TFIDF vs. Word Embedding Demo  .ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings; warnings.simplefilter('ignore')\n",
    "from sklearn.datasets import fetch_20newsgroups\n",
    "from gensim import models\n",
    "import scipy.spatial.distance as distance\n",
    "import numpy as np\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from keras.utils import to_categorical\n",
    "from sklearn.model_selection import train_test_split\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.models import load_model, Model, Sequential\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.utils import to_categorical\n",
    "from keras.layers import Embedding, Input, Conv1D, MaxPooling1D, Flatten, Dense, GlobalMaxPooling1D\n",
    "from keras.layers import Multiply, concatenate, Activation, Dropout, Reshape, Lambda\n",
    "from keras import regularizers\n",
    "from keras.callbacks import ModelCheckpoint, EarlyStopping"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras import backend as K"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_history(history):\n",
    "    ## As loss always exists\n",
    "    epochs = range(1,len(history.history['loss']) + 1)\n",
    "    \n",
    "    ## Loss\n",
    "    plt.figure(1)\n",
    "    plt.plot(epochs, history.history['loss'], 'b', label='Training loss ')\n",
    "    plt.plot(epochs, history.history['val_loss'], 'g', label='Validation loss ')\n",
    "    \n",
    "    plt.title('Loss')\n",
    "    plt.xlabel('Epochs')\n",
    "    plt.ylabel('Loss')\n",
    "    plt.legend()\n",
    "    \n",
    "    ## Accuracy\n",
    "    plt.figure(2)\n",
    "    plt.plot(epochs, history.history['acc'], 'b', label='Training accuracy ')\n",
    "    plt.plot(epochs, history.history['val_acc'], 'g', label='Validation accuracy ')\n",
    "\n",
    "    plt.title('Accuracy')\n",
    "    plt.xlabel('Epochs')\n",
    "    plt.ylabel('Accuracy')\n",
    "    plt.legend()\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Get the Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = fetch_20newsgroups(subset= 'all',shuffle=True, random_state=1)#, remove=('headers', 'footers', 'quotes'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.corpus import stopwords\n",
    "from string import punctuation\n",
    "def clean_text(texts,  min_length = 2):\n",
    "    clean = []\n",
    "    #don't remove apostrophes\n",
    "    translator = str.maketrans(punctuation.replace('\\'',' '), ' '*len(punctuation))\n",
    "    for text in texts:\n",
    "        text = text.translate(translator)\n",
    "        tokens = text.split()\n",
    "        # remove not alphabetic tokens\n",
    "        tokens = [word.lower() for word in tokens if word.isalpha()]\n",
    "        # filter out stop words\n",
    "        stop_words = stopwords.words('english')\n",
    "        tokens = [w for w in tokens if not w in stop_words]\n",
    "        # filter out short tokens\n",
    "        tokens = [word for word in tokens if len(word) >= min_length]\n",
    "        tokens = ' '.join(tokens)\n",
    "        clean.append(tokens)\n",
    "    return clean"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# tagged data from doc2vec training\n",
    "def tag_text(all_text, tag_type =''):\n",
    "    tagged_text = []\n",
    "    for i, text in enumerate(all_text):\n",
    "        tag = i\n",
    "        tagged_text.append(models.doc2vec.TaggedDocument(text.split(), [tag]))\n",
    "    return tagged_text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_samples = len(dataset.data)\n",
    "data = clean_text(dataset.data)\n",
    "tagged_data = tag_text(data)\n",
    "data_labels = dataset.target\n",
    "data_label_names = dataset.target_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "cat_data_labels = to_categorical(data_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "20\n"
     ]
    }
   ],
   "source": [
    "num_labels = len(data_label_names)\n",
    "print(num_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 20 Topics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['alt.atheism',\n",
       " 'comp.graphics',\n",
       " 'comp.os.ms-windows.misc',\n",
