Uda_MER.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import os\n",
    "from sklearn.model_selection import train_test_split\n",
    "from keras.preprocessing import image  \n",
    "import cv2\n",
    "import string"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "src_path=r'/input/uda_MER'\n",
    "img_path=os.path.join(src_path,'train')\n",
    "csv_file=pd.read_csv(os.path.join(src_path,'train.csv'))\n",
    "# img path and label\n",
    "paths=[i for i in csv_file['filename']] \n",
    "labels=[i for i in csv_file['label']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Max.Len=11\n",
      "Min.Len=7\n"
     ]
    }
   ],
   "source": [
    "#define characters in the expression\n",
    "chars = '0123456789+-*()=' \n",
    "chars_array = np.array(list(chars))\n",
    "\n",
    "#\n",
    "n_class = len(chars)+1\n",
    "#Max length of the expression\n",
    "len_label=[len(i) for i in labels]\n",
    "len_label.sort()\n",
    "n_seq=len_label[-1]\n",
    "min_seq=len_label[0]\n",
    "print('Max.Len={}'.format(n_seq))\n",
    "print('Min.Len={}'.format(min_seq))\n",
    "#"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "80000 10000 10000\n",
      "train/31546.jpg\n",
      "Example of path: train/31546.jpg\n",
      "Example of label: 7*(6-5)=7\n"
     ]
    }
   ],
   "source": [
    "#split the data set, get 100000 test data \n",
    "img_path_rest, img_path_test, labels_rest, labels_test = train_test_split(paths, labels, test_size=10000, random_state=10)\n",
    "#split the data set, get 100000 valid data\n",
    "img_path_train, img_path_valid, labels_train, labels_valid = train_test_split(img_path_rest, labels_rest, test_size=10000, random_state=10)\n",
    "print(len(img_path_train),len(img_path_valid),len(img_path_test))\n",
    "print(img_path_train[0])\n",
    "print('Example of path: {}'.format(img_path_train[0]))\n",
    "print('Example of label: {}'.format(labels_train[0]))\n",
    "#should be updated"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "#get label length\n",
    "def get_label_len(labels):\n",
    "    length=np.zeros((len(labels),1))\n",
    "    for i,label in enumerate(labels):\n",
    "        length[i] = len(label)\n",
    "    return length\n",
    "\n",
    "#turn a path list into numpy array\n",
    "def tensors_trans(paths):\n",
    "    img_list=[]\n",
    "    for element in paths:\n",
    "        ele=os.path.join(src_path,element)\n",
    "        img=cv2.imread(ele)\n",
    "        img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)\n",
    "        img_gray = np.expand_dims(image.img_to_array(img_gray),axis=0)\n",
    "        img_list.append(img_gray)\n",
    "    img_tensor=np.vstack(img_list)\n",
    "    return img_tensor\n",
    "#label to index reffers to chars\n",
    "def get_index(label):\n",
    "    index = np.zeros(n_seq)\n",
    "    for i, char in enumerate(label):\n",
    "        index[i] = chars.find(char)\n",
    "    return index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "#maybe ues the iterator will be better\n",
    "X_train, X_valid, X_test = [tensors_trans(img_path) for img_path in [img_path_train, img_path_valid, img_path_test]]\n",
    "y_train = np.array([get_index(label) for label in labels_train])\n",
    "y_valid = np.array([get_index(label) for label in labels_valid])\n",
    "y_test = np.array([get_index(label) for label in labels_test])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "label_length_train = get_label_len(labels_train)\n",
    "label_length_valid = get_label_len(labels_valid)\n",
    "label_length_test = get_label_len(labels_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "plt.subplot();\n",
    "plt.hist(x=[label_length_train,label_length_valid,label_length_test],\n",
    "         bins=[7,8,9,10,11,12],align='left',density=1,);\n",
    "plt.title('Label length distribution');\n",
    "plt.legend(['train','valid','test']);\n",
