Untitled.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "43e6e6ce-268b-48cf-ac48-0ebd6f227706",
   "metadata": {},
   "source": [
    "## AlexNet 編\n",
    "\n",
    "#### 参考文献\n",
    "[1] [Probabilistic Machine Learning: An Introduction](https://probml.github.io/pml-book/book1.html) （テキスト）\n",
    "\n",
    "#### AlexNet とは [1]\n",
    "2012 年の ImageNet challenge において誤り率の記録をそれまでのものから塗り替えた畳み込みニューラルネットモデルである。  \n",
    "訓練には当時で2枚のGPUを要した。\n",
    "\n",
    "#### ImageNet の利用について\n",
    "ImageNet を以下のコードだけかいてロードしようとするとエラーが出る。エラーに「このファイルがない」とあるように [ImageNet 公式](https://image-net.org/index.php) から必要なファイルをダウンロードする必要がある。ダウンロードの許可を得るにはフリーでないメールアドレスを用いてサインアップする必要があるが、何もなければすぐ許可が得られるはずである。しかし、データが大きいため私の環境ではダウンロードに膨大な時間がかかる。\n",
    "```python\n",
    "import torchvision\n",
    "trainset = torchvision.datasets.ImageNet(root=root, train=True, download=True)\n",
    "trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True)\n",
    "testset = torchvision.datasets.ImageNet(root=root, train=False, download=True)\n",
    "testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=False)\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "cd800190-8ed0-4cbe-9555-8f2f1027e175",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "◆ テキスト474ページの図14.16を実現するのに必要な層を特定する\n",
      "※ バッチサイズ32, 3チャネル, 224x224ピクセルのダミーデータを流す。\n",
      "入力直後\n",
      "torch.Size([32, 3, 224, 224])\n",
      "1番目の畳み込み後\n",
      "torch.Size([32, 96, 54, 54])\n",
      "1番目のプール後\n",
      "torch.Size([32, 96, 26, 26])\n",
      "2番目の畳み込み後\n",
      "torch.Size([32, 256, 26, 26])\n",
      "2番目のプール後\n",
      "torch.Size([32, 256, 12, 12])\n",
      "3番目の畳み込み後\n",
      "torch.Size([32, 384, 12, 12])\n",
      "4番目の畳み込み後\n",
      "torch.Size([32, 384, 12, 12])\n",
      "5番目の畳み込み後\n",
      "torch.Size([32, 384, 12, 12])\n",
      "3番目のプール後\n",
      "torch.Size([32, 384, 5, 5])\n",
      "リシェイプした後\n",
      "torch.Size([32, 9600])\n",
      "1番目の全結合した後\n",
      "torch.Size([32, 4096])\n",
      "2番目の全結合した後\n",
      "torch.Size([32, 4096])\n",
      "3番目の全結合した後\n",
      "torch.Size([32, 1000])\n"
     ]
    }
   ],
   "source": [
    "import warnings\n",
    "warnings.simplefilter('ignore')\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "print('◆ テキスト474ページの図14.16を実現するのに必要な層を特定する')\n",
    "batch_size = 32\n",
    "channel = 3\n",
    "height = 224\n",
    "width = 224\n",
    "print(f'※ バッチサイズ{batch_size}, {channel}チャネル, {height}x{width}ピクセルのダミーデータを流す。')\n",
    "x = torch.randn(batch_size, channel, height, width)\n",
    "print(f'入力直後\\n{x.size()}')\n",
    "x = F.relu(nn.Conv2d(3, 96, 11, stride=4)(x))\n",
    "print(f'1番目の畳み込み後\\n{x.size()}')\n",
    "x = nn.MaxPool2d((3, 3), stride=2)(x)\n",
    "print(f'1番目のプール後\\n{x.size()}')\n",
    "x = F.relu(nn.Conv2d(96, 256, 5, padding=2)(x))\n",
    "print(f'2番目の畳み込み後\\n{x.size()}')\n",
    "x = nn.MaxPool2d((3, 3), stride=2)(x)\n",
    "print(f'2番目のプール後\\n{x.size()}')\n",
    "x = F.relu(nn.Conv2d(256, 384, 3, padding=1)(x))\n",
    "print(f'3番目の畳み込み後\\n{x.size()}')\n",
    "x = F.relu(nn.Conv2d(384, 384, 3, padding=1)(x))\n",
    "print(f'4番目の畳み込み後\\n{x.size()}')\n",
    "x = F.relu(nn.Conv2d(384, 384, 3, padding=1)(x))\n",
    "print(f'5番目の畳み込み後\\n{x.size()}')\n",
    "x = nn.MaxPool2d((3, 3), stride=2)(x)\n",
    "print(f'3番目のプール後\\n{x.size()}')\n",
    "size = x.size()[1:]\n",
    "num_features = 1\n",
    "for s in size:\n",
    "    num_features *= s\n",
    "x = x.view(-1, num_features)\n",
    "print(f'リシェイプした後\\n{x.size()}')\n",
    "x = F.relu(nn.Linear(num_features, 4096)(x))\n",
