text_classification_LSTM_usingKeras.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "import sys, os, re, csv, codecs, numpy as np, pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.layers import Dense, Input, LSTM, Embedding, Dropout, Activation, Conv1D\n",
    "from keras.layers import Bidirectional, GlobalMaxPooling1D, MaxPooling1D, SpatialDropout1D, BatchNormalization\n",
    "from keras.models import Model\n",
    "from keras import initializers, regularizers, constraints, optimizers, layers\n",
    "from tqdm import tqdm\n",
    "import gc"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Import Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dtypes = {\n",
    "    'comment_text': np.unicode,\n",
    "    \"toxic\": np.float32,\n",
    "    \"severe_toxic\": np.float32,\n",
    "    \"obscene\": np.float32,\n",
    "    \"threat\": np.float32,\n",
    "    \"insult\": np.float32,\n",
    "    \"identity_hate\": np.float32\n",
    "}\n",
    "train = pd.read_csv(\"data/train.csv\", dtype=dtypes) #, encoding='utf-8'\n",
    "test = pd.read_csv(\"data/test.csv\", dtype=dtypes) # , encoding='utf-8'\n",
    "\n",
    "train.comment_text.fillna(\"unknown\", inplace=True)\n",
    "test.comment_text.fillna(\"unknown\", inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "classes = [\"toxic\", \"severe_toxic\", \"obscene\", \"threat\", \"insult\",\n",
    "           \"identity_hate\"]\n",
    "train_data = train['comment_text']  \n",
    "labels = train[classes].values\n",
    "list_sentences_train = train[\"comment_text\"]\n",
    "list_sentences_test = test[\"comment_text\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    " ## This is used to represent the corpus in the format of a  dictionary where every index represent a unique word and it is arranged from the most frequent to the least frequent"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "str"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(list(list_sentences_train)[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "max_features = 20000 # Number of unique words in the corpus, needs optimization\n",
    "tokenizer = Tokenizer(num_words=max_features)\n",
    "tokenizer.fit_on_texts(list(list_sentences_train))\n",
    "tokenized_train = tokenizer.texts_to_sequences(list_sentences_train)\n",
    "tokenized_test = tokenizer.texts_to_sequences(list_sentences_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 153397 unique tokens.\n"
     ]
    }
   ],
   "source": [
    "word_index = tokenizer.word_index\n",
    "print('Found %d unique tokens.' % len(word_index))\n",
    "word_count = tokenizer.word_counts\n",
    "# print('Top 5 Frequent Words count. ' % word_count[5])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[813, 4855, 200, 6773, 40, 7, 162, 8, 393, 234, 19, 20, 350, 14245]]\n"
     ]
    }
   ],
   "source": [
    "print(tokenized_train[:1]) # Take a lot at the NN input"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## This is done to normalize the length of the text sequences to the maxlen value\n",
    "\n",
    "    This is called [Padding], it includes shortenning the long sentences and longing the short sentences by zeros so they equal the maxlen"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "maxlen = 200 # needs optimization\n",
    "X = pad_sequences(tokenized_train, maxlen=maxlen)\n",
    "X_t =pad_sequences(tokenized_test, maxlen=maxlen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYcAAAD8CAYAAACcjGjIAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nktRhOEiSOgwHSVKH4SBJ6jAcJEkd/x+cJ9x5ROE/zgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f9eba9a8590>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot a histogram between the number of comments and the length of the comment\n",
    "totalNumWords = [len(comment) for comment in tokenized_train]\n",
    "plt.hist(totalNumWords) #, bins=np.range(0,500,10)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Let's Start building the Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"images/txJomEa.png\" >"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# input is a list of encoded sentences, each sentence with a maxlen=200\n",
    "input_layer = Input(shape=(maxlen, )) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2196017it [01:53, 19297.09it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 2196016 word vectors.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# Prepaing the Embedding Layer using Glove Word Vectors\n",