    "plt.xlabel('Label length');\n",
    "plt.ylabel('ratio');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "h = cv2.equalizeHist(gray)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "show_pic(img_path_train[0],labels_train[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "import keras.backend as K\n",
    "from keras.models import *\n",
    "from keras.layers import *\n",
    "from keras.regularizers import L1L2\n",
    "import matplotlib.pyplot as plt\n",
    "import re\n",
    "from keras.callbacks import *\n",
    "from keras.optimizers import SGD\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "height, width, channel = X_train.shape[1:]\n",
    "input_tensor = Input(shape=[height, width, channel])\n",
    "x = input_tensor\n",
    "\n",
    "x = Conv2D(32*2**0, 3, strides=1, padding='same', activation='relu',kernel_initializer='he_normal')(x)\n",
    "x = MaxPooling2D((2, 2), strides=2, padding='same')(x)\n",
    "x = Dropout(rate=0.25)(x)\n",
    "x = Conv2D(32*2**1, 3, strides=1, padding='same', activation='relu',kernel_initializer='he_normal')(x)\n",
    "x = MaxPooling2D((2, 2), strides=2, padding='same')(x)\n",
    "x = Dropout(rate=0.25)(x)\n",
    "x = Conv2D(32*2**2, 3, strides=1, padding='same', activation='relu',kernel_initializer='he_normal')(x)\n",
    "x = MaxPooling2D((2, 2), strides=2, padding='same')(x)\n",
    "x = Dropout(rate=0.25)(x)\n",
    "x = Conv2D(256, 3, strides=1, padding='same', activation='relu',kernel_initializer='he_normal')(x)\n",
    "x = BatchNormalization(axis=-1,momentum=0.99, epsilon=0.001)(x)\n",
    "\n",
    "x = Conv2D(512, 3, strides=1, padding='same', activation='relu')(x)\n",
    "x = BatchNormalization(axis=-1,momentum=0.99, epsilon=0.001)(x)\n",
    "\n",
    "x = MaxPooling2D((1, 2), strides=2, padding='same')(x)\n",
    "x = Dropout(rate=0.5)(x)\n",
    "\n",
    "x = Permute((2,1,3))(x)\n",
    "\n",
    "x = TimeDistributed(Flatten())(x)\n",
    "\n",
    "x = Bidirectional(CuDNNLSTM(64, return_sequences=True))(x)\n",
    "x = Bidirectional(CuDNNLSTM(64, return_sequences=True))(x)\n",
    "\n",
    "y_pred = Dense(n_class,activation='softmax')(x)\n",
    "\n",
    "base_model = Model(inputs=input_tensor,outputs = y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "chars_space = chars + u' '#unicode\n",
    "def decode(pred):\n",
    "    pred = pred.argmax(axis=2)#get the most possible char\n",
    "    orig_labels = []\n",
    "    new_labels = []\n",
    "    for i, code in enumerate(pred):\n",
    "        orig_labels.append(''.join([chars_space[x] for x in code]))\n",
    "    for string in orig_labels:\n",
    "        new_labels.append(re.sub(r'\\s','',re.sub(r'(.)\\1+', r'\\1', string)))#remove uncessary space and ```                              \n",
    "    return [np.array(orig_labels), np.array(new_labels)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "def evalute_valid(model):\n",
    "    pred = model.predict(X_valid)\n",
    "    org_labels, new_labels = decode(pred)\n",
    "    return (new_labels==labels_valid).mean()*100#valid data set accuracy\n",
    "\n",
    "class Evaluate(Callback):\n",
    "    def __init__(self):\n",
    "        self.accs = []\n",
    "    \n",
    "    def on_epoch_end(self, epoch, logs=None):\n",
    "        acc = evalute_valid(base_model)\n",
    "        self.accs.append(acc)\n",
    "        print('valid acc: %f%%'%acc)\n",
    "\n",