    "print(f'1番目の全結合した後\\n{x.size()}')\n",
    "x = F.relu(nn.Linear(4096, 4096)(x))\n",
    "print(f'2番目の全結合した後\\n{x.size()}')\n",
    "x = nn.Linear(4096, 1000)(x)\n",
    "print(f'3番目の全結合した後\\n{x.size()}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "ff33ac99-9790-4910-aaa5-7d2ead76d020",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "◆ 今回学習するネットワーク\n",
      "AlexNet(\n",
      "  (max_pool): MaxPool2d(kernel_size=(3, 3), stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (conv1): Conv2d(3, 96, kernel_size=(11, 11), stride=(4, 4))\n",
      "  (conv2): Conv2d(96, 256, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))\n",
      "  (conv3): Conv2d(256, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (conv4): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (conv5): Conv2d(384, 384, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (fc1): Linear(in_features=9600, out_features=4096, bias=True)\n",
      "  (fc2): Linear(in_features=4096, out_features=4096, bias=True)\n",
      "  (fc3): Linear(in_features=4096, out_features=1000, bias=True)\n",
      ")\n",
      "\n",
      "◆ 今回学習するパラメータたち\n",
      "conv1.weight   torch.Size([96, 3, 11, 11])\n",
      "conv1.bias     torch.Size([96])\n",
      "conv2.weight   torch.Size([256, 96, 5, 5])\n",
      "conv2.bias     torch.Size([256])\n",
      "conv3.weight   torch.Size([384, 256, 3, 3])\n",
      "conv3.bias     torch.Size([384])\n",
      "conv4.weight   torch.Size([384, 384, 3, 3])\n",
      "conv4.bias     torch.Size([384])\n",
      "conv5.weight   torch.Size([384, 384, 3, 3])\n",
      "conv5.bias     torch.Size([384])\n",
      "fc1.weight     torch.Size([4096, 9600])\n",
      "fc1.bias       torch.Size([4096])\n",
      "fc2.weight     torch.Size([4096, 4096])\n",
      "fc2.bias       torch.Size([4096])\n",
      "fc3.weight     torch.Size([1000, 4096])\n",
      "fc3.bias       torch.Size([1000])\n"
     ]
    }
   ],
   "source": [
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "class AlexNet(nn.Module):\n",
    "\n",
    "    def __init__(self):\n",
    "        super(AlexNet, self).__init__()\n",
    "        self.max_pool = nn.MaxPool2d((3, 3), stride=2)\n",
    "        self.conv1 = nn.Conv2d(3, 96, 11, stride=4)\n",
    "        self.conv2 = nn.Conv2d(96, 256, 5, padding=2)\n",
    "        self.conv3 = nn.Conv2d(256, 384, 3, padding=1)\n",
    "        self.conv4 = nn.Conv2d(384, 384, 3, padding=1)\n",
    "        self.conv5 = nn.Conv2d(384, 384, 3, padding=1)\n",
    "        self.fc1 = nn.Linear(9600, 4096)  # TODO: ここが 224x224 ピクセル決め打ちになっている\n",
    "        self.fc2 = nn.Linear(4096, 4096)\n",
    "        self.fc3 = nn.Linear(4096, 1000)\n",
    "\n",
    "    def forward(self, x):\n",
    "        # TODO: ドロップアウト\n",
    "        x = self.max_pool(F.relu(self.conv1(x)))\n",
    "        x = self.max_pool(F.relu(self.conv2(x)))\n",
    "        x = F.relu(self.conv3(x))\n",
    "        x = F.relu(self.conv4(x))\n",
    "        x = F.relu(self.conv5(x))\n",
    "        x = x.view(-1, 9600)\n",
    "        x = F.relu(self.fc1(x))\n",
    "        x = F.relu(self.fc2(x))\n",
    "        x = self.fc3(x)\n",
    "        return x\n",
    "\n",
    "net = AlexNet()\n",
    "\n",
    "print('◆ 今回学習するネットワーク')\n",
    "print(net)\n",
    "print('\\n◆ 今回学習するパラメータたち')\n",
    "for name, param in net.named_parameters():\n",
    "    print(name.ljust(14), param.size())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "75e79400-39d3-4ac6-8b55-4bb514fbc202",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}