       " 'comp.sys.ibm.pc.hardware',\n",
       " 'comp.sys.mac.hardware',\n",
       " 'comp.windows.x',\n",
       " 'misc.forsale',\n",
       " 'rec.autos',\n",
       " 'rec.motorcycles',\n",
       " 'rec.sport.baseball',\n",
       " 'rec.sport.hockey',\n",
       " 'sci.crypt',\n",
       " 'sci.electronics',\n",
       " 'sci.med',\n",
       " 'sci.space',\n",
       " 'soc.religion.christian',\n",
       " 'talk.politics.guns',\n",
       " 'talk.politics.mideast',\n",
       " 'talk.politics.misc',\n",
       " 'talk.religion.misc']"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_label_names"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 18846 Posts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "18846"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Wrappers to perform training-testing and cross-validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {},
   "outputs": [],
   "source": [
    "#randomly split the data set and the labels into n_cv folds and select fold_n as testing data and the rest as training\n",
    "def create_single_fold(data, labels, n_cv, fold_n):\n",
    "    data_size = data.shape[0]\n",
    "    cv_folds = np.random.randint(0, high = n_cv, size = data_size)\n",
    "    train_data = data[cv_folds != fold_n]\n",
    "    train_labels = labels[cv_folds != fold_n]\n",
    "    test_data = data[cv_folds == fold_n]\n",
    "    test_labels = labels[cv_folds == fold_n]\n",
    "    return (train_data, train_labels, test_data, test_labels)\n",
    "\n",
    "#randomly split the data set and the labels into n_cv folds and yeild testing data and training data for each fold\n",
    "def create_all_folds(data, labels, n_cv):\n",
    "    data_size = data.shape[0]\n",
    "    cv_folds = np.random.randint(0, high = n_cv, size = data_size)\n",
    "    for fold_n in range(n_cv):\n",
    "        train_data = data[cv_folds != fold_n]\n",
    "        train_labels = labels[cv_folds != fold_n]\n",
    "        test_data = data[cv_folds == fold_n]\n",
    "        test_labels = labels[cv_folds == fold_n]\n",
    "        yield (train_data, train_labels, test_data, test_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tfidf representation of the document-word matrix\n",
    "\n",
    "Tfidf - term frequency inverse document frequency -   measures the importance of terms in the documents based on how frequently they occur in the corpus of documents. It gives more weight to terms appearing in fewer documents.\n",
    "\n",
    "The 'raw' formula for tfidf is \n",
    "\n",
    "$tfidf_{i,j} = tf_{i,j} \\times idf_i =  tf_{i,j} \\times log(\\frac{N}{df_i})$\n",
    "\n",
    "Here $tf_{i,j}$ is the frequency of term $i$ in document $j$, $idf_i$ is the inverse document frequency, $N$ is the number of documents, and $df_i$ is the number of documents in which term $i$ appears.\n",
    "\n",
    "In practice the raw formula is modified. 1 is added to the definition of $idf_i$ to avoid the inverse document frequency turning to 0 when the term is present in all documents. \n",
    "\n",
    "In addition TfidfVectorizer provides an option of either $l1$ or $l2$ normalization for $tfidf_{i,j}$\n",
    "\n",
    "Tfidf representation is used with  Dense Neural Networks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def sklearn_tokenize(input_texts, vocab_size, ngram_range, mystopwords = [],  max_df = 1.0, min_df = 1):\n",
    "    vectorizer = TfidfVectorizer(max_df=max_df, max_features=vocab_size, ngram_range = ngram_range,\n",
    "                                     min_df=min_df,  norm='l2',\n",
    "                                     use_idf=True, smooth_idf = False)\n",
    "    X = vectorizer.fit_transform(input_texts)\n",
    "    return (X.todense(), vectorizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(18846, 5000)\n"
     ]
    }
   ],
   "source": [
    "#pick the top VOCAB_SIZE words from the corpus based on their counts\n",
    "VOCAB_SIZE = 5000\n",
    "\n",
    "#tokenize raw texts using sklearn\n",
    "tfidf_data, vectorizer = sklearn_tokenize(data,  VOCAB_SIZE, (1,1), max_df = 0.5)\n",
    "print(tfidf_data.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "# split the tfidf data into training/validation = 4/1 \n",
