06_mlp_with_tensorflow_core.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# TensorFlow Coreで多層パーセプトロン\n",
    "\n",
    "参考資料\n",
    "\n",
    "- https://github.com/aymericdamien/TensorFlow-Examples/blob/master/notebooks/4_Utils/tensorboard_basic.ipynb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# TensorFlowの読み込み\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# graphs02ディレクトリがあったら削除\n",
    "if tf.gfile.Exists('./graphs02'):\n",
    "    tf.gfile.DeleteRecursively('./graphs02/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully downloaded train-images-idx3-ubyte.gz 9912422 bytes.\n",
      "Extracting ../datasets/mnist/train-images-idx3-ubyte.gz\n",
      "Successfully downloaded train-labels-idx1-ubyte.gz 28881 bytes.\n",
      "Extracting ../datasets/mnist/train-labels-idx1-ubyte.gz\n",
      "Successfully downloaded t10k-images-idx3-ubyte.gz 1648877 bytes.\n",
      "Extracting ../datasets/mnist/t10k-images-idx3-ubyte.gz\n",
      "Successfully downloaded t10k-labels-idx1-ubyte.gz 4542 bytes.\n",
      "Extracting ../datasets/mnist/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "# MNIST データセットの読み込み\n",
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "mnist = input_data.read_data_sets('../datasets/mnist/', one_hot=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(55000, 784)\n",
      "(5000, 784)\n",
      "(10000, 784)\n"
     ]
    }
   ],
   "source": [
    "print(mnist.train.images.shape)\n",
    "print(mnist.validation.images.shape)\n",
    "print(mnist.test.images.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(55000, 10)\n",
      "(5000, 10)\n",
      "(10000, 10)\n"
     ]
    }
   ],
   "source": [
    "print(mnist.train.labels.shape)\n",
    "print(mnist.validation.labels.shape)\n",
    "print(mnist.test.labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[7 3 4 6 1 8 1 0 9 8]\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAA9CAYAAABbalkHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x124a10048>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# 画像とラベルを確認\n",
    "fig, ax = plt.subplots(1, 10)\n",
    "for ii in range(10):\n",
    "    ax[ii].imshow(mnist.train.images[ii].reshape([28, 28]), 'gray')\n",
    "    ax[ii].axis('off')\n",
    "    \n",
    "print(np.argmax(mnist.train.labels[:10], axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# ハイパーパラメータ\n",
    "learning_rate = 0.01\n",
    "training_epochs = 150\n",
    "batch_size = 100\n",
    "display_step = 1  # エポック\n",
    "logdir = './graphs02'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = tf.placeholder(tf.float32, [None, 784], name='InputData')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y = tf.placeholder(tf.float32, [None, 10], name='LabelData')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# パラメータ\n",
    "W = tf.Variable(tf.zeros([784, 10]), name='Weights')\n",
    "b =  tf.Variable(tf.zeros([10]), name='Bias')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "TensorBoardでみやすくなるようスコープに入れていく"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# Model\n",
    "#with tf.name_scope('Model'):\n",
    "logit = tf.matmul(x, W) + b\n",
    "pred = tf.nn.softmax(logit)\n",
    "\n",
    "#with tf.name_scope('Loss'):\n",
    "# 数値的に不安定な実装なので要注意\n",
    "loss = tf.reduce_mean(- tf.reduce_sum(y * tf.log(pred), reduction_indices=1))\n",
    "# 数値的に安定な実装\n",
    "#loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=logit))\n",
    "\n",
    "#with tf.name_scope('SGD'):\n",
    "optimizer = tf.train.GradientDescentOptimizer(learning_rate)\n",
    "train_op = optimizer.minimize(loss)\n",
    "\n",
    "#with tf.name_scope('Accuracy'):\n",
    "accuracy = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))\n",
    "accuracy = tf.reduce_mean(tf.cast(accuracy, tf.float32))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# 変数の初期化\n",
    "init = tf.global_variables_initializer()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "tf.summary.scalar('loss', loss)\n",
