capstone_multi2_CNN.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2 识别数字序列的卷积神经网络"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "from six.moves import cPickle as pickle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "% matplotlib inline\n",
    "import seaborn\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 加载经过预处理的数据"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "IMAGE_SIZE = 32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "image_size_40px = ''\n",
    "# image_size_40px = '_40'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dataset shape:  (33402, 40, 40)\n",
      "labels shape:  (33402, 5)\n"
     ]
    }
   ],
   "source": [
    "pickle_file = 'svhn_multi_train%s.pickle' % image_size_40px\n",
    "\n",
    "with open(pickle_file, 'rb') as f:\n",
    "    dataset = pickle.load(f)\n",
    "    images = dataset['dataset']\n",
    "    labels = dataset['labels'][:, 1:6]\n",
    "    \n",
    "print('dataset shape: ', images.shape)\n",
    "print('labels shape: ', labels.shape)"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {
    "collapsed": false
   },
   "source": [
    "pickle_file = 'svhn_multi_extra%s.pickle' % image_size_40px\n",
    "\n",
    "with open(pickle_file, 'rb') as f:\n",
    "    dataset = pickle.load(f)\n",
    "    images = np.vstack((images, dataset['dataset']))\n",
    "    labels = np.vstack((labels, dataset['labels'][:, 1:6]))\n",
    "\n",
    "print('dataset shape: ', images.shape)\n",
    "print('labels shape: ', labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 1,  9, 10, 10, 10],\n",
       "       [ 2,  3, 10, 10, 10],\n",
       "       [ 2,  5, 10, 10, 10],\n",
       "       [ 9,  3, 10, 10, 10],\n",
       "       [ 3,  1, 10, 10, 10],\n",
       "       [ 3,  3, 10, 10, 10],\n",
       "       [ 2,  8, 10, 10, 10],\n",
       "       [ 7,  4,  4, 10, 10],\n",
       "       [ 1,  2,  8, 10, 10],\n",
       "       [ 1,  6, 10, 10, 10]], dtype=int8)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 构建交叉验证集\n",
    "\n",
    "从训练数据中选择5%作为交叉验证集。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.cross_validation import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training dataset shape:  (31731, 40, 40)\n",
      "Training labels shape:  (31731, 5)\n",
      "Validing dataset shape:  (1671, 40, 40)\n",
      "Validing labels shape:  (1671, 5)\n"
     ]
    }
   ],
   "source": [
    "train_dataset, valid_dataset , train_labels, valid_labels = train_test_split(images, labels, test_size=0.05, random_state=42)\n",
    "\n",
    "print('Training dataset shape: ', train_dataset.shape)\n",
    "print('Training labels shape: ', train_labels.shape)\n",
    "print('Validing dataset shape: ', valid_dataset.shape)\n",
    "print('Validing labels shape: ', valid_labels.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "NUM_DIGITS = 5\n",
    "NUM_LABELS = 11 # 0-9 + 10==doesn't exist\n",
    "NUM_CHANNELS = 1 # grayscale"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def reformat(dataset, labels):\n",
    "    dataset = dataset.reshape((-1, IMAGE_SIZE, IMAGE_SIZE, NUM_CHANNELS)).astype(np.float32)\n",
    "\n",
    "    return dataset, labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training dataset shape:  (31731, 40, 40, 1)\n",
      "Validing dataset shape:  (1671, 40, 40, 1)\n"
     ]
    }
   ],
   "source": [
    "train_dataset, train_labels = reformat(train_dataset, train_labels)\n",
    "valid_dataset, valid_labels = reformat(valid_dataset, valid_labels)\n",
    "\n",
    "print('Training dataset shape: ', train_dataset.shape)\n",
    "print('Validing dataset shape: ', valid_dataset.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 构建计算图\n",