    "evaluator = Evaluate()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_9 (InputLayer)            (None, 64, 300, 1)   0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_71 (Conv2D)              (None, 64, 300, 32)  320         input_9[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_28 (MaxPooling2D) (None, 32, 150, 32)  0           conv2d_71[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dropout_13 (Dropout)            (None, 32, 150, 32)  0           max_pooling2d_28[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_72 (Conv2D)              (None, 32, 150, 64)  18496       dropout_13[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_29 (MaxPooling2D) (None, 16, 75, 64)   0           conv2d_72[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dropout_14 (Dropout)            (None, 16, 75, 64)   0           max_pooling2d_29[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_73 (Conv2D)              (None, 16, 75, 128)  73856       dropout_14[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_30 (MaxPooling2D) (None, 8, 38, 128)   0           conv2d_73[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dropout_15 (Dropout)            (None, 8, 38, 128)   0           max_pooling2d_30[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_74 (Conv2D)              (None, 8, 38, 256)   295168      dropout_15[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_27 (BatchNo (None, 8, 38, 256)   1024        conv2d_74[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_75 (Conv2D)              (None, 8, 38, 512)   1180160     batch_normalization_27[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_28 (BatchNo (None, 8, 38, 512)   2048        conv2d_75[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_31 (MaxPooling2D) (None, 4, 19, 512)   0           batch_normalization_28[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dropout_16 (Dropout)            (None, 4, 19, 512)   0           max_pooling2d_31[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "permute_8 (Permute)             (None, 19, 4, 512)   0           dropout_16[0][0]                 \n",
      "__________________________________________________________________________________________________\n",
      "time_distributed_6 (TimeDistrib (None, 19, 2048)     0           permute_8[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_12 (Bidirectional (None, 19, 128)      1082368     time_distributed_6[0][0]         \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_13 (Bidirectional (None, 19, 128)      99328       bidirectional_12[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "dense_12 (Dense)                (None, 19, 17)       2193        bidirectional_13[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "the_labels (InputLayer)         (None, None)         0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_length (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "label_length (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "ctc (Lambda)                    (None, 1)            0           dense_12[0][0]                   \n",
      "                                                                 the_labels[0][0]                 \n",
      "                                                                 input_length[0][0]               \n",
      "                                                                 label_length[0][0]               \n",
      "==================================================================================================\n",
      "Total params: 2,754,961\n",
      "Trainable params: 2,753,425\n",
      "Non-trainable params: 1,536\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "#  CTC-Loss function\n",
    "def ctc_lambda_func(args):\n",
    "    y_pred, labels, input_length, label_length = args\n",
    "    return K.ctc_batch_cost(labels, y_pred, input_length, label_length)\n",
    "labels = Input(name='the_labels', shape=[None], dtype='float32')\n",
    "input_length = Input(name='input_length', shape=[1], dtype='int64')\n",
    "label_length = Input(name='label_length', shape=[1], dtype='int64')\n",
    "loss_out = Lambda(ctc_lambda_func, output_shape=(1,), name='ctc')([y_pred, labels, input_length, label_length])\n",
    "\n",
    "model = Model(inputs=[input_tensor, labels, input_length, label_length], outputs=[loss_out])\n",
    "model.compile(loss={'ctc': lambda y_true, y_pred: y_pred}, optimizer='adam')# \n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_file_path = '/output/model01.hd5'#weight file\n",
