Intent_Classification_Keras_Glove.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Intent classification using Convolutional Neural Networks"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Importing libraries and setting parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import sys\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.utils import to_categorical\n",
    "from keras.layers import Dense, Input, Flatten\n",
    "from keras.layers import Reshape, Dropout, Concatenate\n",
    "from keras.layers import Conv2D, MaxPool2D, Embedding\n",
    "from keras.models import Model\n",
    "\n",
    "BASE_DIR = 'C:/Users/username/Documents/chatbot/textcnn'\n",
    "GLOVE_DIR = BASE_DIR + '/glove.6B/'\n",
    "TEXT_DATA_DIR = BASE_DIR + '/data/processed_data'\n",
    "MAX_SEQUENCE_LENGTH = 50 # Maximum number of words in a sentence\n",
    "MAX_NB_WORDS = 20000 # Vocabulary size\n",
    "EMBEDDING_DIM = 100 # Dimensions of Glove word vectors \n",
    "VALIDATION_SPLIT = 0.10"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### loading and processing data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'AddToPlaylist': 0, 'BookRestaurant': 1, 'GetWeather': 2, 'RateBook': 3, 'SearchCreativeWork': 4, 'SearchScreeningEvent': 5}\n"
     ]
    }
   ],
   "source": [
    "texts = np.load(open('data/train_text.npy', 'rb')).tolist()\n",
    "labels = np.load(open('data/train_label.npy', 'rb')).tolist()\n",
    "\n",
    "test_texts = np.load(open('data/test_text.npy', 'rb')).tolist()\n",
    "test_labels = np.load(open('data/test_label.npy', 'rb')).tolist()\n",
    "\n",
    "# Label Encoding: converting text labels into the integers\n",
    "from sklearn import preprocessing\n",
    "\n",
    "train_le = preprocessing.LabelEncoder()\n",
    "train_le.fit(labels)\n",
    "\n",
    "from sklearn.externals import joblib #Saving label encoder, we will need it later\n",
    "joblib.dump(train_le, 'data/label_encoder.pkl') \n",
    "\n",
    "labels = train_le.transform(labels)\n",
    "test_labels = train_le.transform(test_labels)\n",
    "\n",
    "label_dict = dict(zip(list(train_le.classes_), train_le.transform(list(train_le.classes_))))\n",
    "print(label_dict)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Tokenizing text and creating word index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 9869 unique tokens.\n",
      "Shape of data tensor: (11784, 50)\n",
      "Shape of label tensor: (11784, 6)\n",
      "Preparing embedding matrix.\n"
     ]
    }
   ],
   "source": [
    "tokenizer = Tokenizer(num_words=MAX_NB_WORDS)\n",
    "tokenizer.fit_on_texts(texts)\n",
    "sequences = tokenizer.texts_to_sequences(texts)\n",
    "\n",
    "word_index = tokenizer.word_index\n",
    "print('Found %s unique tokens.' % len(word_index))\n",
    "\n",
    "data = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH)\n",
    "\n",
    "labels = to_categorical(np.asarray(labels))\n",
    "print('Shape of data tensor:', data.shape)\n",
    "print('Shape of label tensor:', labels.shape)\n",
    "\n",
    "# split the data into a training set and a validation set\n",
    "indices = np.arange(data.shape[0])\n",
    "np.random.shuffle(indices)\n",
    "data = data[indices]\n",
    "labels = labels[indices]\n",
    "num_validation_samples = int(VALIDATION_SPLIT * data.shape[0])\n",
    "\n",
    "x_train = data[:-num_validation_samples]\n",
    "y_train = labels[:-num_validation_samples]\n",
    "x_val = data[-num_validation_samples:]\n",
    "y_val = labels[-num_validation_samples:]\n",
    "\n",