    "\n",
    "识别数字序列的卷积神经网络结构包含3层卷积层，1层全连接层和一层输出层。所有的连接均为前向，从一层到相邻的下一层。3层卷积层分别包含[24, 48, 80]张特征图。全连接层包含256个单元。每一层卷积层的输出通过最大值池化，池化窗的大小为2x2，步长为2。训练时，全连接层的输出以0.6的概率随机将值赋为0。所有卷积核的大小为5x5。\n",
    "\n",
    "训练的过程中，避免过拟合的方法：  \n",
    "\n",
    "- 参数使用二范数正则化\n",
    "- 早期停止\n",
    "- 随机抛弃"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def createWeight(shape, name):\n",
    "    return tf.Variable(tf.truncated_normal(shape, stddev=0.1), name=name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def createBiase(shape, name):\n",
    "    return tf.Variable(tf.zeros(shape), name=name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "BATCH_SIZE = 64\n",
    "PATCH_SIZE = 5\n",
    "DEPTH_1 = 24\n",
    "DEPTH_2 = 48\n",
    "DEPTH_3 = 80\n",
    "NUM_HIDDEN = 256\n",
    "SEED = 42"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "graph = tf.Graph()\n",
    "with graph.as_default():\n",
    "    # 输入数据\n",
    "    tf_train_dataset = tf.placeholder(tf.float32, shape=(None, IMAGE_SIZE, IMAGE_SIZE, NUM_CHANNELS))\n",
    "    tf_train_labels = tf.placeholder(tf.int64, shape=(None, NUM_DIGITS))\n",
    "    tf_valid_dataset = tf.constant(valid_dataset)\n",
    "    tf_valid_labels = tf.constant(valid_labels, dtype=tf.int64)\n",
    "    \n",
    "    # 模型参数-卷积层\n",
    "    conv1_W = createWeight([PATCH_SIZE, PATCH_SIZE, NUM_CHANNELS, DEPTH_1], 'conv1_weights')\n",
    "    conv1_B = createBiase([DEPTH_1], 'conv1_biases')\n",
    "    conv2_W = createWeight([PATCH_SIZE, PATCH_SIZE, DEPTH_1, DEPTH_2], 'conv2_weights')\n",
    "    conv2_B = createBiase([DEPTH_2], 'conv2_biases')\n",
    "    conv3_W = createWeight([PATCH_SIZE, PATCH_SIZE, DEPTH_2, DEPTH_3], 'conv3_weights')\n",
    "    conv3_B = createBiase([DEPTH_3], 'conv3_biases')\n",
    "    fc1_W = createWeight([IMAGE_SIZE//8 * IMAGE_SIZE//8 * DEPTH_3, NUM_HIDDEN], 'fc1_weights')\n",
    "    fc1_B = createBiase([NUM_HIDDEN], 'fc1_biases')\n",
    "    \n",
    "    # 模型参数-全连接层\n",
    "    fc_d1_W = createWeight([NUM_HIDDEN, NUM_LABELS], 'fc_digit1_weights')\n",
    "    fc_d1_B = createBiase([NUM_LABELS], 'fc_digit1_biases')\n",
    "    fc_d2_W = createWeight([NUM_HIDDEN, NUM_LABELS], 'fc_digit2_weights')\n",
    "    fc_d2_B = createBiase([NUM_LABELS], 'fc_digit2_biases')\n",
    "    fc_d3_W = createWeight([NUM_HIDDEN, NUM_LABELS], 'fc_digit3_weights')\n",
    "    fc_d3_B = createBiase([NUM_LABELS], 'fc_digit3_biases')\n",
    "    fc_d4_W = createWeight([NUM_HIDDEN, NUM_LABELS], 'fc_digit4_weights')\n",
    "    fc_d4_B = createBiase([NUM_LABELS], 'fc_digit4_biases')\n",
    "    fc_d5_W = createWeight([NUM_HIDDEN, NUM_LABELS], 'fc_digit5_weights')\n",
    "    fc_d5_B = createBiase([NUM_LABELS], 'fc_digit5_biases')\n",
    "\n",
    "    saver = tf.train.Saver(tf.trainable_variables()) # defaults to saving all variables\n",
    "    \n",
    "    # 数据通过神经网络前向传播\n",
    "    def model(data, train=False):\n",
    "        conv = tf.nn.conv2d(data, conv1_W, strides=[1, 1, 1, 1], padding='SAME')\n",
    "        relu = tf.nn.relu(tf.nn.bias_add(conv, conv1_B))\n",
    "#         relu = tf.nn.local_response_normalization(relu)\n",
    "        pool = tf.nn.max_pool(relu, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')\n",
    "        conv = tf.nn.conv2d(pool, conv2_W, strides=[1, 1, 1, 1], padding='SAME')\n",
    "        relu = tf.nn.relu(tf.nn.bias_add(conv, conv2_B))\n",
    "#         relu = tf.nn.local_response_normalization(relu)\n",
    "        pool = tf.nn.max_pool(relu, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')\n",
    "        conv = tf.nn.conv2d(pool, conv3_W, strides=[1, 1, 1, 1], padding='SAME')\n",