    "\n",
    "#base_model.load_weights(model_file_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 80000 samples, validate on 10000 samples\n",
      "Epoch 1/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 13.0448 - val_loss: 39.5771\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 39.57708, saving model to /output/model01.hd5\n",
      "valid acc: 0.000000%\n",
      "Epoch 2/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 1.1339 - val_loss: 0.5028\n",
      "\n",
      "Epoch 00002: val_loss improved from 39.57708 to 0.50276, saving model to /output/model01.hd5\n",
      "valid acc: 89.570000%\n",
      "Epoch 3/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.5184 - val_loss: 0.2570\n",
      "\n",
      "Epoch 00003: val_loss improved from 0.50276 to 0.25700, saving model to /output/model01.hd5\n",
      "valid acc: 94.510000%\n",
      "Epoch 4/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.3516 - val_loss: 0.2008\n",
      "\n",
      "Epoch 00004: val_loss improved from 0.25700 to 0.20075, saving model to /output/model01.hd5\n",
      "valid acc: 95.740000%\n",
      "Epoch 5/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.2713 - val_loss: 0.1531\n",
      "\n",
      "Epoch 00005: val_loss improved from 0.20075 to 0.15308, saving model to /output/model01.hd5\n",
      "valid acc: 96.720000%\n",
      "Epoch 6/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.2255 - val_loss: 0.1317\n",
      "\n",
      "Epoch 00006: val_loss improved from 0.15308 to 0.13167, saving model to /output/model01.hd5\n",
      "valid acc: 97.430000%\n",
      "Epoch 7/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1869 - val_loss: 0.1475\n",
      "\n",
      "Epoch 00007: val_loss did not improve from 0.13167\n",
      "valid acc: 96.640000%\n",
      "Epoch 8/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1635 - val_loss: 0.0923\n",
      "\n",
      "Epoch 00008: val_loss improved from 0.13167 to 0.09229, saving model to /output/model01.hd5\n",
      "valid acc: 98.030000%\n",
      "Epoch 9/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1426 - val_loss: 0.1111\n",
      "\n",
      "Epoch 00009: val_loss did not improve from 0.09229\n",
      "valid acc: 97.920000%\n",
      "Epoch 10/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1319 - val_loss: 0.0896\n",
      "\n",
      "Epoch 00010: val_loss improved from 0.09229 to 0.08963, saving model to /output/model01.hd5\n",
      "valid acc: 98.210000%\n",
      "Epoch 11/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1251 - val_loss: 0.0678\n",
      "\n",
      "Epoch 00011: val_loss improved from 0.08963 to 0.06778, saving model to /output/model01.hd5\n",
      "valid acc: 98.400000%\n",
      "Epoch 12/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1138 - val_loss: 0.0819\n",
      "\n",
      "Epoch 00012: val_loss did not improve from 0.06778\n",
      "valid acc: 98.060000%\n",
      "Epoch 13/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1035 - val_loss: 0.0802\n",
      "\n",
      "Epoch 00013: val_loss did not improve from 0.06778\n",
      "valid acc: 98.390000%\n",
      "Epoch 14/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.1008 - val_loss: 0.0648\n",
      "\n",
      "Epoch 00014: val_loss improved from 0.06778 to 0.06477, saving model to /output/model01.hd5\n",
      "valid acc: 98.590000%\n",
      "Epoch 15/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0991 - val_loss: 0.0619\n",
      "\n",
      "Epoch 00015: val_loss improved from 0.06477 to 0.06190, saving model to /output/model01.hd5\n",
      "valid acc: 98.730000%\n",
      "Epoch 16/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0927 - val_loss: 0.0601\n",
      "\n",
      "Epoch 00016: val_loss improved from 0.06190 to 0.06014, saving model to /output/model01.hd5\n",