    "print('Preparing embedding matrix.')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Creating embedding matrix from Glove vectors."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    " Specifically, we will use the 100-dimensional GloVe embeddings of 400k words computed on a 2014 dump of English Wikipedia. You can download them from http://nlp.stanford.edu/data/glove.6B.zip and extract it under the directory, BASE_DIR + '/glove.6B/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Indexing word vectors.\n",
      "Found 400000 word vectors.\n"
     ]
    }
   ],
   "source": [
    "print('Indexing word vectors.')\n",
    "\n",
    "embeddings_index = {}\n",
    "f = open(os.path.join(GLOVE_DIR, 'glove.6B.100d.txt'), encoding='utf-8')\n",
    "for line in f:\n",
    "    values = line.split()\n",
    "    word = values[0]\n",
    "    coefs = np.asarray(values[1:], dtype='float32')\n",
    "    embeddings_index[word] = coefs\n",
    "f.close()\n",
    "\n",
    "print('Found %s word vectors.' % len(embeddings_index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_words = min(MAX_NB_WORDS, len(word_index))\n",
    "embedding_matrix = np.zeros((len(word_index) + 1, EMBEDDING_DIM))\n",
    "for word, i in word_index.items():\n",
    "    embedding_vector = embeddings_index.get(word)\n",
    "    if embedding_vector is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix[i] = embedding_vector"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Model Creation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We will now try to build model described in \"Convolutional Neural Networks for Sentence Classification\n",
    "\", https://arxiv.org/abs/1408.5882\n",
    "\n",
    "It is highly recommended to read, \"A Sensitivity Analysis of (and Practitioners' Guide to) Convolutional Neural Networks for Sentence Classification\", https://arxiv.org/abs/1510.03820 to get an intuition behind the various hyperparameters chosen to build our model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Creating Model...\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_2 (InputLayer)            (None, 50)           0                                            \n",
      "__________________________________________________________________________________________________\n",
      "embedding_2 (Embedding)         (None, 50, 100)      987000      input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "reshape_2 (Reshape)             (None, 50, 100, 1)   0           embedding_2[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_4 (Conv2D)               (None, 49, 1, 512)   102912      reshape_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_5 (Conv2D)               (None, 48, 1, 512)   154112      reshape_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "conv2d_6 (Conv2D)               (None, 46, 1, 512)   256512      reshape_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2D)  (None, 1, 1, 512)    0           conv2d_4[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_5 (MaxPooling2D)  (None, 1, 1, 512)    0           conv2d_5[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_6 (MaxPooling2D)  (None, 1, 1, 512)    0           conv2d_6[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_2 (Concatenate)     (None, 3, 1, 512)    0           max_pooling2d_4[0][0]            \n",
      "                                                                 max_pooling2d_5[0][0]            \n",
      "                                                                 max_pooling2d_6[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "flatten_2 (Flatten)             (None, 1536)         0           concatenate_2[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dropout_2 (Dropout)             (None, 1536)         0           flatten_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 6)            9222        dropout_2[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 1,509,758\n",