    "train_size = 0.8\n",
    "x_tfidf_train, x_tfidf_test,y_tfidf_train,  y_tfidf_test = \\\n",
    "                        train_test_split(tfidf_data, cat_data_labels, train_size = train_size)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## DenseNN  with 1 Hidden Layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCH_SIZE = 128"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_size = VOCAB_SIZE\n",
    "dropout = 0.3\n",
    "activation = 'tanh'\n",
    "layer_size = 60\n",
    "inputs = Input(shape = (input_size,),  name='dnn_input')\n",
    "x = Dense(layer_size, activation=activation, name ='hidden_dnn')(inputs)\n",
    "x = Dropout(dropout, name='dropout')(x)\n",
    "preds = Dense(num_labels, activation='softmax', name = 'output_dnn')(x)\n",
    "\n",
    "model = Model(inputs, preds)\n",
    "model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dnn_input (InputLayer)       (None, 5000)              0         \n",
      "_________________________________________________________________\n",
      "hidden_dnn (Dense)           (None, 60)                300060    \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 60)                0         \n",
      "_________________________________________________________________\n",
      "output_dnn (Dense)           (None, 20)                1220      \n",
      "=================================================================\n",
      "Total params: 301,280\n",
      "Trainable params: 301,280\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Explanation: Output Shape and Number of Parameters \n",
    "\n",
    "```\n",
    "Layer (type)                 Output Shape              Param #   \n",
    "=================================================================\n",
    "dnn_input (InputLayer)       (None, 5000)              0         \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 5000: Vocabular Size\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "hidden_dnn (Dense)           (None, 60)                300060    \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 60: Hidden Layer Size\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- (Vocabulary Size +1)  x Hidden Layer Size =  $5001 * 60 = 300,060$  \n",
    "    - 1 is added for the bias term\n",
    "\n",
    "```     \n",
    "_________________________________________________________________\n",
    "dropout (Dropout)            (None, 60)                0         \n",
    "_________________________________________________________________\n",
    "output_dnn (Dense)           (None, 20)                1220   \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 20: Output Layer Size\n",
    "\n",
    "#### Number of Parameters\n",
    "- (Hidden Layer Size + 1) x  Output Layer Size = $61 * 20 = 1,220$\n",
    "    - 1 is added for the bias term\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15076 samples, validate on 3770 samples\n",
      "Epoch 1/30\n",
      "15076/15076 [==============================] - 2s 155us/step - loss: 2.6063 - acc: 0.6430 - val_loss: 2.0512 - val_acc: 0.7926\n",
      "Epoch 2/30\n",
      "15076/15076 [==============================] - 2s 118us/step - loss: 1.5152 - acc: 0.8252 - val_loss: 1.1280 - val_acc: 0.8276\n",
      "Epoch 3/30\n",
      "15076/15076 [==============================] - 2s 121us/step - loss: 0.8554 - acc: 0.8721 - val_loss: 0.7632 - val_acc: 0.8507\n",
      "Epoch 4/30\n",
      "15076/15076 [==============================] - 2s 122us/step - loss: 0.5729 - acc: 0.9036 - val_loss: 0.6048 - val_acc: 0.8671\n",
      "Epoch 5/30\n",
      "15076/15076 [==============================] - 2s 120us/step - loss: 0.4257 - acc: 0.9240 - val_loss: 0.5222 - val_acc: 0.8735\n",
      "Epoch 6/30\n",
      "15076/15076 [==============================] - 2s 119us/step - loss: 0.3314 - acc: 0.9399 - val_loss: 0.4718 - val_acc: 0.8793\n",
      "Epoch 7/30\n",
      "15076/15076 [==============================] - 2s 117us/step - loss: 0.2658 - acc: 0.9532 - val_loss: 0.4395 - val_acc: 0.8857\n",
      "Epoch 8/30\n",