      "valid acc: 98.600000%\n",
      "Epoch 17/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0828 - val_loss: 0.0658\n",
      "\n",
      "Epoch 00017: val_loss did not improve from 0.06014\n",
      "valid acc: 98.610000%\n",
      "Epoch 18/400\n",
      "80000/80000 [==============================] - 115s 1ms/step - loss: 0.0871 - val_loss: 6.1963\n",
      "\n",
      "Epoch 00018: val_loss did not improve from 0.06014\n",
      "valid acc: 9.860000%\n",
      "Epoch 19/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0818 - val_loss: 0.0472\n",
      "\n",
      "Epoch 00019: val_loss improved from 0.06014 to 0.04718, saving model to /output/model01.hd5\n",
      "valid acc: 98.770000%\n",
      "Epoch 20/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0723 - val_loss: 0.0581\n",
      "\n",
      "Epoch 00020: val_loss did not improve from 0.04718\n",
      "valid acc: 98.660000%\n",
      "Epoch 21/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0732 - val_loss: 0.0509\n",
      "\n",
      "Epoch 00021: val_loss did not improve from 0.04718\n",
      "valid acc: 98.810000%\n",
      "Epoch 22/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0812 - val_loss: 0.0570\n",
      "\n",
      "Epoch 00022: val_loss did not improve from 0.04718\n",
      "valid acc: 98.740000%\n",
      "Epoch 23/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0771 - val_loss: 0.0659\n",
      "\n",
      "Epoch 00023: val_loss did not improve from 0.04718\n",
      "valid acc: 98.710000%\n",
      "Epoch 24/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0652 - val_loss: 0.0460\n",
      "\n",
      "Epoch 00024: val_loss improved from 0.04718 to 0.04598, saving model to /output/model01.hd5\n",
      "valid acc: 98.930000%\n",
      "Epoch 25/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0672 - val_loss: 0.0660\n",
      "\n",
      "Epoch 00025: val_loss did not improve from 0.04598\n",
      "valid acc: 98.800000%\n",
      "Epoch 26/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0661 - val_loss: 0.0479\n",
      "\n",
      "Epoch 00026: val_loss did not improve from 0.04598\n",
      "valid acc: 98.800000%\n",
      "Epoch 27/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0769 - val_loss: 0.0490\n",
      "\n",
      "Epoch 00027: val_loss did not improve from 0.04598\n",
      "valid acc: 98.900000%\n",
      "Epoch 28/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0629 - val_loss: 0.0429\n",
      "\n",
      "Epoch 00028: val_loss improved from 0.04598 to 0.04286, saving model to /output/model01.hd5\n",
      "valid acc: 99.050000%\n",
      "Epoch 29/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0551 - val_loss: 0.0424\n",
      "\n",
      "Epoch 00029: val_loss improved from 0.04286 to 0.04236, saving model to /output/model01.hd5\n",
      "valid acc: 99.080000%\n",
      "Epoch 30/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0609 - val_loss: 1.0487\n",
      "\n",
      "Epoch 00030: val_loss did not improve from 0.04236\n",
      "valid acc: 78.800000%\n",
      "Epoch 31/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0578 - val_loss: 0.0476\n",
      "\n",
      "Epoch 00031: val_loss did not improve from 0.04236\n",
      "valid acc: 98.970000%\n",
      "Epoch 32/400\n",
      "80000/80000 [==============================] - 108s 1ms/step - loss: 0.0562 - val_loss: 0.0500\n",
      "\n",
      "Epoch 00032: val_loss did not improve from 0.04236\n",
      "valid acc: 99.020000%\n",
      "Epoch 33/400\n",
      "80000/80000 [==============================] - 108s 1ms/step - loss: 0.0508 - val_loss: 0.0518\n",
      "\n",
      "Epoch 00033: val_loss did not improve from 0.04236\n",
      "valid acc: 98.780000%\n",
      "Epoch 34/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0508 - val_loss: 0.0416\n",
      "\n",
      "Epoch 00034: val_loss improved from 0.04236 to 0.04156, saving model to /output/model01.hd5\n",
