Add skorch

book/machinelearning/modeltraining.ipynb

@@ -1644,6 +1644,87 @@
     "results = tuner.fit()\n",
     "print(results.get_best_result(metric=\"mean_accuracy\", mode=\"max\").config)"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "62334c73",
+   "metadata": {},
+   "source": [
+    "## Use PyTorch with scikit-learn API with `skorch`"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "4ac45c78",
+   "metadata": {},
+   "source": [
+    "PyTorch and scikit-learn are one of the most popular libraries for ML/DL.\n",
+    "\n",
+    "So, why not combine PyTorch with scikit-learn?\n",
+    "\n",
+    "Try `skorch`!\n",
+    "\n",
+    "`skorch` is a high-level library for PyTorch that provides a scikit-learn-compatible neural network module.\n",
+    "\n",
+    "It allows you to use the simple scikit-learn interface for PyTorch.\n",
+    "\n",
+    "Therefore you can integrate PyTorch models into scikit-learn workflows.\n",
+    "\n",
+    "See below for an example."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "09a00fe5",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install skorch"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ad9ed770",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from torch import nn\n",
+    "from skorch import NeuralNetClassifier\n",
+    "from sklearn.pipeline import Pipeline\n",
+    "from sklearn.preprocessing import StandardScaler\n",
+    "\n",
+    "class MyModule(nn.Module):\n",
+    "    def __init__(self, num_units=10, nonlin=nn.ReLU()):\n",
+    "        super().__init__()\n",
+    "\n",
+    "        self.dense = nn.Linear(20, num_units)\n",
+    "        self.nonlin = nonlin\n",
+    "        self.output = nn.Linear(num_units, 2)\n",
+    "        self.softmax = nn.Softmax(dim=-1)\n",
+    "\n",
+    "    def forward(self, X, **kwargs):\n",
+    "        X = self.nonlin(self.dense(X))\n",
+    "        X = self.dropout(X)\n",
+    "        X = self.softmax(self.output(X))\n",
+    "        return X\n",
+    "\n",
+    "net = NeuralNetClassifier(\n",
+    "    MyModule,\n",
+    "    max_epochs=10,\n",
+    "    lr=0.1,\n",
+    "    iterator_train__shuffle=True,\n",
+    ")\n",
+    "\n",
+    "pipe = Pipeline([\n",
+    "    ('scale', StandardScaler()),\n",
+    "    ('net', net),\n",
+    "])\n",
+    "\n",
+    "pipe.fit(X, y)\n",
+    "y_proba = pipe.predict_proba(X)"
+   ]
   }
  ],
  "metadata": {