    "        relu = tf.nn.relu(tf.nn.bias_add(conv, conv3_B))\n",
    "#         relu = tf.nn.local_response_normalization(relu)\n",
    "        pool = tf.nn.max_pool(relu, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')\n",
    "        shape = pool.get_shape().as_list()\n",
    "        reshape = tf.reshape(pool, [-1, shape[1]*shape[2]*shape[3]])\n",
    "        hidden = tf.nn.relu(tf.matmul(reshape, fc1_W) + fc1_B)\n",
    "        \n",
    "        if train:\n",
    "            hidden = tf.nn.dropout(hidden, 0.6, seed=SEED)\n",
    "            \n",
    "        logit_d1 = tf.matmul(hidden, fc_d1_W) + fc_d1_B\n",
    "        logit_d2 = tf.matmul(hidden, fc_d2_W) + fc_d2_B\n",
    "        logit_d3 = tf.matmul(hidden, fc_d3_W) + fc_d3_B\n",
    "        logit_d4 = tf.matmul(hidden, fc_d4_W) + fc_d4_B\n",
    "        logit_d5 = tf.matmul(hidden, fc_d5_W) + fc_d5_B\n",
    "        \n",
    "        return logit_d1, logit_d2, logit_d3, logit_d4, logit_d5\n",
    "    \n",
    "    # 由logit计算出标签\n",
    "    def predict(logits):\n",
    "        return tf.transpose(tf.pack([tf.argmax(logits[0], 1), tf.argmax(logits[1], 1), tf.argmax(logits[2], 1), \\\n",
    "                        tf.argmax(logits[3], 1), tf.argmax(logits[4], 1)]))\n",
    "    \n",
    "    def accuracy(predictions, labels):\n",
    "        return tf.reduce_mean(tf.cast(tf.reduce_all(\n",
    "                    tf.equal(predictions, labels), reduction_indices=1\n",
    "                ), tf.float32)) * 100\n",
    "    \n",
    "    logits = model(tf_train_dataset, True)\n",
    "    loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits[0], tf_train_labels[:, 0])) + \\\n",
    "            tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits[1], tf_train_labels[:, 1])) + \\\n",
    "            tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits[2], tf_train_labels[:, 2])) + \\\n",
    "            tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits[3], tf_train_labels[:, 3])) + \\\n",
    "            tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits[4], tf_train_labels[:, 4]))\n",
    "    \n",
    "    regularizers = tf.nn.l2_loss(fc1_W) + \\\n",
    "                tf.nn.l2_loss(fc_d1_W) + \\\n",
    "                tf.nn.l2_loss(fc_d2_W) + \\\n",
    "                tf.nn.l2_loss(fc_d3_W) + \\\n",
    "                tf.nn.l2_loss(fc_d4_W) + \\\n",
    "                tf.nn.l2_loss(fc_d5_W)\n",
    "                \n",
    "    loss += 1e-2 * regularizers\n",
    "    \n",
    "    batch = tf.Variable(0, dtype=tf.float32)\n",
    "    \n",
    "    learning_rate = tf.constant(0.04)\n",
    "    \n",
    "    #decayed_learning_rate = learning_rate *\n",
    "    #                    decay_rate ^ (global_step / decay_steps)\n",
    "#     learning_rate = tf.train.exponential_decay(\n",
    "#         0.04, # Base learning rate\n",
    "#         batch * BATCH_SIZE, # Current index into the dataset\n",
    "#         train_labels.shape[0], # Decay step\n",
    "#         0.95, # Decay rate,\n",
    "#         staircase=True\n",
    "#     )\n",
    "    \n",
    "    optimizer = tf.train.AdagradOptimizer(learning_rate, 0.9).minimize(loss, global_step=batch)\n",
    "    \n",
    "    train_prediction = predict(logits)\n",
    "    valid_prediction = predict(model(tf_valid_dataset))\n",
    "    \n",
    "    train_accuracy = accuracy(train_prediction, tf_train_labels)\n",
    "    valid_accuracy = accuracy(valid_prediction, tf_valid_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "def shuffle_in_unison_inplace(a, b):\n",
    "    # 随机打乱数据，同时保证数据和标签仍然相互对应\n",