    "embedding_index_glove = {}\n",
    "f = open(\"../Sentiment_Model_Template/glove.840B.300d.txt\")\n",
    "for line in tqdm(f):\n",
    "    values = line.split()\n",
    "    word = values[0]\n",
    "    coefs = np.asarray(values[1:], dtype='float32')\n",
    "    embedding_index_glove[word] = coefs\n",
    "f.close()\n",
    "print('Found %s word vectors.' % len(embedding_index_glove))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def plot_loss_curve(history):\n",
    "    \"\"\"A function to plot losses and acuracies of a fitted\n",
    "    keras model\"\"\"\n",
    "    # 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', 'valid'], loc='upper left')\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', 'valid'], loc='upper left')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "95851"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_data.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# split the data into a training set and a validation set\n",
    "VALIDATION_SPLIT = 0.2\n",
    "\n",
    "indices = np.arange(X.shape[0])\n",
    "np.random.shuffle(indices)\n",
    "X = X[indices]\n",
    "labels = labels[indices]\n",
    "num_validation_samples = int(VALIDATION_SPLIT * X.shape[0])\n",
    "\n",
    "x_train = X[:-num_validation_samples]\n",
    "y_train = labels[:-num_validation_samples]\n",
    "x_val = X[-num_validation_samples:]\n",
    "y_val = labels[-num_validation_samples:]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 146,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(95851, 6)"
      ]
     },
     "execution_count": 146,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 147,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(76681, 200)\n",
      "(19170, 200)\n",
      "(76681, 6)\n",
      "(19170, 6)\n"
     ]
    }
   ],
   "source": [
    "print (x_train.shape)\n",
    "print (x_val.shape)\n",
    "print (y_train.shape)\n",
    "print (y_val.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 148,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# prepare embedding matrix\n",
    "EMBEDDING_DIM = 300\n",
    "num_words = min(max_features, len(word_index) + 1)\n",
    "embedding_matrix = np.zeros((num_words, EMBEDDING_DIM))\n",
    "for word, i in word_index.items():\n",
    "    if i >= max_features:\n",
    "        continue\n",
    "    embedding_vector = embedding_index_glove.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": "code",
   "execution_count": 149,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "MAX_SEQUENCE_LENGTH = maxlen\n",
    "embedding_layer = Embedding(num_words,\n",
    "                            EMBEDDING_DIM,\n",
    "                            weights=[embedding_matrix],\n",
    "                            input_length=MAX_SEQUENCE_LENGTH,\n",
    "                            trainable=False)(input_layer)\n",
    "\n",
    "# embedding_layer = Embedding(num_words,\n",
    "#                             EMBEDDING_DIM,\n",
    "#                             input_length=MAX_SEQUENCE_LENGTH,\n",
    "#                             trainable=False)(input_layer)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 150,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = LSTM(units=60, return_sequences=True, name='lstm_layer')(embedding_layer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 151,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = GlobalMaxPool1D()(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## This is done to achive Generalization, by doing some kind of regualization called [Droupout] \n",
    "    Which is basiclly done by droping the learned weights of a neuron in a NN to zero (dropout_ate=0.1) means that 10% of the neurons/nodes weights will be dropped to zero every iteration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = Dropout(rate=0.1)(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## A densely connected layer anf the output passes through a ReLU functin"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = Dense(units=50, activation=\"relu\")(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = Dropout(rate=0.1)(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Output layer has an activation [Sigmoid] to reduce the probabilities to a real number between [0, 1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x = Dense(units=6, activation=\"sigmoid\")(x) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 162,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model = Model(inputs=input_layer, outputs=x)\n",