    "tf.summary.scalar('accuracy', accuracy)\n",
    "summary_op = tf.summary.merge_all()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "tensorboardを立ち上げるには以下のコードを打つ\n",
    "\n",
    "```\n",
    "tensorboard --logdir='./graphs02' --port=6006\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 0001\tloss=1.183409788\n",
      "Epoch: 0002\tloss=0.664995495\n",
      "Epoch: 0003\tloss=0.552615560\n",
      "Epoch: 0004\tloss=0.498578957\n",
      "Epoch: 0005\tloss=0.465489370\n",
      "Epoch: 0006\tloss=0.442536707\n",
      "Epoch: 0007\tloss=0.425493833\n",
      "Epoch: 0008\tloss=0.412156212\n",
      "Epoch: 0009\tloss=0.401333432\n",
      "Epoch: 0010\tloss=0.392394326\n",
      "Epoch: 0011\tloss=0.384756775\n",
      "Epoch: 0012\tloss=0.378210272\n",
      "Epoch: 0013\tloss=0.372383532\n",
      "Epoch: 0014\tloss=0.367303779\n",
      "Epoch: 0015\tloss=0.362699974\n",
      "Epoch: 0016\tloss=0.358551035\n",
      "Epoch: 0017\tloss=0.354876478\n",
      "Epoch: 0018\tloss=0.351449799\n",
      "Epoch: 0019\tloss=0.348308506\n",
      "Epoch: 0020\tloss=0.345436832\n",
      "Epoch: 0021\tloss=0.342737799\n",
      "Epoch: 0022\tloss=0.340235049\n",
      "Epoch: 0023\tloss=0.337912788\n",
      "Epoch: 0024\tloss=0.335688573\n",
      "Epoch: 0025\tloss=0.333609294\n",
      "Epoch: 0026\tloss=0.331787237\n",
      "Epoch: 0027\tloss=0.329947388\n",
      "Epoch: 0028\tloss=0.328229609\n",
      "Epoch: 0029\tloss=0.326601887\n",
      "Epoch: 0030\tloss=0.325061481\n",
      "Epoch: 0031\tloss=0.323592124\n",
      "Epoch: 0032\tloss=0.322125017\n",
      "Epoch: 0033\tloss=0.320808485\n",
      "Epoch: 0034\tloss=0.319512220\n",
      "Epoch: 0035\tloss=0.318328226\n",
      "Epoch: 0036\tloss=0.317105841\n",
      "Epoch: 0037\tloss=0.315976388\n",
      "Epoch: 0038\tloss=0.314902909\n",
      "Epoch: 0039\tloss=0.313864408\n",
      "Epoch: 0040\tloss=0.312858640\n",
      "Epoch: 0041\tloss=0.311902636\n",
      "Epoch: 0042\tloss=0.310931109\n",
      "Epoch: 0043\tloss=0.310021337\n",
      "Epoch: 0044\tloss=0.309167000\n",
      "Epoch: 0045\tloss=0.308319522\n",
      "Epoch: 0046\tloss=0.307478823\n",
      "Epoch: 0047\tloss=0.306738491\n",
      "Epoch: 0048\tloss=0.305920657\n",
      "Epoch: 0049\tloss=0.305174084\n",
      "Epoch: 0050\tloss=0.304496469\n",
      "Epoch: 0051\tloss=0.303763741\n",
      "Epoch: 0052\tloss=0.303102923\n",
      "Epoch: 0053\tloss=0.302446980\n",
      "Epoch: 0054\tloss=0.301792336\n",
      "Epoch: 0055\tloss=0.301166974\n",
      "Epoch: 0056\tloss=0.300528809\n",
      "Epoch: 0057\tloss=0.299989830\n",
      "Epoch: 0058\tloss=0.299395788\n",
      "Epoch: 0059\tloss=0.298800382\n",
      "Epoch: 0060\tloss=0.298224142\n",
      "Epoch: 0061\tloss=0.297687535\n",
      "Epoch: 0062\tloss=0.297183529\n",
      "Epoch: 0063\tloss=0.296671164\n",
      "Epoch: 0064\tloss=0.296143161\n",
      "Epoch: 0065\tloss=0.295676750\n",
      "Epoch: 0066\tloss=0.295198612\n",
      "Epoch: 0067\tloss=0.294731166\n",
      "Epoch: 0068\tloss=0.294241957\n",
      "Epoch: 0069\tloss=0.293799029\n",
      "Epoch: 0070\tloss=0.293330308\n",
      "Epoch: 0071\tloss=0.292943184\n",
      "Epoch: 0072\tloss=0.292510362\n",
      "Epoch: 0073\tloss=0.292073752\n",
      "Epoch: 0074\tloss=0.291668334\n",
      "Epoch: 0075\tloss=0.291273513\n",
      "Epoch: 0076\tloss=0.290902896\n",
      "Epoch: 0077\tloss=0.290526736\n",
      "Epoch: 0078\tloss=0.290142769\n",
      "Epoch: 0079\tloss=0.289769953\n",
      "Epoch: 0080\tloss=0.289391896\n",
      "Epoch: 0081\tloss=0.289042890\n",
      "Epoch: 0082\tloss=0.288688961\n",
      "Epoch: 0083\tloss=0.288333116\n",
      "Epoch: 0084\tloss=0.287975113\n",
      "Epoch: 0085\tloss=0.287659116\n",
      "Epoch: 0086\tloss=0.287343417\n",
      "Epoch: 0087\tloss=0.287004929\n",
      "Epoch: 0088\tloss=0.286696656\n",
      "Epoch: 0089\tloss=0.286394937\n",
