2-Try-Trainer-class_ja.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Trainerを使ってみよう\n",
    "\n",
    "`Trainer`を使うと学習ループを陽に書く必要がなくなります。またいろいろな便利なExtentionを使うことで可視化やログの保存などが楽になります。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 1. データセットの準備"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from chainer.datasets import mnist\n",
    "\n",
    "train, test = mnist.get_mnist()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 2. Iteratorの準備"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from chainer import iterators\n",
    "\n",
    "batchsize = 128\n",
    "\n",
    "train_iter = iterators.SerialIterator(train, batchsize)\n",
    "test_iter = iterators.SerialIterator(test, batchsize, False, False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 3. Modelの準備\n",
    "\n",
    "ここでは、先程と同じモデルを再度用います。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<__main__.MLP at 0x7fb11c1e33c8>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import chainer\n",
    "import chainer.links as L\n",
    "import chainer.functions as F\n",
    "\n",
    "class MLP(chainer.Chain):\n",
    "\n",
    "    def __init__(self, n_mid_units=100, n_out=10):\n",
    "        super(MLP, self).__init__(\n",
    "            l1=L.Linear(None, n_mid_units),\n",
    "            l2=L.Linear(None, n_mid_units),\n",
    "            l3=L.Linear(None, n_out),\n",
    "        )\n",
    "\n",
    "    def __call__(self, x):\n",
    "        h1 = F.relu(self.l1(x))\n",
    "        h2 = F.relu(self.l2(h1))\n",
    "        return self.l3(h2)\n",
    "\n",
    "gpu_id = 0\n",
    "\n",
    "model = MLP()\n",
    "model.to_gpu(gpu_id)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 4. Updaterの準備\n",
    "\n",
    "Trainerは学習に必要な全てのものをひとまとめにするクラスです。それは主に以下のようなものを保持します。\n",
    "\n",
    "- Updater\n",
    "    - Iterator\n",
    "        - Dataset\n",
    "    - Optimizer\n",
    "        - Model\n",
    "\n",
    "`Trainer`オブジェクトを作成するときに渡すのは基本的に`Updater`だけですが、`Updater`は中に`Iterator`と`Optimizer`を持っています。`Iterator`からはデータセットにアクセスすることができ、`Optimizer`は中でモデルへの参照を保持しているので、モデルのパラメータを更新することができます。つまり、`Updater`が内部で\n",
    "\n",
    "1. データセットからデータを取り出し（Iterator）\n",
    "2. モデルに渡してロスを計算し（Model = Optimizer.target）\n",
    "3. Optimizerを使ってモデルのパラメータを更新する（Optimizer）\n",
    "\n",
    "という一連の学習の主要部分を行うことができるということです。では、`Updater`オブジェクトを作成してみます。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from chainer import optimizers\n",
    "from chainer import training\n",
    "\n",
    "max_epoch = 10\n",
    "gpu_id = 0\n",
    "\n",
    "# モデルをClassifierで包んで、ロスの計算などをモデルに含める\n",
    "model = L.Classifier(model)\n",
    "model.to_gpu(gpu_id)\n",
    "\n",
    "# 最適化手法の選択\n",
    "optimizer = optimizers.SGD()\n",
    "optimizer.setup(model)\n",
    "\n",
    "# UpdaterにIteratorとOptimizerを渡す\n",
    "updater = training.StandardUpdater(train_iter, optimizer, device=gpu_id)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### NOTE\n",
    "\n",
    "ここで、上で定義したモデルのオブジェクトを`L.Classifier`に渡して、新しい`Chain`にしています。`L.Classifier`は`Chain`を継承したクラスで、渡された`Chain`を`predictor`というプロパティに保存します。`()`アクセサでデータとラベルを渡すと、中で`__call__`が実行され、まず渡されたデータの方を`predictor`に通し、その出力`y`と同じく渡されていたラベルを、コンストラクタの`lossfun`引数で指定されたロス関数に渡して、その出力`Variable`を返します。`lossfun`はデフォルトで`softmax_cross_entropy`に指定されています。\n",
    "\n",
    "`StandardUpdater`は前述のUpdaterの行う処理を遂行する最もシンプルなクラスです。この他にも複数のGPUを用いるための`ParallelUpdater`などが用意されています。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 5. Trainerの設定\n",
    "\n",
    "最後に、`Trainer`の設定を行います。`Trainer`のオブジェクトを作成する際に必須となるのは、先程作成した`Updater`オブジェクトだけですが、二番目の引数`stop_trigger`に学習をどのタイミングで終了するかを表す`(長さ, 単位)`という形のタプルを与えると、指定したタイミングで学習を自動的に終了することができます。長さには任意の整数、単位には`'epoch'`か`'iteration'`のいずれかの文字列を指定できます。`stop_trigger`を指定しない場合、学習は自動的には止まりません。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# TrainerにUpdaterを渡す\n",