      "15076/15076 [==============================] - 2s 119us/step - loss: 0.2167 - acc: 0.9617 - val_loss: 0.4172 - val_acc: 0.8873\n",
      "Epoch 9/30\n",
      "15076/15076 [==============================] - 2s 117us/step - loss: 0.1803 - acc: 0.9703 - val_loss: 0.4028 - val_acc: 0.8886\n",
      "Epoch 10/30\n",
      "15076/15076 [==============================] - 2s 118us/step - loss: 0.1507 - acc: 0.9772 - val_loss: 0.3925 - val_acc: 0.8905\n",
      "Epoch 11/30\n",
      "15076/15076 [==============================] - 2s 121us/step - loss: 0.1265 - acc: 0.9816 - val_loss: 0.3848 - val_acc: 0.8899\n",
      "Epoch 12/30\n",
      "15076/15076 [==============================] - 2s 121us/step - loss: 0.1051 - acc: 0.9863 - val_loss: 0.3794 - val_acc: 0.8902\n",
      "Epoch 13/30\n",
      "15076/15076 [==============================] - 2s 119us/step - loss: 0.0904 - acc: 0.9896 - val_loss: 0.3772 - val_acc: 0.8910\n",
      "Epoch 14/30\n",
      "15076/15076 [==============================] - 2s 119us/step - loss: 0.0771 - acc: 0.9922 - val_loss: 0.3759 - val_acc: 0.8920\n",
      "Epoch 15/30\n",
      "15076/15076 [==============================] - 2s 119us/step - loss: 0.0665 - acc: 0.9943 - val_loss: 0.3760 - val_acc: 0.8920\n",
      "Epoch 16/30\n",
      "15076/15076 [==============================] - 2s 123us/step - loss: 0.0574 - acc: 0.9948 - val_loss: 0.3767 - val_acc: 0.8934\n",
      "Epoch 17/30\n",
      "15076/15076 [==============================] - 2s 123us/step - loss: 0.0493 - acc: 0.9962 - val_loss: 0.3780 - val_acc: 0.8910\n",
      "Epoch 18/30\n",
      "15076/15076 [==============================] - 2s 121us/step - loss: 0.0447 - acc: 0.9968 - val_loss: 0.3803 - val_acc: 0.8923\n",
      "Epoch 19/30\n",
      "15076/15076 [==============================] - 2s 121us/step - loss: 0.0379 - acc: 0.9979 - val_loss: 0.3833 - val_acc: 0.8907\n"
     ]
    }
   ],
   "source": [
    "earlystop = EarlyStopping(monitor='val_loss', patience=5)\n",
    "history = model.fit(x_tfidf_train, y_tfidf_train, validation_data=(x_tfidf_test, y_tfidf_test), \n",
    "                    epochs=30, batch_size=BATCH_SIZE, callbacks=[earlystop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best validation accuracy is  0.8933687006130775\n"
     ]
    }
   ],
   "source": [
    "print ('Best validation accuracy is ', max(history.history['val_acc']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Overfitting occurs if trained too much \n",
    "\n",
    "Note, early stopping criteria kicked in at Epoch 21"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 798,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_history(history)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"images/TFIDF Accuracy.png\" alt=\"Drawing\" style=\"width: 600px;\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Increase of the vocabulary size allows for increase in the network size without deterioration of accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# DenseNN for Doc2vec"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Document Embedding Dimensions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "DOC2VEC_SIZE = 1000"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train Doc2vec embedding using gensim"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_dbow = models.Doc2Vec(dm=0, dbow_words = 0, vector_size=DOC2VEC_SIZE, min_count=4, alpha=0.065, min_alpha=0.005)\n",
    "model_dbow.build_vocab(tagged_data)\n",
    "model_dbow.train(tagged_data, total_examples=len(tagged_data), epochs=30)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Input to a DNN is the Doc2vec vector"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(18846, 1000)\n"
     ]
    }
   ],
   "source": [
    "dbow_doc2vec = []\n",
    "for i in range(n_samples):\n",
    "    dbow_doc2vec.append(model_dbow.docvecs[i])\n",
    "dbow_doc2vec = np.array(dbow_doc2vec)\n",
    "print (dbow_doc2vec.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### dbow_doc2vec is a matrix with the dimensions Number of Posts x Doc2vec Embedding size = 18846 x 1000 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "# split the tfidf data into training/validation = 4/1 \n",