      "Trainable params: 522,758\n",
      "Non-trainable params: 987,000\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "filter_sizes = [2,3,5]\n",
    "num_filters = 512\n",
    "drop = 0.5\n",
    "\n",
    "print(\"Creating Model...\")\n",
    "inputs = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')\n",
    "embedding = Embedding(input_dim=len(word_index) + 1, output_dim=EMBEDDING_DIM, weights=[embedding_matrix],\n",
    "                      input_length=MAX_SEQUENCE_LENGTH, trainable=False)(inputs)\n",
    "reshape = Reshape((MAX_SEQUENCE_LENGTH,EMBEDDING_DIM,1))(embedding)\n",
    "\n",
    "conv_0 = Conv2D(num_filters, kernel_size=(filter_sizes[0], EMBEDDING_DIM), padding='valid', kernel_initializer='normal', activation='relu')(reshape)\n",
    "conv_1 = Conv2D(num_filters, kernel_size=(filter_sizes[1], EMBEDDING_DIM), padding='valid', kernel_initializer='normal', activation='relu')(reshape)\n",
    "conv_2 = Conv2D(num_filters, kernel_size=(filter_sizes[2], EMBEDDING_DIM), padding='valid', kernel_initializer='normal', activation='relu')(reshape)\n",
    "\n",
    "maxpool_0 = MaxPool2D(pool_size=(MAX_SEQUENCE_LENGTH - filter_sizes[0] + 1, 1), strides=(1,1), padding='valid')(conv_0)\n",
    "maxpool_1 = MaxPool2D(pool_size=(MAX_SEQUENCE_LENGTH - filter_sizes[1] + 1, 1), strides=(1,1), padding='valid')(conv_1)\n",
    "maxpool_2 = MaxPool2D(pool_size=(MAX_SEQUENCE_LENGTH - filter_sizes[2] + 1, 1), strides=(1,1), padding='valid')(conv_2)\n",
    "\n",
    "concatenated_tensor = Concatenate(axis=1)([maxpool_0, maxpool_1, maxpool_2])\n",
    "flatten = Flatten()(concatenated_tensor)\n",
    "dropout = Dropout(drop)(flatten)\n",
    "preds = Dense(len(label_dict), activation='softmax')(dropout)\n",
    "\n",
    "# this creates a model that includes inputs and outputs\n",
    "model = Model(inputs=inputs, outputs=preds)\n",
    "\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['acc'])\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 10606 samples, validate on 1178 samples\n",
      "Epoch 1/10\n",
      "10606/10606 [==============================] - 12s 1ms/step - loss: 0.2478 - acc: 0.9183 - val_loss: 0.0527 - val_acc: 0.9881\n",
      "Epoch 2/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0564 - acc: 0.9839 - val_loss: 0.0370 - val_acc: 0.9898\n",
      "Epoch 3/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0325 - acc: 0.9911 - val_loss: 0.0319 - val_acc: 0.9898\n",
      "Epoch 4/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0201 - acc: 0.9952 - val_loss: 0.0235 - val_acc: 0.9932\n",
      "Epoch 5/10\n",
      "10606/10606 [==============================] - 12s 1ms/step - loss: 0.0147 - acc: 0.9959 - val_loss: 0.0225 - val_acc: 0.9915\n",
      "Epoch 6/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0117 - acc: 0.9968 - val_loss: 0.0278 - val_acc: 0.9924\n",
      "Epoch 7/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0088 - acc: 0.9975 - val_loss: 0.0262 - val_acc: 0.9907\n",
      "Epoch 8/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0053 - acc: 0.9991 - val_loss: 0.0221 - val_acc: 0.9915\n",
      "Epoch 9/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0071 - acc: 0.9975 - val_loss: 0.0182 - val_acc: 0.9932\n",
      "Epoch 10/10\n",
      "10606/10606 [==============================] - 11s 1ms/step - loss: 0.0086 - acc: 0.9978 - val_loss: 0.0249 - val_acc: 0.9932\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(x_train, y_train,\n",
    "          batch_size=64,\n",
    "          epochs=10,\n",
