsu5A9bDhW"
},
"source": [
"PyTorch Lightning trainer object:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "33R2nrv-b62w"
},
"outputs": [],
"source": [
"trainer = pl.Trainer(max_steps=total_steps, accelerator=\"gpu\", devices=1)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "jh91CukM5CkI"
},
"source": [
"Here, we fit the neural network:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "TXZdF8k1bDhY"
},
"outputs": [],
"source": [
"trainer.fit(model, train_dataloaders=train, val_dataloaders=test)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "uYzk0yU4ulfH"
},
"source": [
"## Latent vectors of network\n",
"\n",
"Let's get the input and output of the learned `g` function from the network over some random data:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "s2sQLla5bDhb"
},
"outputs": [],
"source": [
"np.random.seed(0)\n",
"idx = np.random.randint(0, 10000, size=1000)\n",
"\n",
"X_for_pysr = Xt[idx]\n",
"y_i_for_pysr = model.g(X_for_pysr)[:, :, 0]\n",
"y_for_pysr = torch.sum(y_i_for_pysr, dim=1) / y_i_for_pysr.shape[1]\n",
"z_for_pysr = zt[idx] # Use true values.\n",
"\n",
"X_for_pysr.shape, y_i_for_pysr.shape"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "nCCIvvAGuyFi"
},
"source": [
"## Learning over the network:\n",
"\n",
"Now, let's fit `g` using PySR.\n",
"\n",
"> **Warning**\n",
">\n",
"> First, let's save the data, because sometimes PyTorch and PyJulia's C bindings interfere and cause the colab kernel to crash. If we need to restart, we can just load the data without having to retrain the network:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "UX7Am6mZG-KT"
},
"outputs": [],
"source": [
"nnet_recordings = {\n",
" \"g_input\": X_for_pysr.detach().cpu().numpy().reshape(-1, 5),\n",
" \"g_output\": y_i_for_pysr.detach().cpu().numpy().reshape(-1),\n",
" \"f_input\": y_for_pysr.detach().cpu().numpy().reshape(-1, 1),\n",
" \"f_output\": z_for_pysr.detach().cpu().numpy().reshape(-1),\n",
"}\n",
"\n",
"# Save the data for later use:\n",
"import pickle as pkl\n",
"\n",
"with open(\"nnet_recordings.pkl\", \"wb\") as f:\n",
" pkl.dump(nnet_recordings, f)"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "krhaNlwFG-KT"
},
"source": [
"We can now load the data, including after a crash (be sure to re-run the import cells at the top of this notebook, including the one that starts PyJulia)."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "NF9aSFXHG-KT"
},
"outputs": [],
"source": [
"import pickle as pkl\n",
"\n",
"nnet_recordings = pkl.load(open(\"nnet_recordings.pkl\", \"rb\"))\n",
"f_input = nnet_recordings[\"f_input\"]\n",
"f_output = nnet_recordings[\"f_output\"]\n",
"g_input = nnet_recordings[\"g_input\"]\n",
"g_output = nnet_recordings[\"g_output\"]"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "_hTYHhDGG-KT"
},
"source": [
"And now fit using a subsample of the data (symbolic regression only needs a small sample to find the best equation):"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "51QdHVSkbDhc"
},
"outputs": [],
"source": [
"rstate = np.random.RandomState(0)\n",
"f_sample_idx = rstate.choice(f_input.shape[0], size=500, replace=False)\n",
"\n",
"model = PySRRegressor(\n",
" niterations=50,\n",
" binary_operators=[\"+\", \"-\", \"*\"],\n",
" unary_operators=[\"cos\", \"square\"],\n",
")\n",
"model.fit(g_input[f_sample_idx], g_output[f_sample_idx])"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "1a738a33"
},
"source": [
"If this segfaults, restart the notebook, and run the initial imports and PyJulia part, but skip the PyTorch training. This is because PyTorch's C binding tends to interefere with PyJulia. You can then re-run the `pkl.load` cell to import the data."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "xginVMmTu3MZ"
},
"source": [
"## Validation"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "6WuaeqyqbDhe"
},
"source": [
"Recall we are searching for $f$ and $g$ such that:\n",
"$$z=f(\\sum g(x_i))$$\n",
"which approximates the true relation:\n",
"$$ z = y^2,\\quad y = \\frac{1}{10} \\sum(y_i),\\quad y_i = x_{i0}^2 + 6 \\cos(2 x_{i2})$$\n",
"\n",
"Let's see how well we did in recovering $g$:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "E1_VWQ45bDhf"
},
"outputs": [],
"source": [
"model.equations_[[\"complexity\", \"loss\", \"equation\"]]"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "mlU1hidZkgCY"
},
"source": [
"A neural network can easily undo a linear transform (which commutes with the summation), so any affine transform in $g$ is to be expected. The network for $f$ has learned to undo the linear transform.\n",
"\n",
"This likely won't find the exact result, but it should find something similar. You may wish to try again but with many more `total_steps` for the neural network (10,000 is quite small!).\n",
"\n",
"Then, we can learn another analytic equation for $f$."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "TntGlQEwbDhk"
},
"source": [
"**Now, we can compose these together to get the time series model!**\n",
"\n",
"Think about what we just did: we found an analytical equation for $z$ in terms of $500$ datapoints, under the assumption that $z$ is a function of a sum of another function over an axis:\n",
"\n",
"$$ z = f(\\sum_i g(x_i)) $$\n",
"\n",
"And we pulled out analytical copies for $g$ using symbolic regression."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "1QsHVjAVbDhk"
},
"source": [
"# Other PySR Options"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "S5dO61g1bDhk"
},
"source": [
"The full list of PySR parameters can be found here: https://astroautomata.com/PySR/api"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"provenance": []
},
"gpuClass": "standard",
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
|