    "train_size = 0.8\n",
    "x_doc_train, x_doc_test, y_doc_train, y_doc_test = \\\n",
    "                        train_test_split(dbow_doc2vec, cat_data_labels, train_size = train_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_size = DOC2VEC_SIZE\n",
    "dropout = 0.3\n",
    "activation = 'tanh'\n",
    "layer_size = 100\n",
    "inputs = Input(shape = (input_size,),  name='dnn_input')\n",
    "x = Dense(layer_size, activation=activation, name ='hidden_dnn')(inputs)\n",
    "x = Dropout(dropout, name='dropout')(x)\n",
    "preds = Dense(num_labels, activation='softmax', name = 'output_dnn')(x)\n",
    "\n",
    "model = Model(inputs, preds)\n",
    "model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dnn_input (InputLayer)       (None, 1000)              0         \n",
      "_________________________________________________________________\n",
      "hidden_dnn (Dense)           (None, 100)               100100    \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 100)               0         \n",
      "_________________________________________________________________\n",
      "output_dnn (Dense)           (None, 20)                2020      \n",
      "=================================================================\n",
      "Total params: 102,120\n",
      "Trainable params: 102,120\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Explanation of Output Shape and Number of Parameters \n",
    "\n",
    "```\n",
    "Layer (type)                 Output Shape              Param #   \n",
    "=================================================================\n",
    "dnn_input (InputLayer)       (None, 1000)              0         \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1000: Doc2Vec Embedding Size\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "\n",
    "hidden_dnn (Dense)           (None, 100)               100100      \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 100: Hidden Layer Size\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- (Document Embedding Dimensions + 1)  x Hidden Layer Size =  $1001 \\times 100 = 100,100$\n",
    "    - 1 is added for the bias term\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "dropout (Dropout)          (None, 100)               0         \n",
    "_________________________________________________________________\n",
    "output_dnn (Dense)           (None, 20)                2020   \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 20: Output Layer Size\n",
    "\n",
    "#### Number of Parameters\n",
    "- (Hidden Layer Size + 1) x  Output Layer Size = $101 \\times 20 = 2,020$\n",
    "    - 1 is added for the bias term\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15076 samples, validate on 3770 samples\n",
      "Epoch 1/30\n",
      "15076/15076 [==============================] - 1s 66us/step - loss: 1.4975 - acc: 0.6500 - val_loss: 0.7110 - val_acc: 0.8435\n",
      "Epoch 2/30\n",
      "15076/15076 [==============================] - 0s 32us/step - loss: 0.6188 - acc: 0.8394 - val_loss: 0.5147 - val_acc: 0.8650\n",
      "Epoch 3/30\n",
      "15076/15076 [==============================] - 0s 31us/step - loss: 0.4771 - acc: 0.8683 - val_loss: 0.4578 - val_acc: 0.8663\n",
      "Epoch 4/30\n",
      "15076/15076 [==============================] - 0s 29us/step - loss: 0.4094 - acc: 0.8859 - val_loss: 0.4357 - val_acc: 0.8706\n",
      "Epoch 5/30\n",
      "15076/15076 [==============================] - 0s 32us/step - loss: 0.3667 - acc: 0.8941 - val_loss: 0.4256 - val_acc: 0.8666\n",
      "Epoch 6/30\n",
      "15076/15076 [==============================] - 0s 30us/step - loss: 0.3345 - acc: 0.9036 - val_loss: 0.4267 - val_acc: 0.8674\n",
      "Epoch 7/30\n",
      "15076/15076 [==============================] - 0s 29us/step - loss: 0.3181 - acc: 0.9042 - val_loss: 0.4209 - val_acc: 0.8682\n",
      "Epoch 8/30\n",
      "15076/15076 [==============================] - 1s 39us/step - loss: 0.2912 - acc: 0.9128 - val_loss: 0.4265 - val_acc: 0.8653\n",
      "Epoch 9/30\n",
      "15076/15076 [==============================] - 0s 30us/step - loss: 0.2723 - acc: 0.9191 - val_loss: 0.4269 - val_acc: 0.8666\n",
