Implement Generator

alimoezzi · Jul 16, 2022 · 69b4e74 · 69b4e74
1 parent 928c530
commit 69b4e74
Showing 1 changed file with 58 additions and 2 deletions.
diff --git a/mnist_gan_pytorch.ipynb b/mnist_gan_pytorch.ipynb
@@ -261,11 +261,67 @@
     "        for layer in disc_dims:\n",
     "            layers += [\n",
     "                nn.Linear(in_channels, self.hidden_dim * layer),\n",
+    "                nn.LeakyReLU(inplace=True, negative_slope=negative_slope),\n",
     "                nn.Dropout(p=dropout_probability, inplace=True),\n",
-    "                nn.LeakyReLU(inplace=True, negative_slope=negative_slope)\n",
     "            ]\n",
     "            in_channels = self.hidden_dim * layer\n",
-    "        return nn.Sequential(*layers)"
+    "        return nn.Sequential(*layers)\n",
+    "\n",
+    "    def forward(self, x):\n",
+    "        x = self.features(x)\n",
+    "        return self.classifier(x)\n"
+   ],
+   "metadata": {
+    "collapsed": false,
+    "pycharm": {
+     "name": "#%%\n"
+    }
+   }
+  },
+  {
+   "cell_type": "markdown",
+   "source": [
+    "## Generator"
+   ],
+   "metadata": {
+    "collapsed": false,
+    "pycharm": {
+     "name": "#%% md\n"
+    }
+   }
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "outputs": [],
+   "source": [
+    "gen_dims = [4, 2, 1]\n",
+    "class Generator(nn.Module):\n",
+    "    def __init__(self, input_size, hidden_dim, output_size):\n",
+    "        super(Generator, self).__init__()\n",
+    "        self.hidden_dim = hidden_dim\n",
+    "        self.output_size = output_size\n",
+    "        self.input_size = input_size\n",
+    "        self.features = self._make_layers()\n",
+    "\n",
+    "    def _make_layers(self, negative_slope=0.2, dropout_probability=0.3):\n",
+    "        layers = []\n",
+    "        in_channels = self.input_size\n",
+    "        for layer in disc_dims:\n",
+    "            layers += [\n",
+    "                nn.Linear(in_channels, self.hidden_dim * layer),\n",
+    "                nn.LeakyReLU(inplace=True, negative_slope=negative_slope),\n",
+    "                nn.Dropout(p=dropout_probability, inplace=True),\n",
+    "            ]\n",
+    "            in_channels = self.hidden_dim * layer\n",
+    "        layers += [\n",
+    "            nn.Linear(self.hidden_dim, self.output_size),\n",
+    "            nn.Tanh()\n",
+    "        ]\n",
+    "        return nn.Sequential(*layers)\n",
+    "\n",
+    "    def forward(self, x):\n",
+    "        return self.features(x)"
    ],
    "metadata": {
     "collapsed": false,