    "          validation_data=(x_val, y_val))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Wow! That looks amazing! 99 % and 98 % accuracy on train and validation dataset respectively within a minute after 3 epochs!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEWCAYAAAB1xKBvAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x2ed20f28>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEWCAYAAAB1xKBvAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xt4XHd95/H3V6PR/TKyJDuxNYqdq+Mk9ig4JiEpJYRA\nnBTSFhpuSVtaarIPFGghS9gCXbrbhWfLUqANhEC9LYUkpQHaUBxyIQnQDSGxYyd2HN+SOJZsx5at\nm3WXRt/94xxZI1l2RrJGo9F8Xs+jZ86cy8xX89jz0e/8zvn9zN0RERF5LQXZLkBERHKDAkNERNKi\nwBARkbQoMEREJC0KDBERSYsCQ0RE0qLAEJkBZvaPZvY/09x3r5m95XRfR2S2KTBERCQtCgwREUmL\nAkPyRngq6DYze87MeszsH8xskZk9YGbHzOwRM6tJ2f8dZva8mXWY2eNmdmHKtiYzeyY87l+Akgnv\n9VtmtiU89gkzWznNmv/EzPaYWZuZ3W9mi8P1ZmZ/a2aHzazLzLaa2cXhtuvNbHtY234z++S0PjCR\nCRQYkm/eCVwLnA+8HXgA+G9APcH/h48CmNn5wD3Ax8NtG4Afm1mRmRUB/wb8M7AA+NfwdQmPbQLW\nAx8CaoFvAvebWfFUCjWzNwNfAG4CzgReAe4NN78VeGP4e1SH+xwNt/0D8CF3rwQuBh6dyvuKnIwC\nQ/LN37n7IXffD/wS+LW7b3b3fuBHQFO437uBn7j7w+4+BHwJKAXeAFwORIGvuPuQu98HPJ3yHuuA\nb7r7r9096e7/BAyEx03F+4H17v6Muw8AnwauMLOlwBBQCSwHzN1fcPeD4XFDwAozq3L3dnd/Zorv\nKzIpBYbkm0Mpy32TPK8IlxcT/EUPgLuPAM3AknDbfh8/cucrKctnAZ8IT0d1mFkHEA+Pm4qJNXQT\ntCKWuPujwN8DdwCHzewuM6sKd30ncD3wipn93MyumOL7ikxKgSEyuQMEX/xA0GdA8KW/HzgILAnX\njWpMWW4G/trdYyk/Ze5+z2nWUE5wims/gLt/zd1fB6wgODV1W7j+aXe/EVhIcOrs+1N8X5FJKTBE\nJvd94AYzu8bMosAnCE4rPQH8ChgGPmpmUTP7XWBNyrHfAm41s9eHndPlZnaDmVVOsYZ7gA+YWSLs\n//hfBKfQ9prZZeHrR4EeoB8YCftY3m9m1eGptC5g5DQ+B5HjFBgik3D3ncDNwN8BRwg6yN/u7oPu\nPgj8LvCHQBtBf8cPU47dCPwJwSmjdmBPuO9Ua3gE+CzwA4JWzTnAe8LNVQTB1E5w2uoo8DfhtluA\nvWbWBdxK0BcictpMEyiJiEg61MIQEZG0KDBERCQtCgwREUmLAkNERNJSmO0CZlJdXZ0vXbo022WI\niOSMTZs2HXH3+nT2nVeBsXTpUjZu3JjtMkREcoaZvfLaewV0SkpERNKiwBARkbQoMEREJC0Z7cMw\ns+uArwIR4Nvu/sUJ298PfAow4BjwX9z92XDb3nBdEhh299WZrFVE8tPQ0BAtLS309/dnu5SMKikp\noaGhgWg0Ou3XyFhgmFmEYOjla4EW4Gkzu9/dt6fs9jLwm+7ebmZrgbuA16dsv9rdj2SqRhGRlpYW\nKisrWbp0KeMHIJ4/3J2jR4/S0tLCsmXLpv06mTwltQbY4+4vhYO13QvcmLqDuz/h7u3h0yeBhgzW\nIyJygv7+fmpra+dtWACYGbW1tafdispkYCwhmBdgVEu47mT+mGC6zFEOPGJmm8xs3ckOMrN1ZrbR\nzDa2traeVsEikp/mc1iMmonfcU50epvZ1QSB8amU1Ve5ewJYC3zYzN442bHufpe7r3b31fX1ad17\nMs7g8AjfePxFfrlbYSMiciqZDIz9BDOUjWoI141jZiuBbwM3uvvoJPaEcy7j7ocJ5lpeM/HYmRCN\nGN/65Uv8+NkDmXh5EZFT6ujo4Otf//qUj7v++uvp6OjIQEUnl8nAeBo4z8yWmVkRwcQv96fuYGaN\nBBPP3OLuu1LWl4/OThZOS/lWYFsmijQzEvEYm/fN7gcvIgInD4zh4eFTHrdhwwZisVimyppUxq6S\ncvdhM/sI8CDBZbXr3f15M7s13H4n8DmCOYq/Hp5fG718dhHwo3BdIXC3u/80U7Um4jEe23mYrv4h\nqkqmf8mZiMhU3X777bz44oskEgmi0SglJSXU1NSwY8cOdu3axW//9m/T3NxMf38/H/vYx1i3LujS\nHR0Kqbu7m7Vr13LVVVfxxBNPsGTJEv793/+d0tLSGa81o/dhuPsGYMOEdXemLH8Q+OAkx70ErMpk\nbamaGmO4w3PNnVx1Xt1sva2IzDGf//HzbD/QNaOvuWJxFX/59otOuv2LX/wi27ZtY8uWLTz++OPc\ncMMNbNu27fjlr+vXr2fBggX09fVx2WWX8c53vpPa2tpxr7F7927uuecevvWtb3HTTTfxgx/8gJtv\nvnlGfw+YI53e2bayIWjWbWluf409RUQya82aNePulfja177GqlWruPzyy2lubmb37t0nHLNs2TIS\niQQAr3vd69i7d29GaptXo9VOV3VplHMXVqgfQyTPnaolMFvKy8uPLz/++OM88sgj/OpXv6KsrIw3\nvelNk95LUVxcfHw5EonQ19eXkdrUwggl4jG2NHfg7tkuRUTySGVlJceOHZt0W2dnJzU1NZSVlbFj\nxw6efPLJWa5uPLUwQk2NMe7b1EJzWx+NtWXZLkdE8kRtbS1XXnklF198MaWlpSxatOj4tuuuu447\n77yTCy+8kAsuuIDLL788i5UqMI5LxIN+jM3N7QoMEZlVd99996Tri4uLeeCBBybdNtpPUVdXx7Zt\nY3cdfPKTn5zx+kbplFTogkWVlEYj6scQETkJBUaoMFLAJQ3VbGlWYIiITEaBkaKpMcb2A10MDCez\nXYqIyJyjwEjRFI8xmByZ8Rt3RETmAwVGiqbGGgD1Y4iITEKBkWJRVQlnVpeoH0NEZBIKjAmaGmNs\n1hAhIjJLpju8OcBXvvIVent7Z7iik1NgTJCIx2hu6+NI90C2SxGRPJBLgaEb9yYY7cfYsq+Dt6xY\n9Bp7i4icntThza+99loWLlzI97//fQYGBvid3/kdPv/5z9PT08NNN91ES0sLyWSSz372sxw6dIgD\nBw5w9dVXU1dXx2OPPZbxWhUYE1y8uJpIgbGlWYEhknceuB1e3Tqzr3nGJbD2iyfdnDq8+UMPPcR9\n993HU089hbvzjne8g1/84he0trayePFifvKTnwDBGFPV1dV8+ctf5rHHHqOubnamZdApqQlKiyIs\nP6NS/RgiMuseeughHnroIZqamrj00kvZsWMHu3fv5pJLLuHhhx/mU5/6FL/85S+prq7OSn1qYUyi\nqTHGv20+QHLEiRRYtssRkdlyipbAbHB3Pv3pT/OhD33ohG3PPPMMGzZs4DOf+QzXXHMNn/vc52a9\nPrUwJpGI19A9MMyLrd3ZLkVE5rnU4c3f9ra3sX79erq7g++e/fv3c/jwYQ4cOEBZWRk333wzt912\nG88888wJx84GtTAm0dQYzsC3r4PzF1VmuRoRmc9Shzdfu3Yt73vf+7jiiisAqKio4Lvf/S579uzh\ntttuo6CggGg0yje+8Q0A1q1bx3XXXcfixYtnpdPb5tOEQatXr/aNGzee9uuMjDiJv3qIG1aeyRd+\nd+UMVCYic9ULL7zAhRdemO0yZsVkv6uZbXL31ekcr1NSkygoMBKNNRoiREQkhQLjJBLxGLsOHaNn\nYDjbpYiIzAkKjJNoaowx4vBcS2e2SxGRDJtPp+ZPZiZ+RwXGSSQawo5vDUQoMq+VlJRw9OjReR0a\n7s7Ro0cpKSk5rdfRVVInUVNexLK6cjbv0w18IvNZQ0MDLS0ttLa2ZruUjCopKaGhoeG0XkOBcQqJ\neIz/3HMEd8dMN/CJzEfRaJRly5Zlu4ycoFNSp9DUGKP12AAHOvuzXYqISNYpME4hER+7gU9EJN8p\nME5h+RlVFBcWqB9DRAQFxikVFRZw8ZJqXSklIoIC4zU1xWNs3d/JUHIk26WIiGSVAuM1JBpjDAyP\nsOPg7I0IKSIyF2U0MMzsOjPbaWZ7zOz2Sba/38yeM7OtZvaEma1K99jZMjplqyZUEpF8l7HAMLMI\ncAewFlgBvNfMVkzY7WXgN939EuB/AHdN4dhZsbi6hPrKYl0pJSJ5L5MtjDXAHnd/yd0HgXuBG1N3\ncPcn3H30T/cngYZ0j50tZkZTPMZmdXyLSJ7LZGAsAZpTnreE607mj4EHpnqsma0zs41mtjFTt/Yn\nGmO8fKSH9p7BjLy+iEgumBOd3mZ2NUFgfGqqx7r7Xe6+2t1X19fXz3xxQFM86MfY0qJWhojkr0wG\nxn4gnvK8IVw3jpmtBL4N3OjuR6dy7GxZ2VBNgemObxHJb5kMjKeB88xsmZkVAe8B7k/dwcwagR8C\nt7j7rqkcO5vKiws5f1Gl+jFEJK9lbLRadx82s48ADwIRYL27P29mt4bb7wQ+B9QCXw9Hgx0OTy9N\nemymak1HU2OMDVtfZWTEKSjQyLUikn8yOry5u28ANkxYd2fK8geBD6Z7bDYl4jHueaqZl4/2cE59\nRbbLERGZdXOi0zsXjN7Ap34MEclXCow0nVNfQUVxoe74FpG8pcBIU6TAWBXXyLUikr8UGFOQiMd4\n4eAx+gaT2S5FRGTWKTCmoCleQ3LE2XagM9uliIjMOgXGFCQaNWWriOQvBcYU1FUUE19Qqo5vEclL\nCowpSsRr1MIQkbykwJiipniMA539HOrqz3YpIiKzSoExRaP9GJvVyhCRPKPAmKKLFldRFClQP4aI\n5B0FxhQVF0a4cHGV+jFEJO8oMKahKR7juZZOhpMj2S5FRGTWKDCmoakxRt9Qkl2HurNdiojIrFFg\nTMPolK3qxxCRfKLAmIb4glIWlBepH0NE8ooCYxrMjKZ4TFO2ikheUWBMUyIeY8/hbjr7hrJdiojI\nrFBgTNPoDHzPtaiVISL5QYExTSvj1Zhp5FoRyR8KjGmqKolybn2F+jFEJG8oME5DIh5jS3MH7p7t\nUkREMk6BcRoSjTHaegbZ19ab7VJERDJOgXEaRm/g26LTUiKSBxQYp+H8RRWURiMa6lxE8oIC4zQU\nRgpY2VCtjm8RyQsKjNOUaIyx/UAn/UPJbJciIpJRCozT1BSvYSjpbD/Yle1SREQySoFxmpo0ZauI\n5AkFxmlaVFXC4uoSXSklIvOeAmMGJBpjbNHcGCIyzykwZkBTvIbmtj6OdA9kuxQRkYzJaGCY2XVm\nttPM9pjZ7ZNsX25mvzKzATP75IRte81sq5ltMbONmazzdCXCfgwNRCgi81nGAsPMIsAdwFpgBfBe\nM1sxYbc24KPAl07yMle7e8LdV2eqzplw8eJqCgtMU7aKyLyWyRbGGmCPu7/k7oPAvcCNqTu4+2F3\nfxrI6VmISosiLD+zUh3fIjKvZTIwlgDNKc9bwnXpcuARM9tkZutOtpOZrTOzjWa2sbW1dZqlnr6m\neA3PNneSHNHItSIyP83lTu+r3D1BcErrw2b2xsl2cve73H21u6+ur6+f3QpTJOIxugeGebG1O2s1\niIhkUiYDYz8QT3neEK5Li7vvDx8PAz8iOMU1Z43dwKd+DBGZnzIZGE8D55nZMjMrAt4D3J/OgWZW\nbmaVo8vAW4FtGat0BiyrK6e6NKp+DBGZtwoz9cLuPmxmHwEeBCLAend/3sxuDbffaWZnABuBKmDE\nzD5OcEVVHfAjMxut8W53/2mmap0JZkYiHtMQISIyb2UsMADcfQOwYcK6O1OWXyU4VTVRF7Aqk7Vl\nQiIe4+8e3U33wDAVxRn9aEVEZt1c7vTOOU2NMUYcnmtRK0NE5h8FxgxKxMM7vtWPISLzkAJjBsXK\niji7rlz9GCIyLykwZlgiHmNLcwfuuoFPROYXBcYMSzTGaD02wP6OvmyXIiIyoxQYM6wpXgOoH0NE\n5h8FxgxbfmYlxYUF6scQkXknrcAws4+ZWZUF/sHMnjGzt2a6uFwUjRRwyZJqtTBEZN5Jt4XxR+7e\nRTBERw1wC/DFjFWV4xLxGFv3dzI4PJLtUkREZky6gWHh4/XAP7v78ynrZIKmxhoGh0fY8WpXtksR\nEZkx6QbGJjN7iCAwHgwHBtSfzyeROD5yrU5Licj8kW5g/DFwO3CZu/cCUeADGasqxy2uLmFhZbH6\nMURkXkk3MK4Adrp7h5ndDHwG6MxcWbltbORazY0hIvNHuoHxDaDXzFYBnwBeBL6TsarmgabGGvYe\n7aW9ZzDbpYiIzIh0A2PYg7EubgT+3t3vACozV1buOz4QoUauFZF5It3AOGZmnya4nPYnZlZA0I8h\nJ7GyoZoCU8e3iMwf6QbGu4EBgvsxRic9+puMVTUPlBcXcv6iSnV8i8i8kVZghCHxPaDazH4L6Hd3\n9WG8hqbGGrbsa2dkRCPXikjuS3dokJuAp4DfA24Cfm1m78pkYfNBUzxGV/8wLx/tyXYpIiKnLd2J\np/+C4B6MwwBmVg88AtyXqcLmg6aUG/jOqa/IcjUiIqcn3T6MgtGwCB2dwrF565z6CiqLC9nSrPsx\nRCT3pdvC+KmZPQjcEz5/N7AhMyXNHwUFxqp4TFdKici8kG6n923AXcDK8Ocud/9UJgubLxLxGDte\nPUbfYDLbpYiInJZ0Wxi4+w+AH2SwlnmpqTFGcsTZur+TNcsWZLscEZFpO2ULw8yOmVnXJD/HzExj\nd6fh+B3f6scQkRx3yhaGu2v4j9NUW1FM44Iy9WOISM7TlU6zIBGP6Y5vEcl5CoxZkIjHONjZz6ud\n/dkuRURk2hQYs2D0Bj71Y4hILlNgzIIVi6soihSoH0NEcpoCYxYUF0ZYsbiKzerHEJEcpsCYJYl4\njK0tnQwnR7JdiojItGQ0MMzsOjPbaWZ7zOz2SbYvN7NfmdmAmX1yKsfmmqbGGH1DSXYeOpbtUkRE\npiVjgWFmEeAOYC2wAnivma2YsFsb8FHgS9M4Nqc0xWsAzcAnIrkrky