      "Epoch 10/30\n",
      "15076/15076 [==============================] - 0s 30us/step - loss: 0.2598 - acc: 0.9226 - val_loss: 0.4303 - val_acc: 0.8639\n",
      "Epoch 11/30\n",
      "15076/15076 [==============================] - 0s 32us/step - loss: 0.2469 - acc: 0.9262 - val_loss: 0.4361 - val_acc: 0.8647\n",
      "Epoch 12/30\n",
      "15076/15076 [==============================] - 0s 29us/step - loss: 0.2356 - acc: 0.9306 - val_loss: 0.4392 - val_acc: 0.8618\n",
      "Wall time: 6.66 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "earlystop = EarlyStopping(monitor='val_loss', patience=5)\n",
    "history = model.fit(x_doc_train, y_doc_train, validation_data=(x_doc_test, y_doc_test), \n",
    "                    epochs=30, batch_size=BATCH_SIZE, callbacks=[earlystop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best validation accuracy is  0.8933687006130775\n"
     ]
    }
   ],
   "source": [
    "print ('Best validation accuracy is ', max(history.history['val_acc']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Overfitting occurs if trained too much \n",
    "\n",
    "Note, early stopping criteria kicked in at Epoch 13"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "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"
    },
    {
     "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": [
    "plot_history(history)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"images/DOC2VEC Accuracy.png\" alt=\"Drawing\" style=\"width: 600px;\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Word Embedding"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ```make_sequence``` \n",
    "```make_sequence``` converts a list of texts into a matrix of tokens using ```Tokenizer```.\n",
    "```Tokenizer``` turns text (a sentence, a paragraph, etc.) into sequences of tokens: word $\\rightarrow$ token.\n",
    "word_index is the dictionary used to transform the words into the tokens: \n",
    "```\n",
    "{'UNK': 1,\n",
    "'might', 77,\n",
    "'play', 317,\n",
    "'baseball', 616,\n",
    "'lets', 2429,\n",
    "'rain', 3057,\n",
    "'tomorrow', 3102\n",
    " ...\n",
    "```\n",
    "Sequences have to be the same size to be fed into a NN, hence padding with 0's to max_sequence_length.\n",
    "E.g. 2 input sentences *'Lets play baseball tomorrow'* and *'It might rain tomorrow'* and max_sequence_length = 5. The output of ```make_sequence``` will be a matrix (*'it'* is a stopword):\n",
    "\n",
    "$\n",
    "\\begin{array}{cc}\n",
    "\\textit {lets play baseball tomorrow} \\\\\n",
    "\\textit {might rain tomorrow}\\\\\n",
    "\\end{array}\n",
    "\\begin{bmatrix}\n",
    "2429 & 317& 616& 3102& 0 \\\\\n",
    "77 & 3057& 3102& 0& 0\\\\\n",
    "\\end{bmatrix}\n",
    "$\n",
    "\n",
    "\n",
    "## Keras ```Embedding``` Layer\n",
    "```Embedding``` turns each token, and by extension each word, into a vector, the size of the vector is EMBEDDING_DIM.\n",
    "\n",
    "E.g. EMBEDDING_DIM = 3,  max_sequence_length = 5. Sentences *'Lets play baseball tomorrow'* will converted into something like this matrix \n",
    "$\n",
    "\\begin{matrix}\n",
    "lets \\\\\n",
    "play\\\\\n",
    "baseball\\\\\n",
    "tomorrow\\\\ \n",
    "\\\\\n",
    "\\end{matrix}\n",
    "\\begin{pmatrix}\n",
    "2429 \\\\\n",
    "317\\\\\n",
    "616\\\\\n",
    "3102\\\\ \n",
    "0\\\\\n",
    "\\end{pmatrix}\n",
    "\\begin{bmatrix}\n",
    "0.1 & 0.3& 0.7 \\\\\n",
    "0.2& -0.1& 0.6\\\\\n",
    "0.1& 0.6& -0.2\\\\\n",
    "-0.7& 0.1& 0.8\\\\ \n",
    "0.0& 0.0& 0.0\\\\\n",
    "\\end{bmatrix}\n",
    "$\n",
    "\n",
    "and *'It might rain tomorrow'* will converted into something like this matrix\n",
    "\n",
    "$\n",
    "\\begin{matrix}\n",
    "might \\\\\n",
    "rain\\\\\n",
    "tomorrow\\\\ \n",
    "\\\\\n",
    "\\\\\n",
    "\\end{matrix}\n",
    "\\begin{pmatrix}\n",
    "77 \\\\\n",
    "3057\\\\\n",
    "3102\\\\ \n",
    "0\\\\\n",
    "0\\\\\n",
    "\\end{pmatrix}\n",
    "\\begin{bmatrix}\n",
    "-0.4 & 0.2& 0.9 \\\\\n",
    "0.5& 0.1& -0.2\\\\\n",
    "-0.7& 0.1& 0.8\\\\ \n",
    "0.0& 0.0& 0.0\\\\\n",
    "0.0& 0.0& 0.0\\\\\n",