    "    assert len(a) == len(b)\n",
    "    p = np.random.permutation(len(a))\n",
    "    return a[p], b[p]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Random shuffle\n",
      "Step 400, \tMinibatch loss 19.0825, Learning rate 0.0400, Minibatch accuracy 29.69%, Validation accuracy 30.70% \n",
      "Random shuffle\n",
      "Step 800, \tMinibatch loss 14.2333, Learning rate 0.0400, Minibatch accuracy 31.25%, Validation accuracy 45.60% \n",
      "Random shuffle\n",
      "Step 1200, \tMinibatch loss 9.9948, Learning rate 0.0400, Minibatch accuracy 62.50%, Validation accuracy 55.18% \n",
      "Random shuffle\n",
      "Step 1600, \tMinibatch loss 7.9433, Learning rate 0.0400, Minibatch accuracy 65.62%, Validation accuracy 60.38% \n",
      "Random shuffle\n",
      "Step 2000, \tMinibatch loss 6.0067, Learning rate 0.0400, Minibatch accuracy 67.19%, Validation accuracy 63.85% \n",
      "Step 2400, \tMinibatch loss 4.8136, Learning rate 0.0400, Minibatch accuracy 71.88%, Validation accuracy 63.49% \n",
      "Random shuffle\n",
      "Step 2800, \tMinibatch loss 4.2582, Learning rate 0.0400, Minibatch accuracy 73.44%, Validation accuracy 66.61% \n",
      "Random shuffle\n",
      "Step 3200, \tMinibatch loss 3.2431, Learning rate 0.0400, Minibatch accuracy 78.12%, Validation accuracy 70.32% \n",
      "Random shuffle\n",
      "Step 3600, \tMinibatch loss 3.1759, Learning rate 0.0400, Minibatch accuracy 68.75%, Validation accuracy 69.84% \n",
      "Random shuffle\n",
      "Step 4000, \tMinibatch loss 2.7947, Learning rate 0.0400, Minibatch accuracy 73.44%, Validation accuracy 71.99% \n",
      "Step 4400, \tMinibatch loss 2.6937, Learning rate 0.0400, Minibatch accuracy 75.00%, Validation accuracy 72.11% \n",
      "Random shuffle\n",
      "Step 4800, \tMinibatch loss 1.9199, Learning rate 0.0400, Minibatch accuracy 85.94%, Validation accuracy 72.83% \n",
      "Random shuffle\n",
      "Step 5200, \tMinibatch loss 2.0668, Learning rate 0.0400, Minibatch accuracy 76.56%, Validation accuracy 72.77% \n",
      "Random shuffle\n",
      "Step 5600, \tMinibatch loss 1.9112, Learning rate 0.0400, Minibatch accuracy 79.69%, Validation accuracy 73.55% \n",
      "Random shuffle\n",
      "Step 6000, \tMinibatch loss 1.5042, Learning rate 0.0400, Minibatch accuracy 90.62%, Validation accuracy 74.87% \n",
      "Step 6400, \tMinibatch loss 1.8986, Learning rate 0.0400, Minibatch accuracy 76.56%, Validation accuracy 74.03% \n",
      "Random shuffle\n",
      "Step 6800, \tMinibatch loss 1.2337, Learning rate 0.0400, Minibatch accuracy 92.19%, Validation accuracy 75.58% \n",
      "Random shuffle\n",
      "Step 7200, \tMinibatch loss 1.2141, Learning rate 0.0400, Minibatch accuracy 87.50%, Validation accuracy 75.28% \n",
      "Random shuffle\n",
      "Step 7600, \tMinibatch loss 1.2389, Learning rate 0.0400, Minibatch accuracy 85.94%, Validation accuracy 76.66% \n",
      "Random shuffle\n",
      "Step 8000, \tMinibatch loss 1.3510, Learning rate 0.0400, Minibatch accuracy 93.75%, Validation accuracy 76.24% \n",
      "Step 8400, \tMinibatch loss 1.7317, Learning rate 0.0400, Minibatch accuracy 81.25%, Validation accuracy 76.72% \n",
      "Random shuffle\n",
      "Step 8800, \tMinibatch loss 1.5642, Learning rate 0.0400, Minibatch accuracy 85.94%, Validation accuracy 76.60% \n",
      "Random shuffle\n",
      "Step 9200, \tMinibatch loss 1.4071, Learning rate 0.0400, Minibatch accuracy 93.75%, Validation accuracy 76.18% \n",
      "Random shuffle\n",
      "Step 9600, \tMinibatch loss 0.9907, Learning rate 0.0400, Minibatch accuracy 96.88%, Validation accuracy 76.66% \n",
      "Random shuffle\n",