    "model.compile(loss='binary_crossentropy',\n",
    "              optimizer= 'adam',\n",
    "              metrics= ['accuracy']) #{'toxic': 'accuracy', 'severe_toxic': 'accuracy', 'obscene': 'accuracy',\n",
    "                        #'threat': 'accuracy', 'insult': 'accuracy', 'identity_hate': 'accuracy'}) # 'accuracy'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 76681 samples, validate on 19170 samples\n",
      "Epoch 1/10\n",
      "76681/76681 [==============================] - 358s - loss: 0.1426 - acc: 0.9634 - val_loss: 0.1447 - val_acc: 0.9621\n",
      "Epoch 2/10\n",
      "76681/76681 [==============================] - 356s - loss: 0.1422 - acc: 0.9634 - val_loss: 0.1450 - val_acc: 0.9621\n",
      "Epoch 3/10\n",
      "76681/76681 [==============================] - 362s - loss: 0.1418 - acc: 0.9634 - val_loss: 0.1450 - val_acc: 0.9621\n",
      "Epoch 4/10\n",
      "76681/76681 [==============================] - 379s - loss: 0.1408 - acc: 0.9634 - val_loss: 0.1456 - val_acc: 0.9621\n",
      "Epoch 5/10\n",
      "76681/76681 [==============================] - 376s - loss: 0.1404 - acc: 0.9634 - val_loss: 0.1463 - val_acc: 0.9621\n",
      "Epoch 6/10\n",
      "76681/76681 [==============================] - 352s - loss: 0.1405 - acc: 0.9634 - val_loss: 0.1486 - val_acc: 0.9621\n",
      "Epoch 7/10\n",
      "76681/76681 [==============================] - 349s - loss: 0.1395 - acc: 0.9634 - val_loss: 0.1463 - val_acc: 0.9621\n",
      "Epoch 8/10\n",
      "76681/76681 [==============================] - 349s - loss: 0.1387 - acc: 0.9634 - val_loss: 0.1465 - val_acc: 0.9621\n",
      "Epoch 9/10\n",
      "76681/76681 [==============================] - 370s - loss: 0.1373 - acc: 0.9634 - val_loss: 0.1480 - val_acc: 0.9621\n",
      "Epoch 10/10\n",
      "76681/76681 [==============================] - 382s - loss: 0.1356 - acc: 0.9634 - val_loss: 0.1496 - val_acc: 0.9621\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f9f38651c10>"
      ]
     },
     "execution_count": 163,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "batch_size = 32\n",
    "epochs = 10\n",
    "model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, validation_data=(x_val, y_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         (None, 200)               0         \n",
      "_________________________________________________________________\n",
      "embedding_7 (Embedding)      (None, 200, 300)          6000000   \n",
      "_________________________________________________________________\n",
      "lstm_layer (LSTM)            (None, 200, 60)           86640     \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d_6 (Glob (None, 60)                0         \n",
      "_________________________________________________________________\n",
      "dropout_11 (Dropout)         (None, 60)                0         \n",
      "_________________________________________________________________\n",
      "dense_11 (Dense)             (None, 50)                3050      \n",
      "_________________________________________________________________\n",
      "dropout_12 (Dropout)         (None, 50)                0         \n",
      "_________________________________________________________________\n",
      "dense_12 (Dense)             (None, 6)                 306       \n",
      "=================================================================\n",
      "Total params: 6,089,996\n",
      "Trainable params: 89,996\n",
      "Non-trainable params: 6,000,000\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "226998/226998 [==============================] - 205s   \n"
     ]
    }
   ],
   "source": [
    "y_test = model.predict([X_t], batch_size=1024, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sample_submission = pd.read_csv('data/sample_submission.csv')\n",
    "sample_submission[classes] = y_test\n",
    "sample_submission.to_csv('submissions/keras_lstm_submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "# A claimed improvement based on Jermey Howrad's [kernel](https://www.kaggle.com/jhoward/improved-lstm-baseline-glove-dropout)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(-0.0058384896, 0.48782218)"
