MilesCranmer commited on
Commit
67c22c7
1 Parent(s): 4800983

Tweak PySR demo

Browse files
Files changed (1) hide show
  1. examples/pysr_demo.ipynb +11 -7
examples/pysr_demo.ipynb CHANGED
@@ -990,6 +990,7 @@
990
  ]
991
  },
992
  {
 
993
  "cell_type": "markdown",
994
  "metadata": {
995
  "id": "3hS2kTAbbDhL"
@@ -999,9 +1000,9 @@
999
  "\n",
1000
  "Let's consider a time series problem:\n",
1001
  "\n",
1002
- "$$ z = y^2,\\quad y = \\frac{1}{100} \\sum(y_i),\\quad y_i = x_{i0}^2 + 6 \\cos(2*x_{i2})$$\n",
1003
  "\n",
1004
- "Imagine our time series is 100 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",
1005
  "\n",
1006
  "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",
1007
  "\n",
@@ -1018,7 +1019,7 @@
1018
  "source": [
1019
  "###### np.random.seed(0)\n",
1020
  "N = 100000\n",
1021
- "Nt = 100\n",
1022
  "X = 6 * np.random.rand(N, Nt, 5) - 3\n",
1023
  "y_i = X[..., 0] ** 2 + 6 * np.cos(2 * X[..., 2])\n",
1024
  "y = np.sum(y_i, axis=1) / y_i.shape[1]\n",
@@ -1299,6 +1300,7 @@
1299
  ]
1300
  },
1301
  {
 
1302
  "cell_type": "markdown",
1303
  "metadata": {
1304
  "id": "6WuaeqyqbDhe"
@@ -1306,7 +1308,7 @@
1306
  "source": [
1307
  "Recall we are searching for $y_i$ above:\n",
1308
  "\n",
1309
- "$$ z = y^2,\\quad y = \\frac{1}{100} \\sum(y_i),\\quad y_i = x_{i0}^2 + 6 \\cos(2 x_{i2})$$"
1310
  ]
1311
  },
1312
  {
@@ -1373,11 +1375,13 @@
1373
  },
1374
  "gpuClass": "standard",
1375
  "kernelspec": {
1376
- "display_name": "Python 3",
1377
- "name": "python3"
 
1378
  },
1379
  "language_info": {
1380
- "name": "python"
 
1381
  }
1382
  },
1383
  "nbformat": 4,
 
990
  ]
991
  },
992
  {
993
+ "attachments": {},
994
  "cell_type": "markdown",
995
  "metadata": {
996
  "id": "3hS2kTAbbDhL"
 
1000
  "\n",
1001
  "Let's consider a time series problem:\n",
1002
  "\n",
1003
+ "$$ z = y^2,\\quad y = \\frac{1}{10} \\sum(y_i),\\quad y_i = x_{i0}^2 + 6 \\cos(2*x_{i2})$$\n",
1004
  "\n",
1005
+ "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",
1006
  "\n",
1007
  "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",
1008
  "\n",
 
1019
  "source": [
1020
  "###### np.random.seed(0)\n",
1021
  "N = 100000\n",
1022
+ "Nt = 10\n",
1023
  "X = 6 * np.random.rand(N, Nt, 5) - 3\n",
1024
  "y_i = X[..., 0] ** 2 + 6 * np.cos(2 * X[..., 2])\n",
1025
  "y = np.sum(y_i, axis=1) / y_i.shape[1]\n",
 
1300
  ]
1301
  },
1302
  {
1303
+ "attachments": {},
1304
  "cell_type": "markdown",
1305
  "metadata": {
1306
  "id": "6WuaeqyqbDhe"
 
1308
  "source": [
1309
  "Recall we are searching for $y_i$ above:\n",
1310
  "\n",
1311
+ "$$ z = y^2,\\quad y = \\frac{1}{10} \\sum(y_i),\\quad y_i = x_{i0}^2 + 6 \\cos(2 x_{i2})$$"
1312
  ]
1313
  },
1314
  {
 
1375
  },
1376
  "gpuClass": "standard",
1377
  "kernelspec": {
1378
+ "display_name": "Python (main_ipynb)",
1379
+ "language": "python",
1380
+ "name": "main_ipynb"
1381
  },
1382
  "language_info": {
1383
+ "name": "python",
1384
+ "version": "3.10.9"
1385
  }
1386
  },
1387
  "nbformat": 4,