    "trainer = training.Trainer(updater, (max_epoch, 'epoch'),\n",
    "                           out='mnist_result')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "`out`引数では、この次に説明する`Extension`を使って、ログファイルやロスの変化の過程を描画したグラフの画像ファイルなどを保存するディレクトリを指定しています。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 6. TrainerにExtensionを追加する\n",
    "\n",
    "\n",
    "`Trainer`を使う利点として、\n",
    "\n",
    "- ログを自動的にファイルに保存（`LogReport`)\n",
    "- ターミナルに定期的にロスなどの情報を表示（`PrintReport`）\n",
    "- ロスを定期的にグラフで可視化して画像として保存（`PlotReport`)\n",
    "- 定期的にモデルやOptimizerの状態を自動シリアライズ（`snapshot`/`snapshot_object`）\n",
    "- 学習の進捗を示すプログレスバーを表示（`ProgressBar`）\n",
    "- モデルの構造をGraphvizのdot形式で保存（`dump_graph`）\n",
    "\n",
    "などなどの様々な便利な機能を簡単に利用することができる点があります。これらの機能を利用するには、`Trainer`オブジェクトに対して`extend`メソッドを使って追加したい`Extension`のオブジェクトを渡してやるだけです。では実際に幾つかの`Extension`を追加してみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from chainer.training import extensions\n",
    "\n",
    "trainer.extend(extensions.LogReport())\n",
    "trainer.extend(extensions.snapshot(filename='snapshot_epoch-{.updater.epoch}'))\n",
    "trainer.extend(extensions.snapshot_object(model.predictor, filename='model_epoch-{.updater.epoch}'))\n",
    "trainer.extend(extensions.Evaluator(test_iter, model, device=gpu_id))\n",
    "trainer.extend(extensions.PrintReport(['epoch', 'main/loss', 'main/accuracy', 'validation/main/loss', 'validation/main/accuracy', 'elapsed_time']))\n",
    "trainer.extend(extensions.PlotReport(['main/loss', 'validation/main/loss'], x_key='epoch', file_name='loss.png'))\n",
    "trainer.extend(extensions.PlotReport(['main/accuracy', 'validation/main/accuracy'], x_key='epoch', file_name='accuracy.png'))\n",
    "trainer.extend(extensions.dump_graph('main/loss'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### `LogReport`\n",
    "\n",
    "`epoch`や`iteration`ごとの`loss`, `accuracy`などを自動的に集計し、`Trainer`の`out`引数で指定した出力ディレクトリに`log`というファイル名で保存します。\n",
    "\n",
    "### `snapshot`\n",
    "\n",
    "`Trainer`の`out`引数で指定した出力ディレクトリに`Trainer`オブジェクトを指定されたタイミング（デフォルトでは1エポックごと）に保存します。`Trainer`オブジェクトは上述のように`Updater`を持っており、この中に`Optimizer`とモデルが保持されているため、この`Extension`でスナップショットをとっておけば、学習の復帰や学習済みモデルを使った推論などが学習終了後にも可能になります。\n",
    "\n",
    "### `snapshot_object`\n",
    "\n",
    "しかし、`Trainer`ごと保存した場合、しばしば中身のモデルだけ取り出すのが面倒な場合があります。そこで、`snapshot_object`を使って指定したオブジェクト（ここでは`Classifier`で包まれたモデル）だけを、`Trainer`とは別に保存するようにします。`Classifier`は第1引数に渡された`Chain`オブジェクトを自身の`predictor`というプロパティとして保持してロスの計算を行う`Chain`であり、`Classifier`はそもそもモデル以外にパラメータを持たないので、ここでは後々学習済みモデルを推論に使うことを見越して`model.predictor`を保存対象として指定しています。\n",
    "\n",
    "### `dump_graph`\n",
    "\n",
    "指定された`Variable`オブジェクトから辿れる計算グラフをGraphvizのdot形式で保存します。保存先は`Trainer`の`out`引数で指定した出力ディレクトリです。\n",
    "\n",
    "### `Evaluator`\n",
    "\n",
    "評価用のデータセットの`Iterator`と、学習に使うモデルのオブジェクトを渡しておくことで、学習中のモデルを指定されたタイミングで評価用データセットを用いて評価します。\n",
    "\n",
    "### `PrintReport`\n",
    "\n",
    "`Reporter`によって集計された値を標準出力に出力します。このときどの値を出力するかを、リストの形で与えます。\n",
    "\n",
    "### `PlotReport`\n",
    "\n",
    "引数のリストで指定された値の変遷を`matplotlib`ライブラリを使ってグラフに描画し、出力ディレクトリに`file_name`引数で指定されたファイル名で画像として保存します。\n",
    "\n",
    "---\n",
    "\n",
    "これらの`Extension`は、ここで紹介した以外にも、例えば`trigger`によって個別に作動するタイミングを指定できるなどのいくつかのオプションを持っており、より柔軟に組み合わせることができます。詳しくは公式のドキュメントを見てください：[Trainer extensions](http://docs.chainer.org/en/latest/reference/extensions.html)。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 7. 学習を開始する\n",
    "\n",