      ]
     },
     "execution_count": 167,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_embs = np.stack(embedding_index_glove.values())\n",
    "emb_mean,emb_std = all_embs.mean(), all_embs.std()\n",
    "emb_mean,emb_std"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# prepare embedding matrix\n",
    "EMBEDDING_DIM = 300\n",
    "num_words = min(max_features, len(word_index) + 1)\n",
    "embedding_matrix_modified = np.random.normal(emb_mean, emb_std, (num_words, EMBEDDING_DIM))\n",
    "for word, i in word_index.items():\n",
    "    if i >= max_features:\n",
    "        continue\n",
    "    embedding_vector = embedding_index_glove.get(word)\n",
    "    if embedding_vector is not None:\n",
    "        # words not found in embedding index will be all-zeros.\n",
    "        embedding_matrix_modified[i] = embedding_vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 169,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "inp = Input(shape=(maxlen,))\n",
    "x = Embedding(max_features, EMBEDDING_DIM, weights=[embedding_matrix_modified])(inp)\n",
    "x = Bidirectional(LSTM(50, return_sequences=True, dropout=0.1, recurrent_dropout=0.1))(x)\n",
    "x = GlobalMaxPool1D()(x)\n",
    "x = Dense(50, activation=\"relu\")(x)\n",
    "x = Dropout(0.1)(x)\n",
    "x = Dense(6, activation=\"sigmoid\")(x)\n",
    "model_mod = Model(inputs=inp, outputs=x)\n",
    "model_mod.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 173,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 173,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 76681 samples, validate on 19170 samples\n",
      "Epoch 1/2\n",
      "76681/76681 [==============================] - 885s - loss: 0.1468 - acc: 0.9631 - val_loss: 0.1446 - val_acc: 0.9621\n",
      "Epoch 2/2\n",
      "76681/76681 [==============================] - 795s - loss: 0.1410 - acc: 0.9634 - val_loss: 0.1461 - val_acc: 0.9621\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f9f37352a90>"
      ]
     },
     "execution_count": 174,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "batch_size = 32\n",
    "epochs = 2\n",
    "model_mod.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, validation_split=0.1, \n",
    "              validation_data=(x_val, y_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_3 (InputLayer)         (None, 200)               0         \n",
      "_________________________________________________________________\n",
      "embedding_8 (Embedding)      (None, 200, 300)          6000000   \n",
      "_________________________________________________________________\n",
      "bidirectional_1 (Bidirection (None, 200, 100)          140400    \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d_7 (Glob (None, 100)               0         \n",
      "_________________________________________________________________\n",
      "dense_13 (Dense)             (None, 50)                5050      \n",
      "_________________________________________________________________\n",
      "dropout_13 (Dropout)         (None, 50)                0         \n",
      "_________________________________________________________________\n",
      "dense_14 (Dense)             (None, 6)                 306       \n",
      "=================================================================\n",
      "Total params: 6,145,756\n",
      "Trainable params: 6,145,756\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_mod.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "226998/226998 [==============================] - 183s   \n"
     ]
    }
   ],
   "source": [
    "y_test_mod = model.predict([X_t], batch_size=1024, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 177,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sample_submission = pd.read_csv('data/sample_submission.csv')\n",
    "sample_submission[classes] = y_test_mod\n",
    "sample_submission.to_csv('submissions/keras_lstm_mod_submission.csv', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# I will Try to improve the Keras Model by doing the following:\n",
    "- Build Word embedding using the data training text data using [gensim](https://radimrehurek.com/gensim/models/word2vec.html) package"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.tokenize import WordPunctTokenizer\n",