      "Epoch: 0090\tloss=0.286032521\n",
      "Epoch: 0091\tloss=0.285784305\n",
      "Epoch: 0092\tloss=0.285442320\n",
      "Epoch: 0093\tloss=0.285155036\n",
      "Epoch: 0094\tloss=0.284871966\n",
      "Epoch: 0095\tloss=0.284618939\n",
      "Epoch: 0096\tloss=0.284330175\n",
      "Epoch: 0097\tloss=0.284034647\n",
      "Epoch: 0098\tloss=0.283756155\n",
      "Epoch: 0099\tloss=0.283452341\n",
      "Epoch: 0100\tloss=0.283236738\n",
      "Epoch: 0101\tloss=0.282966306\n",
      "Epoch: 0102\tloss=0.282685788\n",
      "Epoch: 0103\tloss=0.282448681\n",
      "Epoch: 0104\tloss=0.282187632\n",
      "Epoch: 0105\tloss=0.281972968\n",
      "Epoch: 0106\tloss=0.281681564\n",
      "Epoch: 0107\tloss=0.281482377\n",
      "Epoch: 0108\tloss=0.281201127\n",
      "Epoch: 0109\tloss=0.281019025\n",
      "Epoch: 0110\tloss=0.280759126\n",
      "Epoch: 0111\tloss=0.280526239\n",
      "Epoch: 0112\tloss=0.280276654\n",
      "Epoch: 0113\tloss=0.280070695\n",
      "Epoch: 0114\tloss=0.279845023\n",
      "Epoch: 0115\tloss=0.279618723\n",
      "Epoch: 0116\tloss=0.279444497\n",
      "Epoch: 0117\tloss=0.279213919\n",
      "Epoch: 0118\tloss=0.278967843\n",
      "Epoch: 0119\tloss=0.278753648\n",
      "Epoch: 0120\tloss=0.278613835\n",
      "Epoch: 0121\tloss=0.278340328\n",
      "Epoch: 0122\tloss=0.278191007\n",
      "Epoch: 0123\tloss=0.277974586\n",
      "Epoch: 0124\tloss=0.277760205\n",
      "Epoch: 0125\tloss=0.277572658\n",
      "Epoch: 0126\tloss=0.277380315\n",
      "Epoch: 0127\tloss=0.277163677\n",
      "Epoch: 0128\tloss=0.277043809\n",
      "Epoch: 0129\tloss=0.276840233\n",
      "Epoch: 0130\tloss=0.276591606\n",
      "Epoch: 0131\tloss=0.276459099\n",
      "Epoch: 0132\tloss=0.276273833\n",
      "Epoch: 0133\tloss=0.276084155\n",
      "Epoch: 0134\tloss=0.275920965\n",
      "Epoch: 0135\tloss=0.275754005\n",
      "Epoch: 0136\tloss=0.275569838\n",
      "Epoch: 0137\tloss=0.275408052\n",
      "Epoch: 0138\tloss=0.275224384\n",
      "Epoch: 0139\tloss=0.275087507\n",
      "Epoch: 0140\tloss=0.274877455\n",
      "Epoch: 0141\tloss=0.274746140\n",
      "Epoch: 0142\tloss=0.274579546\n",
      "Epoch: 0143\tloss=0.274415551\n",
      "Epoch: 0144\tloss=0.274249402\n",
      "Epoch: 0145\tloss=0.274095260\n",
      "Epoch: 0146\tloss=0.273949389\n",
      "Epoch: 0147\tloss=0.273781400\n",
      "Epoch: 0148\tloss=0.273623386\n",
      "Epoch: 0149\tloss=0.273479539\n",
      "Epoch: 0150\tloss=0.273332215\n"
     ]
    }
   ],
   "source": [
    "with tf.Session() as sess:\n",
    "    sess.run(init)\n",
    "    \n",
    "    # ログのTensorBoardへの書き出し\n",
    "    summary_writer = tf.summary.FileWriter(logdir, graph=tf.get_default_graph())\n",
    "    \n",
    "    #　学習\n",
    "    for epoch in range(training_epochs):\n",
    "        \n",
    "        average_loss = 0.\n",
    "        total_batch = int(mnist.train.num_examples / batch_size)\n",
    "        \n",
    "        for i in range(total_batch):\n",
    "            \n",
    "            # ミニバッチを取得\n",
    "            batch_xs, batch_ys = mnist.train.next_batch(batch_size)\n",
    "            \n",
    "            # 計算グラフを実行\n",
    "            _, loss_value, summary = sess.run([train_op, loss, summary_op],\n",
    "                                             feed_dict={x: batch_xs, y: batch_ys})\n",
    "            \n",
    "            #  ステップごとにログを書き出す\n",
    "            summary_writer.add_summary(summary, epoch * total_batch + i)\n",
    "            \n",
    "            # 平均損失を計算\n",
    "            average_loss += loss_value / total_batch\n",
    "            \n",
    "        # エポックの平均損失を時々表示\n",
    "        if (epoch + 1) % display_step == 0:\n",
    "            print(\"Epoch: %04d\\tloss=%.9f\" % (epoch+1, average_loss))"
   ]
  }
 ],
 "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": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}