    "学習を開始するには、`Trainer`オブジェクトのメソッド`run`を呼ぶだけです。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch       main/loss   main/accuracy  validation/main/loss  validation/main/accuracy  elapsed_time\n",
      "\u001b[J1           1.49348     0.623801       0.753058              0.838113                  2.52215       \n",
      "\u001b[J2           0.583053    0.854611       0.44962               0.885186                  4.76435       \n",
      "\u001b[J3           0.422944    0.885028       0.366528              0.89913                   6.96108       \n",
      "\u001b[J4           0.364944    0.896735       0.329752              0.908228                  9.15184       \n",
      "\u001b[J5           0.33318     0.904501       0.306012              0.914656                  11.3642       \n",
      "\u001b[J6           0.311984    0.910681       0.290334              0.918216                  13.5463       \n",
      "\u001b[J7           0.295134    0.914645       0.275341              0.920985                  15.7157       \n",
      "\u001b[J8           0.282358    0.918436       0.264523              0.923457                  17.8918       \n",
      "\u001b[J9           0.27013     0.922408       0.257301              0.926325                  20.0797       \n",
      "\u001b[J10          0.259913    0.925373       0.247199              0.929688                  22.3097       \n"
     ]
    }
   ],
   "source": [
    "trainer.run()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "保存されているロスのグラフを確認してみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAg0AAAEKCAYAAACR9WaxAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nERGRZlFoEBERkWZRaBAREZFm+f9kvFwvG2xRXQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import Image\n",
    "Image(filename='mnist_result/loss.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "精度のグラフも見てみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAi0AAAEKCAYAAADeqGVhAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nACmIFgAghQv9SoHVrfJg0wAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Image(filename='mnist_result/accuracy.png')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "ついでに、`dump_graph`という`Extension`が出力した計算グラフを、`Graphviz`を使って画像化して見てみましょう。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "%%bash\n",
    "dot -Tpng mnist_result/cg.dot -o mnist_result/cg.png"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "![](mnist_result/cg.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "上から下へ向かって、データやパラメータがどのような`Function`に渡されて計算が行われ、ロスが出力されたかが分かります。"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## 8. 学習済みモデルで推論する"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nH0iK8ANJEX4gKcIPJEX4gaQIP5DUXwFGhz+pWT5yuQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fb0fe2b2ba8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label: 7\n",
      "predicted_label: 7\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from chainer import serializers\n",
    "from chainer.cuda import to_gpu\n",
    "from chainer.cuda import to_cpu\n",
    "\n",
    "model = MLP()\n",
    "serializers.load_npz('mnist_result/model_epoch-10', model)\n",
    "model.to_gpu(gpu_id)\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "x, t = test[0]\n",
    "plt.imshow(x.reshape(28, 28), cmap='gray')\n",
    "plt.show()\n",
    "print('label:', t)\n",
    "\n",
    "x = to_gpu(x[None, ...])\n",
    "y = model(x)\n",
    "y = to_cpu(y.data)\n",
    "\n",
    "print('predicted_label:', y.argmax(axis=1)[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "無事正解できました。"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.4.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}