      "Step 10000, \tMinibatch loss 1.0777, Learning rate 0.0400, Minibatch accuracy 87.50%, Validation accuracy 75.58% \n"
     ]
    }
   ],
   "source": [
    "SHOW_STATE_AFTER = 400\n",
    "\n",
    "train_accuracies = []\n",
    "valid_accuracies = []\n",
    "\n",
    "best_valid_accuracy = 0\n",
    "best_accuracy_step = 0\n",
    "stop_after_step_no_progress = 5200\n",
    "\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "    tf.initialize_all_variables().run()\n",
    "    print('Initialized')\n",
    "    offset = 0\n",
    "    \n",
    "    step = 0\n",
    "    while True:\n",
    "        step += 1\n",
    "        # 扫描一遍训练数据后，随机打乱训练数据的顺序\n",
    "        if(offset == 0):\n",
    "            train_dataset, train_labels = shuffle_in_unison_inplace(train_dataset, train_labels)\n",
    "            print('Random shuffle')\n",
    "        feed_dict = {\n",
    "            tf_train_dataset : train_dataset[offset:(offset + BATCH_SIZE)], \n",
    "            tf_train_labels : train_labels[offset:(offset + BATCH_SIZE)]\n",
    "        }\n",
    "        offset += BATCH_SIZE\n",
    "        if offset+BATCH_SIZE > train_labels.shape[0]:\n",
    "            offset = 0\n",
    "        _, l, lr = session.run([optimizer, loss, learning_rate], feed_dict=feed_dict)\n",
    "        \n",
    "        \n",
    "        if (step % SHOW_STATE_AFTER == 0):\n",
    "            # 训练集准确率和交叉验证集准确率只需要在打印输出的时候计算\n",
    "            train_acc, valid_acc = session.run([train_accuracy, valid_accuracy], feed_dict=feed_dict)\n",
    "            \n",
    "            # 星号表示交叉验证集上的准确率有提升\n",
    "            improved_str = ''\n",
    "#             if valid_acc > best_valid_accuracy:\n",
    "#                 saver.save(session, 'multi22.ckpt', write_meta_graph=False)\n",
    "#                 best_valid_accuracy = valid_acc\n",
    "#                 best_accuracy_step = step\n",
    "#                 improved_str = '*'\n",
    "            \n",
    "            train_accuracies.append(train_acc)\n",
    "            valid_accuracies.append(valid_acc)\n",
    "            print('Step %d, \\tMinibatch loss %.4f, Learning rate %.4f, Minibatch accuracy %.2f%%, Validation accuracy %.2f%% %s'\\\n",
    "                  % (step, l, lr, train_acc, valid_acc, improved_str))\n",
    "            \n",
    "            # stop_after_step_no_progress步内，交叉验证集准确率没有提升，终止训练\n",
    "#             if step - best_accuracy_step > stop_after_step_no_progress:\n",
    "#                 print(\"Valid accuracy makes no progress in {} steps\".format(stop_after_step_no_progress))\n",
    "#                 break\n",
    "            if step >= 10000:\n",
    "                saver.save(session, 'multi22.ckpt', write_meta_graph=False)\n",
    "                break"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 学习曲线\n",
    "\n",
    "在训练过程中，准确率在训练集和交叉验证集上的变化情况。"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f089c0dca58>"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAtAAAAIACAYAAACraVuzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nmoiIiIjIAP8PJqvYao475+8AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f08b759dd30>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(range(0, step, SHOW_STATE_AFTER), train_accuracies)\n",
    "plt.plot(range(0, step, SHOW_STATE_AFTER), valid_accuracies)\n",
    "plt.title('Learning curve')\n",
    "plt.xlabel('Number of steps')\n",
    "plt.ylabel('Accuracy(%)')\n",
    "plt.legend(['Train accuracy', 'Valid accuracy'], loc='best')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [Root]",
   "language": "python",
   "name": "Python [Root]"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}