2MNcAed3/J3QeBewnmBD/O\n3Q+7+9PA0FSPzTXxBaXUlhfpfgwRyVmZDIwlQHPK85Zw3Ywea2brzGyjmW1sbW2dVqGzwcxIxGNs\n3qdLa0UkN+V8p7e73+Xuq919dX19fbbLOaWmxhgvtvbQ2TexQSUiMvdlMjD2A/GU5w3hukwfO2cl\nwn6MZ3VaSkRyUCYD42ngPDNbZmZFwHuA+2fh2DlrZbwaM9SPISI5Ke35MKbK3YfN7CPAg0AEWO/u\nz5vZreH2O83sDGAjUAWMmNnHgRXu3jXZsZmqdbZUlUQ5t75CgSEiOSljgQHg7huYMJWru9+Zsvwq\nwemmtI6dD5oaYzy8/RDujplluxwRkbTlfKd3rknEa2jvHWJfW2+2SxERmRIFxiwbHblWN/CJSK5R\nYMyy8xdVUlYUUT+GiOQcBcYsixQYKxuqdQOfiOQcBUYWJOI1bD/YRf9QMtuliIikTYGRBU2NMYaS\nzvMHurJdiohI2hQYWdAUH52yVf0YIpI7FBhZsLCqhCWxUvVjiEhOUWBkSSIeUwtDRHKKAiNLEvEY\nLe19tB4byHYpIiJpUWBkyegNfGpliEiuUGBkycVLqiksMPVjiEjOUGBkSUk0woVnVqmFISI5Q4GR\nRYl4jGebO0iOeLZLERF5TQqMLGpqjNEzmGTP4e5slyIi8poUGFmUiI+OXKt+DBGZ+xQYWbSsrpzq\n0qj6MUQkJygwssjMSMRjmhtDRHKCAiPLmhpj7Dp8jO6B4WyXIiJySgqMLEvEY7jDczotJSJznAIj\ny453fCswRGSOU2BkWaysiLPrytXxLSJzngJjDkg0Bh3f7rqBT0TmLgXGHNAUj3Gke4D9HX3ZLkVE\n5KQUGHNAU2MNgC6vFZE5TYExB1xwRiXFhQXqxxCROU2BMQdEIwWsbKjWECEiMqcpMOaIRDzGtgNd\nDA6PZLsUEZFJKTDmiKbGGgaHR3RaSkTmLAXGHHHZ0gVUFBfyB+uf4ms/203/UDLbJYmIjKPAmCPq\nK4t54GO/wZuXL+TLD+/imv/zc/7juQO6N0NE5oyMBoaZXWdmO81sj5ndPsl2M7OvhdufM7NLU7bt\nNbOtZrbFzDZmsk6e/AbsexKy/OUcX1DGHe+/lHvXXU5VaZSP3L2Zd3/zSbbt78xqXSIiAJapv2DN\nLALsAq4FWoCngfe6+/aUfa4H/hS4Hng98FV3f324bS+w2t2PpPueq1ev9o0bp5gtA93w5RUw0AkL\nzoHE+2DVe6F6ydReZ4YlR5zvb2zmSw/upK13kHevjvOJt15AfWVxVusSkfnFzDa5++p09s1kC2MN\nsMfdX3L3QeBe4MYJ+9wIfMcDTwIxMzszgzWdqLgC/nw73Ph1qDwDHv0f8JWL4Z9/F7b9AIb6Z7Wc\nUZEC471rGnnstjfxwauWcd+mFq7+0uPc9YsXdSWViGRFJgNjCdCc8rwlXJfuPg48YmabzGxdxqqE\nIDSa3g8f2AAf3Qy/8Uk4sgvu+yP4P+fDf/w57N+UlVNWVSVR/uKGFTz4Z29kzbIF/K8NO3jbV37B\nI9sPqX9DRGbVXO70vsrdE8Ba4MNm9sbJdjKzdWa20cw2tra2nv67Ljgb3vwX8LHn4JZ/g/PeClu+\nB996M3z9Cnji76D78Om/zxSdU1/B+j+8jH/8wGUUGHzwOxv5/fVPsfvQsVmvRUTyUyb7MK4A/ru7\nvy18/mkAd/9Cyj7fBB5393vC5zuBN7n7wQmv9d+Bbnf/0qnec1p9GOno74RtPwyCo+VpsAic/7ag\nv+O8t0Fh0cy/5ykMJUf47pOv8LcP76JnMMktl5/Fx99yHrGy2a1DRHLfVPowMhkYhQSd3tcA+wk6\nvd/n7s+n7HMD8BHGOr2/5u5rzKwcKHD3Y+Hyw8BfuftPT/WeGQuMVK07Ycvd8Oy90P0qlNXCyndD\n4v1wxsWZfe8J2noG+fLDO7n71/uoKo3y59eez/vWNFIYmcsNRxGZS+ZEYISFXA98BYgA6939r83s\nVgB3v9PMDPh74DqgF/iAu280s7OBH4UvUwjc7e5//VrvNyuBMSo5DC8+Clu+CzsfgOQgnLkqCI5L\nfg/KFsxOHcCOV7v4qx9v54kXj3L+ogo+91sXcdV5dbP2/iKSu+ZMYMy2WQ2MVL1tsPW+IDwOPguR\nIrhgLSRuhnPeDJHCjJfg7jy0/RB//ZMX2NfWy7UrFvEX11/I0rryjL+3iOQuBUY2vbot6Ot47l+g\n9yhUnAGr3h2ER/35GX/7/qEk6//fy9zx6B6Gks4HrlrKR64+l8qSaMbfW0RyjwJjLhgehN0PBeGx\n60HwJDRcFnSUX/xOKKnO6Nsf7urnfz+4k/s2tVBXUcx/fdsFvOt1DRQUWEbfV0RyiwJjruk+HLQ4\nNn8PWl+AwhK48O1Bf8ey34SCzHVSP9vcwed//DzP7OvgkiXV/OXbV7B66ez1r4jI3KbAmKvc4cDm\noNWx9V+Dy3WrGiDx3qDlseDsDL2tc/+zB/jiAzs42NnP21ct5va1y1kSK83I+4lI7lBg5IKhftj5\nk+AS3RcfBR+BxjfA4iaIxSHWCNXhY2lsRt6yd3CYO3/+Et/8+YuYwYfeeA63/uY5lBZFZuT1RST3\nKDByTdcBePYeeP5HcPRFGOodv724anyAxBpTQqUxuITX0u+baGnv5QsP7OAnzx1kcXUJt19/IW9f\neSY2hdcQkflBgZHL3IPLdDtegc5m6NgHHaOP4c/ghOFAouUntkpicYidFSyX108aKE+93Mbnf/w8\nzx/oYvVZNfzl2y/ikobMdsaLyNyiwJjP3KG/YyxEOieESWcz9LWPP6awJAyS1FA5C2JxklVx7ts1\nxN88tJujPYO869IGbrvuAhZWlmTn9xORWaXAyHf9XWGQjIbKvvEtld4JU4xEihipXMIrI7Vs7Khg\nP4uILF5F4yVv4MpVK6ir0BwcIvOVAkNObbAHOltObJl07GO4bR+FvYeO77rfa9lXfAGcmWDxRW+g\n8aI3YOW1WSxeRGaSAkNOz8Ax/OBzvLrjSTpffIrKtq0sSe4/vrmt6EyGF62i5tzXE41fCmcmZuxK\nLhGZXVMJjMwPciS5p7gSW3olZy69ktHpD1uPHGbb07/g6O5fU35kKyteeYaFzWODBw/HllHYcGlw\nWfDipmAgxuLKrJQ/7wwPBi3AkWGIloU/pUHfVAZv+hSZSC0MmbL+oSRPvnSUX23bw6EdT7K4dweX\nFLzE66J7WTgSTGLlGFZ33liALG6CMy6BIg2GeIKRJBw7CO2vBFfHTXzsOkAwAeUkCkuD8BgNkaKU\nQDn+WJoSNBPXhY9FE49JeSzQfTrzmU5Jyaxxd3YeOsbPXjjMz144xL7mV7jYXuby0mZ+o7yFc4Z2\nUdIX9olYAdQvHx8iiy4KvpjmM3foORL2F+09MRA6mmFkKOUAg6rFwZVsNeGl0bGzIFoCg70w1Bfc\nqzPusQ+GeiZZFy4P9gbL494nTZGi8WFTXgcVi6DyjJTHM6ByUfBYVquWTw5RYEjWHOke4PGdrfzs\nhUP8YlcrPYNJGgo7+b3FR7m6qoXzk3soOfws9ITT6RYUwsILx4fIwhVQmGNXZvV3Td46aH8lCIqh\nnvH7l9WmBELq41Kobsjc758cOjFMhnonCaCU0BlMCaLBnuAqu2OHggnE+jtPfI+CQihfOBYgJ3us\nWAgRjaI8ZQPH4MguaN0FrTuC5eQQ3HzftF5OgSFzwsBwkqdebgtaHzsO0dzWB8BFZ1Zy49lwbewg\nSwd2YQc3B2Ns9bUFBxZEg5ZH/fLgr+pIUfgTnfA4g8uvddplqD/oR2h/ZfJWwsR7X4oqJwmDs8Zu\npiyumPkPPBuG+qD70FiAHDsUnF7rPgTHXh17nHgpNwAWBOe4lkrq45lj4RLNw/uCeo7CkZ1BKLTu\nCpd3QtfYBSgURKH23OD/yzu/PaURH0YpMGTOcXf2HO7mkRcO8+iOQ2x6pZ0Rh/rKYt58wULevLye\nNy7so7T1uSA8DmyGtpeDmQyTg8FfUMlBSA5kpkArOHkwDXQHX4apIsXBF/+4MGgcayWU1kzrP++8\nlRwKRm1ajVtDAAALUUlEQVQeDZWTPh4KpgKYqKR6klZKSkulYlGwriSWW5+7e9BHNRoGrTvD1sOO\nYD6dUdEyqDsf6i8IH5cHyzVLT7uVpsCQOa+tZ5Cf7zrMIy8c5hc7Wzk2MExRYQFXnF3LWy5cyNXL\nF9JQU3bige5BJ/FkQXJ8efAky6+1/STLhaUnthIqFuk8fSaMjARflOOC5NUTWyzdh2C4/8TjI0Vh\nkCw8MUwqFqVsWzS7pz1HktC+NwyEnSkthl3jh/opiYVhcD7UXTC2XNWQsX9vCgzJKUPJEZ7e23a8\n43zv0WDwxbPryllaV068ppT4grLgp6aM+IJSzSCY79xhoGusVdJ9aKwF03045TTZoZOcDiP4cj4h\nTCaESuUZU2stDg8EA4iO9i2MthqO7hnfOq44I2ghTGwxnGTct0xSYEhOe7G1m0dfOMzTe9tobu+j\npa2XYwPD4/apKYumBEgQIqPLS2KlFBXqr38JJYeCiyxGQ+VYSqik/hw7BMN9Jx5fEB0LkHFhsii4\nF+bonrEO6Pa9KafULGiR1l0QtBLqlwfLdefNqRtdFRgyr7g7nX1D7Gvrpbmtj+b23nC5l5b2Plra\nexlKjv07LjA4o6qEhgVlNKa0ShrDVkp9RbGmqpUTuQdXIB0Pk9RgmRA0Pa0cvzemoDDoeB7tY6hf\nHizXnhvc3zLH6U5vmVfMjFhZEbGyIlY2nPiXWXLEOdTVT3NbGCRhq2RfWy+/3N3Koa7xHeVFhQVj\np7lqwlBZUEpDTRmNtWVU6XRXfjKDkqrgp+7cU++bHA5OdQ32BBc75MnlwQoMyXmRAmNxrJTFsVJe\nf/aJAyP2DyVpaQ9aJqNBMtpS2fRKO8f6x5/uqi6NjrVIaso4o7qEWFk0CK3S4LGmLEplSZSIWir5\nKVIY9G/kGQWGzHsl0QjnLqzg3IWT3/vQ2Ts07jRXsNzHjoPHeGT7YQaTI5MeZwZVJdEJYRKlpqyI\n6tTlsiix0mA5pqCRHKbAkLxXXRaluqyai5ecONvgyIjT0TdER+8g7b1DdPYN0tE7FP4MhtuGaO8d\npL13kJeP9NDRO0jXhFZLKrOgFRMrjVIdtlZGWy7VpdHgeRg0NSlBVFUSVd+LZJUCQ+QUCgqMBeVF\nLCgvmtJxw8kRuvqHTwia9t4hOicETVvPIC+19tDeO3jC6bFU41o0YdiMhknq89HWTawsSnVpEEK6\nakxmggJDJAMKIwUZCZrOvqHjYdPRN8S+oz109A3R2TfEqS54LC+KHG/BBEEyFiix4y2cseej20uj\nESyX7pyWjFJgiMwh0w2akRHnWP8wHX1hqISB0tk7eHy5IyWA9hzuPn6qLfWS5ImKIgXH+2BGA2VB\neZTaimJqy4uorShiQXnqchHFhRoO/XQMJUfoH0rSPxQ8DgyPLR9/HLcuSVFhAb9/xdKM16bAEJkH\nCgos7IuZ2uWd7k7fUHKsX6ZvkM5xATO+36alvZet+wc52j3I8MjkQVNZXHg8PCYGS93o+vJiaiuK\nqCkrmvOny9ydgeER+gaT9A0l6R1M0jeYpHdwmL7wCzv1y7t/ePSLfuxLfmCSL/n+oRH6h5MMTDg2\neZLP9VRqy4sUGCKSWWZGWVEhZUWFLI6lPy+Ju9PVP8zR7gHaegY50h30xRztHuBozyBHewZp6xmg\nua2XLc0dtPUMnvSLsKqkkLqK4jBgJgRLxfjWy4KyIgojJwZMcsTDL/Ph8Mt87It94vrR52Nf/MFP\nf+r6CcEwje9wohGjpDBCcTRCSbSAktHHwghlRYUsKC8IthWeuL0kGqE4XC4+vi1CSWHK8uj6cJ/Z\noMAQkSkzM6rDDvaz6197/5ERp6t/6MRg6Q6C5UjPIG3dwVVmm15pp61n8KRf0qN9LkNJP/4FPzA8\n+aXPJxMpMMqiEUqKIpQVRSiNBo9lRYXUVhSHy8EX8+j60X1Kw+ej20tO8oU+Hy+dVmCISMYVFIzd\nrZ+O5EgwHMwJwRIGTkffEEWRguNf7KXHHwspC7/YS4oi4XLh8e2j+xZFCtSZPw0ZDQwzuw74KhAB\nvu3uX5yw3cLt1wO9wB+6+zPpHCsi81ck5XLm87JdjByXsRNfZhYB7gDWAiuA95rZigm7rQXOC3/W\nAd+YwrEiIjKLMtlTsgbY4+4vufsgcC9w44R9bgS+44EngZiZnZnmsSIiMosyGRhLgOaU5y3hunT2\nSedYAMxsnZltNLONra2tp120iIhMbm5fAJ0Gd7/L3Ve7++r6+jQu1xARkWnJZKf3fiCe8rwhXJfO\nPtE0jhURkVmUyRbG08B5ZrbMzIqA9wD3T9jnfuD3LXA50OnuB9M8VkREZlHGWhjuPmxmHwEeJLg0\ndr27P29mt4bb7wQ2EFxSu4fgstoPnOrYTNUqIiKvTXN6i4jksanM6T2vAsPMWoFXpnl4HXBkBsvJ\nZfosxtPnMZ4+jzHz4bM4y93TumJoXgXG6TCzjemm7Hynz2I8fR7j6fMYk2+fRc5fVisiIrNDgSEi\nImlRYIy5K9sFzCH6LMbT5zGePo8xefVZqA9DRETSohaGiIikRYEhIiJpyfvAMLPrzGynme0xs9uz\nXU82mVnczB4zs+1m9ryZfSzbNWWbmUXMbLOZ/Ue2a8k2M4uZ2X1mtsPMXjCzK7JdUzaZ2Z+F/0+2\nmdk9ZlaS7ZoyLa8DQxM1nWAY+IS7rwAuBz6c558HwMeAF7JdxBzxVeCn7r4cWEUefy5mtgT4KLDa\n3S8mGMLoPdmtKvPyOjDQRE3juPvB0Sly3f0YwRfCpPOQ5AMzawBuAL6d7VqyzcyqgTcC/wDg7oPu\n3pHdqrKuECg1s0KgDDiQ5XoyLt8DI+2JmvKNmS0FmoBfZ7eSrPoK8F+BkWwXMgcsA1qB/xueovu2\nmZVnu6hscff9wJeAfcBBgpG2H8puVZmX74EhkzCzCuAHwMfdvSvb9WSDmf0WcNjdN2W7ljmiELgU\n+Ia7NwE9QN72+ZlZDcHZiGXAYqDczG7OblWZl++Bkc4kT3nFzKIEYfE9d/9htuvJoiuBd5jZXoJT\nlW82s+9mt6SsagFa3H20xXkfQYDkq7cAL7t7q7sPAT8E3pDlmjIu3wNDEzWlMDMjOEf9grt/Odv1\nZJO7f9rdG9x9KcG/i0fdfd7/BXky7v4q0GxmF4SrrgG2Z7GkbNsHXG5mZeH/m2vIg4sAMjlF65yn\niZpOcCVwC7DVzLaE6/6bu2/IYk0yd/wp8L3wj6uXCCc8y0fu/mszuw94huDqws3kwTAhGhpERETS\nku+npEREJE0KDBERSYsCQ0RE0qLAEBGRtCgwREQkLQoMkTnAzN6kEXFlrlNgiIhIWhQYIlNgZjeb\n2VNmtsXMvhnOl9FtZn8bzo3wMzOrD/dNmNmTZvacmf0oHH8IMzvXzB4xs2fN7BkzOyd8+YqU+Sa+\nF95BLDJnKDBE0mRmFwLvBq509wSQBN4PlAMb3f0i4OfAX4aHfAf4lLuvBLamrP8ecIe7ryIYf+hg\nuL4J+DjB3CxnE9x5LzJn5PXQICJTdA3wOuDp8I//UuAwwfDn/xLu813gh+H8ETF3/3m4/p+AfzWz\nSmCJu/8IwN37AcLXe8rdW8LnW4ClwH9m/tcSSY8CQyR9BvyTu3963Eqzz07Yb7rj7QykLCfR/0+Z\nY3RKSiR9PwPeZWYLAcxsgZmdRfD/6F3hPu8D/tPdO4F2M/uNcP0twM/DmQxbzOy3w9coNrOyWf0t\nRKZJf8GIpMndt5vZZ4CHzKwAGAI+TDCZ0Jpw22GCfg6APwDuDAMhdXTXW4Bvmtlfha/xe7P4a4hM\nm0arFTlNZtbt7hXZrkMk03RKSkRE0qIWhoiIpEUtDBERSYsCQ0RE0qLAEBGRtCgwREQkLQoMERFJ\ny/8H2MWOMEWaEukAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x307247b8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# summarize history for accuracy\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='lower right')\n",
    "plt.show()\n",
    "# summarize history for loss\n",
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.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": "markdown",
   "metadata": {},
   "source": [
    "#### Test dataset evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Processing text to be compatible with model input\n",
    "\n",
    "test_sequences = tokenizer.texts_to_sequences(test_texts)\n",
    "test_input = pad_sequences(test_sequences, maxlen=MAX_SEQUENCE_LENGTH)\n",
    "\n",
    "# Get predictions\n",
    "test_predictions_probas = model.predict(test_input)\n",
    "test_predictions = test_predictions_probas.argmax(axis=-1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy:  0.9866666666666667\n",
      "Precision, Recall and F1-Score:\n",
      "\n",
      "                       precision    recall  f1-score   support\n",
      "\n",
      "       AddToPlaylist       1.00      1.00      1.00       100\n",
      "      BookRestaurant       1.00      1.00      1.00       100\n",
      "          GetWeather       1.00      0.98      0.99       100\n",
      "            RateBook       1.00      1.00      1.00       100\n",
      "  SearchCreativeWork       0.95      0.99      0.97       100\n",
      "SearchScreeningEvent       0.97      0.95      0.96       100\n",
      "\n",
      "         avg / total       0.99      0.99      0.99       600\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\ajinkya\\Documents\\chatbot\\chatbot_env\\lib\\site-packages\\sklearn\\preprocessing\\label.py:171: DeprecationWarning: The truth value of an empty array is ambiguous. Returning False, but in future this will result in an error. Use `array.size > 0` to check that an array is not empty.\n",
      "  if diff:\n",
      "C:\\Users\\ajinkya\\Documents\\chatbot\\chatbot_env\\lib\\site-packages\\sklearn\\preprocessing\\label.py:171: DeprecationWarning: The truth value of an empty array is ambiguous. Returning False, but in future this will result in an error. Use `array.size > 0` to check that an array is not empty.\n",
      "  if diff:\n"
    ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report\n",
    "\n",
    "test_intent_predictions = train_le.inverse_transform(test_predictions)\n",
    "test_intent_original = train_le.inverse_transform(test_labels)\n",
    "\n",
    "print('accuracy: ', sum(test_intent_predictions == test_intent_original) / len(test_labels))\n",
    "print(\"Precision, Recall and F1-Score:\\n\\n\", classification_report(test_intent_original, test_intent_predictions))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So we saw how we can reach 99% accuracy using Convolutional Neural Network for our intent classification task within a minute!"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}