Create 2.1-xtal2png-cnn-regression.ipynb

@faris-k, just leftover from before I messaged you. No need to use this, but the final presentation will probably be in notebook form

notebooks/2.1-xtal2png-cnn-regression.ipynb

@@ -0,0 +1,81 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "\"\"\"Modified from https://skorch.readthedocs.io/en/stable/user/quickstart.html\"\"\"\n",
+    "from skorch.regressor import NeuralNetRegressor\n",
+    "import numpy as np\n",
+    "from sklearn.datasets import make_classification\n",
+    "from torch import nn\n",
+    "import torch.nn.functional as F\n",
+    "import torch\n",
+    "import torch.nn as nn\n",
+    "import torch.optim as optim\n",
+    "from sklearn.metrics import mean_absolute_error, mean_squared_error\n",
+    "from skorch.callbacks import EarlyStopping\n",
+    "\n",
+    "\n",
+    "# Regular PyTorch Module\n",
+    "class ConvNet(torch.nn.Module):\n",
+    "    def __init__(self, in_channels, num_classes):\n",
+    "        super().__init__()\n",
+    "\n",
+    "        # num_classes is used by the corn loss function\n",
+    "        self.num_classes = num_classes\n",
+    "\n",
+    "        # Initialize CNN layers\n",
+    "        all_layers = [\n",
+    "            torch.nn.Conv2d(in_channels=in_channels, out_channels=3, \n",
+    "                            kernel_size=(3, 3), stride=(1, 1), \n",
+    "                            padding=1),\n",
+    "            torch.nn.MaxPool2d(kernel_size=(2, 2),  stride=(2, 2)),\n",
+    "            torch.nn.Conv2d(in_channels=3, out_channels=6, \n",
+    "                            kernel_size=(3, 3), stride=(1, 1), \n",
+    "                            padding=1),\n",
+    "            torch.nn.MaxPool2d(kernel_size=(2, 2),  stride=(2, 2)),\n",
+    "            torch.nn.Flatten()\n",
+    "        ]\n",
+    "        \n",
+    "        # CORN output layer --------------------------------------\n",
+    "        # Regular classifier would use num_classes instead of \n",
+    "        # num_classes-1 below\n",
+    "        output_layer = torch.nn.Linear(294, num_classes-1)\n",
+    "        # ---------------------------------------------------------\n",
+    "        \n",
+    "        all_layers.append(output_layer)\n",
+    "        self.model = torch.nn.Sequential(*all_layers)\n",
+    "        \n",
+    "    def forward(self, x):\n",
+    "        x = self.model(x)\n",
+    "        return x\n",
+    "\n",
+    "net = NeuralNetRegressor(\n",
+    "    ConvNet(dropout=0.1),\n",
+    "    criterion=nn.MSELoss,\n",
+    "    max_epochs=100,\n",
+    "    optimizer=optim.Adam,\n",
+    "    optimizer__lr = .005,\n",
+    "    callbacks=[EarlyStopping()],\n",
+    ")\n",
+    "\n",
+    "net.fit(X_train, y_train)\n",
+    "y_pred = net.predict(X_val)\n",
+    "\n",
+    "mae = mean_absolute_error(y_val, y_pred)\n",
+    "rmse = mean_squared_error(y_val, y_pred, squared=False)"
+   ]
+  }
+ ],
+ "metadata": {
+  "language_info": {
+   "name": "python"
+  },
+  "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}