    "from collections import Counter\n",
    "from string import punctuation, ascii_lowercase\n",
    "import regex as re"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 322849/322849 [00:25<00:00, 12862.69it/s]\n"
     ]
    }
   ],
   "source": [
    "# replace urls\n",
    "re_url = re.compile(r\"((http|https)\\:\\/\\/)?[a-zA-Z0-9\\.\\/\\?\\:@\\-_=#]+\\\n",
    "                    .([a-zA-Z]){2,6}([a-zA-Z0-9\\.\\&\\/\\?\\:@\\-_=#])*\",\n",
    "                    re.MULTILINE|re.UNICODE)\n",
    "# replace ips\n",
    "re_ip = re.compile(\"\\d{1,3}\\.\\d{1,3}\\.\\d{1,3}\\.\\d{1,3}\")\n",
    "\n",
    "# setup tokenizer\n",
    "tokenizer = WordPunctTokenizer()\n",
    "\n",
    "vocab = Counter()\n",
    "\n",
    "def text_to_wordlist(text, lower=False):\n",
    "    # replace URLs\n",
    "    text = re_url.sub(\"URL\", text)\n",
    "    \n",
    "    # replace IPs\n",
    "    text = re_ip.sub(\"IPADDRESS\", text)\n",
    "    \n",
    "    # Tokenize\n",
    "    text = tokenizer.tokenize(text)\n",
    "    \n",
    "    # optional: lower case\n",
    "    if lower:\n",
    "        text = [t.lower() for t in text]\n",
    "    \n",
    "    # Return a list of words\n",
    "    vocab.update(text)\n",
    "    return text\n",
    "\n",
    "def process_comments(list_sentences, lower=False):\n",
    "    comments = []\n",
    "    for text in tqdm(list_sentences):\n",
    "        txt = text_to_wordlist(text, lower=lower)\n",
    "        comments.append(txt)\n",
    "    return comments\n",
    "\n",
    "list_sentences_train = list(train[\"comment_text\"].fillna(\"unknown\").values)\n",
    "list_sentences_test = list(test[\"comment_text\"].fillna(\"unknown\").values)\n",
    "\n",
    "comments = process_comments(list_sentences_train + list_sentences_test, lower=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Let's build a CBOW Word2Vec Model by specifing `sg=0` which means to deactivate SkipGram Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from gensim.models import Word2Vec\n",
    "model = Word2Vec(comments, size=100, window=5, min_count=5, workers=8, sg=0,\n",
    "                 negative=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of word vectors: 70910\n"
     ]
    }
   ],
   "source": [
    "word_vectors =  model.wv\n",
    "print(\"Number of word vectors: {}\".format(len(word_vectors.vocab)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('prince', 0.9886927008628845),\n",
       " ('princess', 0.9826782941818237),\n",
       " ('emperor', 0.945073127746582),\n",
       " ('mary', 0.9410067200660706),\n",
       " ('bishop', 0.9352464079856873),\n",
       " ('solomon', 0.9321256875991821),\n",
       " ('throne', 0.9311854839324951),\n",
       " ('lincoln', 0.9298420548439026),\n",
       " ('shah', 0.9296499490737915),\n",
       " ('heir', 0.9235137104988098)]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.wv.most_similar_cosmul(positive=['woman', 'king'], negative=['man'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Initialize the Word Embeddings in Keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Shape of data tensor:', (95851, 200))\n",
      "('Shape of label tensor:', (95851, 6))\n",
      "('Shape of test_data tensor:', (226998, 200))\n"
     ]
    }
   ],
   "source": [
    "max_features = len(word_vectors.vocab)\n",
    "maxlen = 200\n",
    "\n",
    "\n",
    "word_index = {t[0]: i+1 for i,t in enumerate(vocab.most_common(max_features))}\n",
    "train_sequences = [[word_index.get(t, 0) for t in comment]\n",
    "             for comment in comments[:len(list_sentences_train)]]\n",
    "test_sequences = [[word_index.get(t, 0)  for t in comment] \n",
    "                  for comment in comments[len(list_sentences_train):]]\n",
    "\n",
    "# pad\n",
    "train_data = pad_sequences(train_sequences, maxlen=maxlen, \n",
    "                     padding=\"pre\", truncating=\"post\")\n",
    "list_classes = [\"toxic\", \"severe_toxic\", \"obscene\", \"threat\", \"insult\", \"identity_hate\"]\n",
    "y = train[list_classes].values\n",
    "print('Shape of data tensor:', train_data.shape)\n",
    "print('Shape of label tensor:', y.shape)\n",
    "\n",
    "test_data = pad_sequences(test_sequences, maxlen=maxlen, padding=\"pre\",\n",
    "                          truncating=\"post\")\n",