    "\\end{bmatrix}\n",
    "$\n",
    "\n",
    "## Combination\n",
    "Combination of ```make_sequence``` and ```Embedding``` converts a list of texts into a 3D tensor:\n",
    "<img src=\"images/3dtensor.png\" alt=\"Drawing\" style=\"width: 300px;\" />"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_sequences(input_texts, max_nb_words, max_sequence_length = 0, percentile = 99, verbose = 0, output_tokenizer = False):\n",
    "# input_texts is a list of strings/texts\n",
    "# if max_sequence_length is not given, it is set to the percentile of all the lengths of input_texts\n",
    "# select top vocab_size words based on the word counts\n",
    "# word_index is the dictionary used to transform the words into the tokens. \n",
    "    tokenizer = Tokenizer(oov_token='UNK',num_words=max_nb_words+1)# <= +1 because tokenizer is 1 indexed\n",
    "    tokenizer.fit_on_texts(input_texts)\n",
    "    sequences = tokenizer.texts_to_sequences(input_texts)\n",
    "    ml = [len(s) for s in sequences]\n",
    "    if max_sequence_length == 0:\n",
    "        max_sequence_length = int(np.percentile(ml, percentile))\n",
    "    #padding can be done before 'pre' ore after 'post' the existing text\n",
    "    data = pad_sequences(sequences, maxlen=max_sequence_length, padding='post', truncating='post')\n",
    "#the fix below is found in https://github.com/keras-team/keras/issues/8092\n",
    "#TODO follow on how they fix it\n",
    "    word_index = {e:i for e,i in tokenizer.word_index.items()  if i <= max_nb_words} # <= because tokenizer is 1 indexed\n",
    "#    word_index[tokenizer.oov_token] = max_nb_words + 1\n",
    "    if verbose > 0:\n",
    "        print('Found %s unique tokens.' % len(tokenizer.word_index))\n",
    "        print (len(input_texts))\n",
    "        print (len(sequences))\n",
    "    if verbose > 1:\n",
    "        out = plt.plot(ml, 'b.')\n",
    "    if output_tokenizer:\n",
    "        return (data , word_index, tokenizer)\n",
    "    return (data , word_index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(18846, 1100)\n",
      "1100\n"
     ]
    }
   ],
   "source": [
    "VOCAB_SIZE = 5000\n",
    "# word_index is the dictionary used to transform the words into the tokens. \n",
    "#use the default of 99 percentile to define MAX_SEQUENCE_LENGTH\n",
    "sequence_data, word_index = make_sequences(data, VOCAB_SIZE, percentile=99)\n",
    "\n",
    "print (sequence_data.shape)\n",
    "MAX_SEQUENCE_LENGTH = sequence_data.shape[1]\n",
    "print (MAX_SEQUENCE_LENGTH)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### sequence_data is a matrix with the dimensions Number of Posts x MAX_SEQUENCE_LENGTH = 18846 x 11000 "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Find the tokens for the words used in the explanation above"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[['might', 77],\n",
       " ['play', 317],\n",
       " ['baseball', 616],\n",
       " ['lets', 2429],\n",
       " ['rain', 3057],\n",
       " ['tomorrow', 3102]]"
      ]
     },
     "execution_count": 78,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "words = ['lets','play', 'baseball',  'tomorrow','it','rain','might']\n",
    "[[i,e] for i,e in word_index.items() if i in words]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Dense Neural Network (DNN) with Word Embedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [],
   "source": [
    "# split the tfidf data into training/validation = 4/1 \n",
    "train_size = 0.8\n",
    "x_train, x_test, y_train, y_test = \\\n",
    "                        train_test_split(sequence_data, cat_data_labels, train_size = train_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "EMBEDDING_DIM = 64\n",
    "layer_size = 25\n",
    "dropout = 0.1\n",
    "sequence_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32', name='dnn_input')\n",
    "embedding_layer = Embedding(len(word_index) + 1, EMBEDDING_DIM, input_length=MAX_SEQUENCE_LENGTH, name = 'embedding_dnn')\n",
    "embedded_sequences = embedding_layer(sequence_input)\n",
    "x = Flatten(name='flatten_dnn')(embedded_sequences)\n",
    "x = Dense(layer_size, activation='relu', name ='hidden_dense_dnn')(x)\n",