      "valid acc: 99.130000%\n",
      "Epoch 35/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0484 - val_loss: 0.0449\n",
      "\n",
      "Epoch 00035: val_loss did not improve from 0.04156\n",
      "valid acc: 99.050000%\n",
      "Epoch 36/400\n",
      "80000/80000 [==============================] - 108s 1ms/step - loss: 0.0504 - val_loss: 0.1480\n",
      "\n",
      "Epoch 00036: val_loss did not improve from 0.04156\n",
      "valid acc: 96.480000%\n",
      "Epoch 37/400\n",
      "80000/80000 [==============================] - 108s 1ms/step - loss: 0.0521 - val_loss: 0.0567\n",
      "\n",
      "Epoch 00037: val_loss did not improve from 0.04156\n",
      "valid acc: 98.900000%\n",
      "Epoch 38/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0478 - val_loss: 0.0370\n",
      "\n",
      "Epoch 00038: val_loss improved from 0.04156 to 0.03698, saving model to /output/model01.hd5\n",
      "valid acc: 99.050000%\n",
      "Epoch 39/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0456 - val_loss: 0.0341\n",
      "\n",
      "Epoch 00039: val_loss improved from 0.03698 to 0.03413, saving model to /output/model01.hd5\n",
      "valid acc: 99.210000%\n",
      "Epoch 40/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0426 - val_loss: 0.0407\n",
      "\n",
      "Epoch 00040: val_loss did not improve from 0.03413\n",
      "valid acc: 99.160000%\n",
      "Epoch 41/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0537 - val_loss: 0.0340\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 00041: val_loss improved from 0.03413 to 0.03396, saving model to /output/model01.hd5\n",
      "valid acc: 99.090000%\n",
      "Epoch 42/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0485 - val_loss: 0.0549\n",
      "\n",
      "Epoch 00042: val_loss did not improve from 0.03396\n",
      "valid acc: 98.630000%\n",
      "Epoch 43/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0421 - val_loss: 0.0461\n",
      "\n",
      "Epoch 00044: val_loss did not improve from 0.02951\n",
      "valid acc: 99.190000%\n",
      "Epoch 45/400\n",
      "80000/80000 [==============================] - 112s 1ms/step - loss: 0.0442 - val_loss: 0.0362\n",
      "\n",
      "Epoch 00045: val_loss did not improve from 0.02951\n",
      "valid acc: 99.130000%\n",
      "Epoch 46/400\n",
      "80000/80000 [==============================] - 109s 1ms/step - loss: 0.0443 - val_loss: 0.0372\n",
      "\n",
      "Epoch 00046: val_loss did not improve from 0.02951\n",
      "valid acc: 99.210000%\n",
      "Epoch 47/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0463 - val_loss: 0.0463\n",
      "\n",
      "Epoch 00047: val_loss did not improve from 0.02951\n",
      "valid acc: 98.830000%\n",
      "Epoch 48/400\n",
      "80000/80000 [==============================] - 115s 1ms/step - loss: 0.0398 - val_loss: 0.0346\n",
      "\n",
      "Epoch 00048: val_loss did not improve from 0.02951\n",
      "valid acc: 99.280000%\n",
      "Epoch 49/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0383 - val_loss: 0.0335\n",
      "\n",
      "Epoch 00049: val_loss did not improve from 0.02951\n",
      "valid acc: 99.230000%\n",
      "Epoch 50/400\n",
      "80000/80000 [==============================] - 115s 1ms/step - loss: 0.0418 - val_loss: 0.0419\n",
      "\n",
      "Epoch 00050: val_loss did not improve from 0.02951\n",
      "valid acc: 99.020000%\n",
      "Epoch 51/400\n",
      "80000/80000 [==============================] - 115s 1ms/step - loss: 0.0433 - val_loss: 0.0294\n",
      "\n",
      "Epoch 00051: val_loss improved from 0.02951 to 0.02938, saving model to /output/model01.hd5\n",
      "valid acc: 99.200000%\n",
      "Epoch 52/400\n",
      "80000/80000 [==============================] - 115s 1ms/step - loss: 0.0416 - val_loss: 0.0411\n",
      "\n",
      "Epoch 00052: val_loss did not improve from 0.02938\n",