    "print('Shape of test_data tensor:', test_data.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# prepare embedding matrix\n",
    "EMBEDDING_DIM =  100\n",
    "num_words = min(max_features, len(word_vectors.vocab))\n",
    "embedding_matrix = (np.random.rand(num_words, EMBEDDING_DIM) - 0.5) / 5.0 \n",
    "for word, i in word_index.items():\n",
    "    if i >= max_features:\n",
    "        continue\n",
    "    try: \n",
    "        embedding_vector = word_vectors[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\n",
    "    except:\n",
    "        pass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 86265 samples, validate on 9586 samples\n",
      "Epoch 1/10\n",
      "86265/86265 [==============================] - 343s - loss: 0.2408 - acc: 0.9062 - val_loss: 0.0633 - val_acc: 0.9791\n",
      "Epoch 2/10\n",
      "86265/86265 [==============================] - 291s - loss: 0.0650 - acc: 0.9784 - val_loss: 0.0577 - val_acc: 0.9801\n",
      "Epoch 3/10\n",
      "86265/86265 [==============================] - 284s - loss: 0.0588 - acc: 0.9794 - val_loss: 0.0550 - val_acc: 0.9799\n",
      "Epoch 4/10\n",
      "86265/86265 [==============================] - 309s - loss: 0.0553 - acc: 0.9802 - val_loss: 0.0541 - val_acc: 0.9808\n",
      "Epoch 5/10\n",
      "86265/86265 [==============================] - 326s - loss: 0.0531 - acc: 0.9807 - val_loss: 0.0526 - val_acc: 0.9811\n",
      "Epoch 6/10\n",
      "86265/86265 [==============================] - 299s - loss: 0.0520 - acc: 0.9810 - val_loss: 0.0528 - val_acc: 0.9813\n",
      "Epoch 7/10\n",
      "86265/86265 [==============================] - 304s - loss: 0.0498 - acc: 0.9816 - val_loss: 0.0509 - val_acc: 0.9816\n",
      "Epoch 8/10\n",
      "86265/86265 [==============================] - 356s - loss: 0.0485 - acc: 0.9819 - val_loss: 0.0516 - val_acc: 0.9810\n",
      "Epoch 9/10\n",
      "86265/86265 [==============================] - 341s - loss: 0.0472 - acc: 0.9824 - val_loss: 0.0525 - val_acc: 0.9804\n",
      "Epoch 10/10\n",
      "86265/86265 [==============================] - 303s - loss: 0.0465 - acc: 0.9825 - val_loss: 0.0506 - val_acc: 0.9815\n"
     ]
    }
   ],
   "source": [
    "inp = Input(shape=(maxlen,), dtype='int32')\n",
    "x = Embedding(num_words, EMBEDDING_DIM, mask_zero=False, weights=[embedding_matrix],\n",
    "              input_length=maxlen, trainable=False)(inp)\n",
    "x = SpatialDropout1D(0.2)(x)\n",
    "x = Bidirectional(LSTM(64, return_sequences=False))(x)\n",
    "x = Dropout(0.2)(x)\n",
    "x = BatchNormalization()(x)\n",
    "x = Dense(6, activation=\"sigmoid\")(x)\n",
    "model_wv = Model(inputs=inp, outputs=x)\n",
    "model_wv.compile(loss='binary_crossentropy',\n",
    "                  optimizer=optimizers.Adam(lr=0.001, clipnorm=.25, beta_1=0.7, beta_2=0.99),\n",
    "                  metrics=['accuracy'])\n",
    "model_wv_history = model_wv.fit([train_data], y, validation_split=0.1,\n",
    "                 epochs=10, batch_size=256, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEWCAYAAAB1xKBvAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff9a6174710>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAEWCAYAAABMoxE0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nKQ4SMzObFAeJmZlNioPEzMwm5f8HXaE7429AueAAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff97b546e90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_loss_curve(model_wv_history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_6 (InputLayer)         (None, 200)               0         \n",
      "_________________________________________________________________\n",
      "embedding_6 (Embedding)      (None, 200, 100)          7091000   \n",
      "_________________________________________________________________\n",
      "spatial_dropout1d_6 (Spatial (None, 200, 100)          0         \n",
      "_________________________________________________________________\n",
      "bidirectional_6 (Bidirection (None, 128)               84480     \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_5 (Batch (None, 128)               512       \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 6)                 774       \n",
      "=================================================================\n",
      "Total params: 7,176,766\n",
      "Trainable params: 85,510\n",
      "Non-trainable params: 7,091,256\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_wv.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/alaa/anaconda2/lib/python2.7/site-packages/keras/backend/tensorflow_backend.py:1154: calling reduce_max (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "keep_dims is deprecated, use keepdims instead\n",
      "Train on 86265 samples, validate on 9586 samples\n",
      "Epoch 1/10\n",
      "86265/86265 [==============================] - 290s - loss: 0.0456 - acc: 0.9828 - val_loss: 0.0520 - val_acc: 0.9812\n",
      "Epoch 2/10\n",
      "86265/86265 [==============================] - 279s - loss: 0.0446 - acc: 0.9831 - val_loss: 0.0512 - val_acc: 0.9817\n",
      "Epoch 3/10\n",
      "86265/86265 [==============================] - 283s - loss: 0.0445 - acc: 0.9830 - val_loss: 0.0501 - val_acc: 0.9821\n",
      "Epoch 4/10\n",
      "86265/86265 [==============================] - 286s - loss: 0.0433 - acc: 0.9836 - val_loss: 0.0503 - val_acc: 0.9819\n",
      "Epoch 5/10\n",
      "86265/86265 [==============================] - 284s - loss: 0.0429 - acc: 0.9836 - val_loss: 0.0507 - val_acc: 0.9817\n",
      "Epoch 6/10\n",
      "86265/86265 [==============================] - 309s - loss: 0.0423 - acc: 0.9837 - val_loss: 0.0509 - val_acc: 0.9818\n",
      "Epoch 7/10\n",
      "86265/86265 [==============================] - 294s - loss: 0.0420 - acc: 0.9837 - val_loss: 0.0507 - val_acc: 0.9821\n",
      "Epoch 8/10\n",
      "86265/86265 [==============================] - 313s - loss: 0.0413 - acc: 0.9840 - val_loss: 0.0523 - val_acc: 0.9807\n",
      "Epoch 9/10\n",
      "86265/86265 [==============================] - 303s - loss: 0.0408 - acc: 0.9841 - val_loss: 0.0500 - val_acc: 0.9818\n",
      "Epoch 10/10\n",
      "86265/86265 [==============================] - 348s - loss: 0.0405 - acc: 0.9843 - val_loss: 0.0505 - val_acc: 0.9819\n"
     ]
    }
   ],
   "source": [
    "# Apossible altrnative architecture\n",
    "\n",
    "inp = Input(shape=(maxlen,), dtype='int32')\n",
    "x = Embedding(num_words, EMBEDDING_DIM, mask_zero=False, weights=[embedding_matrix],\n",
    "              input_length=maxlen, trainable=False)(inp)\n",
    "x = Conv1D(128, 5, activation='relu')(x)\n",
    "x = MaxPooling1D(5)(x)\n",
    "x = Conv1D(128, 5, activation='relu')(x)\n",
    "x = GlobalMaxPooling1D()(x)\n",
    "x = Dense(6, activation=\"sigmoid\")(x)\n",
    "model_wv_cv = Model(inputs=inp, outputs=x)\n",
    "model_wv_cv.compile(loss='binary_crossentropy',\n",
    "                  optimizer=optimizers.Adam(lr=0.001, clipnorm=.25, beta_1=0.7, beta_2=0.99),\n",
    "                  metrics=['accuracy'])\n",
    "model_wv_cv_history = model_wv.fit([train_data], y, validation_split=0.1,\n",
    "                 epochs=10, batch_size=256, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZgAAAEWCAYAAABbgYH9AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff97b447b50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAEWCAYAAABMoxE0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSEREJCoKEhERiYqCREREovL/A4LB3Adtj0h6AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff977de6a50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_loss_curve(model_wv_cv_history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_10 (InputLayer)        (None, 200)               0         \n",
      "_________________________________________________________________\n",
      "embedding_10 (Embedding)     (None, 200, 100)          7091000   \n",
      "_________________________________________________________________\n",
      "conv1d_4 (Conv1D)            (None, 196, 128)          64128     \n",
      "_________________________________________________________________\n",
      "max_pooling1d_2 (MaxPooling1 (None, 39, 128)           0         \n",
      "_________________________________________________________________\n",
      "conv1d_5 (Conv1D)            (None, 35, 128)           82048     \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d_1 (Glob (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 6)                 774       \n",
      "=================================================================\n",
      "Total params: 7,237,950\n",
      "Trainable params: 146,950\n",
      "Non-trainable params: 7,091,000\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_wv_cv.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}