      "valid acc: 99.090000%\n",
      "Epoch 53/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0394 - val_loss: 0.0328\n",
      "\n",
      "Epoch 00054: val_loss did not improve from 0.02938\n",
      "valid acc: 99.310000%\n",
      "Epoch 55/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0374 - val_loss: 0.0356\n",
      "\n",
      "Epoch 00055: val_loss did not improve from 0.02938\n",
      "valid acc: 99.190000%\n",
      "Epoch 56/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0381 - val_loss: 0.0423\n",
      "\n",
      "Epoch 00056: val_loss did not improve from 0.02938\n",
      "valid acc: 99.220000%\n",
      "Epoch 57/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0345 - val_loss: 0.0604\n",
      "\n",
      "Epoch 00057: val_loss did not improve from 0.02938\n",
      "valid acc: 98.940000%\n",
      "Epoch 58/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0392 - val_loss: 0.0311\n",
      "\n",
      "Epoch 00058: val_loss did not improve from 0.02938\n",
      "valid acc: 99.420000%\n",
      "Epoch 59/400\n",
      "80000/80000 [==============================] - 115s 1ms/step - loss: 0.0424 - val_loss: 0.0346\n",
      "\n",
      "Epoch 00060: val_loss did not improve from 0.02938\n",
      "valid acc: 99.280000%\n",
      "Epoch 61/400\n",
      "80000/80000 [==============================] - 114s 1ms/step - loss: 0.0314 - val_loss: 0.0328\n",
      "\n",
      "Epoch 00061: val_loss did not improve from 0.02938\n",
      "valid acc: 99.360000%\n",
      "Epoch 00061: early stopping\n"
     ]
    }
   ],
   "source": [
    "checkpointer = ModelCheckpoint(filepath=model_file_path, \n",
    "                               verbose=1, monitor='val_loss', save_best_only=True)\n",
    "esp = EarlyStopping(monitor='val_loss', patience=10, verbose=1)\n",
    "tb = TensorBoard(log_dir='./output', update_freq='epoch')\n",
    "\n",
    "#\n",
    "input_len_to_ctc = np.array(np.ones((len(X_train),1))*base_model.output_shape[1])\n",
    "label_len_to_ctc = label_length_train\n",
    "\n",
    "input_len_to_ctc2 = np.array(np.ones((len(X_valid),1))*base_model.output_shape[1])\n",
    "label_len_to_ctc2 = label_length_valid\n",
    "\n",
    "\n",
    "fited = model.fit(x=[X_train, y_train, input_len_to_ctc, label_len_to_ctc], \n",
    "                  y=np.ones(len(y_train)),\n",
    "                  validation_data=([X_valid, y_valid, input_len_to_ctc2, label_len_to_ctc2], np.ones(len(y_valid))),\n",
    "                  batch_size=200, epochs=400, callbacks=[checkpointer, evaluator, esp, tb], verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "test_acc: 99.26%\n"
     ]
    }
   ],
   "source": [
    "base_model.load_weights(model_file_path)\n",
    "pred = base_model.predict(X_test)\n",
    "sequence, new_labels = decode(pred)\n",
    "test_acc = (new_labels==labels_test).mean()*100\n",
    "print('test_acc: {}%'.format(round(test_acc,2)))\n",
    "#1round acc:98.09%\n",
    "#2round acc:97.05%\n",
    "#```\n",
    "#finial"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.save('model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'loss'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(fited.history.keys())\n",
    "# loss during the training process\n",
    "plt.plot(fited.history['loss'])\n",
    "plt.plot(fited.history['val_loss'])\n",
    "plt.title('Model Loss')\n",
    "plt.ylabel('Loss')\n",
    "plt.xlabel('Epoch')\n",
    "plt.legend(['train', 'test'], loc='upper right')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "acc = evaluator.accs\n",
    "plt.plot(acc)\n",
    "plt.title('Model Acc')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.xlabel('Epoch')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "print(img_path_test[:3])\n",
    "print(sequence[:3])\n",
    "print(new_labels[:3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [default]",
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}