    "x = Dropout(dropout, name='dropout')(x)\n",
    "preds = Dense(num_labels, activation='softmax', name = 'output_dnn')(x)\n",
    "\n",
    "model = Model(sequence_input, preds)\n",
    "model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dnn_input (InputLayer)       (None, 1100)              0         \n",
      "_________________________________________________________________\n",
      "embedding_dnn (Embedding)    (None, 1100, 64)          320064    \n",
      "_________________________________________________________________\n",
      "flatten_dnn (Flatten)        (None, 70400)             0         \n",
      "_________________________________________________________________\n",
      "hidden_dense_dnn (Dense)     (None, 25)                1760025   \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 25)                0         \n",
      "_________________________________________________________________\n",
      "output_dnn (Dense)           (None, 20)                520       \n",
      "=================================================================\n",
      "Total params: 2,080,609\n",
      "Trainable params: 2,080,609\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Explanation of Output Shape and Number of Parameters \n",
    "\n",
    "```\n",
    "Layer (type)                 Output Shape              Param #   \n",
    "=================================================================\n",
    "dnn_input (InputLayer)         (None, 1100)              0      \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1100: Sequence Length\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "embedding_dnn (Embedding)      (None, 1100, 64)          320064      \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1100: Sequence Length\n",
    "- 64: Word Embedding Dimensions\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- (Vocabulary Size + 1)  x Word Embedding Dimensions =  $5001 \\times 64 = 320,064$  \n",
    "    - 1 is added for the 0's of the padding\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "flatten_dnn (Flatten)          (None, 70400)             0\n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 70400: Sequence Length x Word Embedding Dimensions = $1100 \\times 64 = 70,400$\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "hidden_dense_dnn (Dense)              (None, 25)                1760025   \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 25: Hidden Layer Size\n",
    "\n",
    "#### Number of Parameters\n",
    "- (Flattened Layer Size + 1) x Hidden Layer Size = $70401 \\times 25 = 1,760,025$ \n",
    "- 1 is added for the bias term\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "dropout (Dropout)          (None, 25)                0         \n",
    "_________________________________________________________________\n",
    "output_dnn (Dense)              (None, 20)                520   \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 20: Output Layer Size\n",
    "\n",
    "#### Number of Parameters\n",
    "- (Hidden Layer Size + 1) x  Output Layer Size = $26 \\times 20 = 520$\n",
    "    - 1 is added for the bias term\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15076 samples, validate on 3770 samples\n",
      "Epoch 1/30\n",
      "15076/15076 [==============================] - 7s 444us/step - loss: 2.9998 - acc: 0.0550 - val_loss: 2.9688 - val_acc: 0.0740\n",
      "Epoch 2/30\n",
      "15076/15076 [==============================] - 6s 388us/step - loss: 2.7849 - acc: 0.1144 - val_loss: 2.2253 - val_acc: 0.2780\n",
      "Epoch 3/30\n",
      "15076/15076 [==============================] - 6s 389us/step - loss: 1.3981 - acc: 0.5759 - val_loss: 0.8808 - val_acc: 0.7623\n",
      "Epoch 4/30\n",
      "15076/15076 [==============================] - 6s 394us/step - loss: 0.5725 - acc: 0.8471 - val_loss: 0.6529 - val_acc: 0.8175\n",
      "Epoch 5/30\n",
      "15076/15076 [==============================] - 6s 403us/step - loss: 0.3249 - acc: 0.9209 - val_loss: 0.5754 - val_acc: 0.8302\n",
      "Epoch 6/30\n",
      "15076/15076 [==============================] - 6s 389us/step - loss: 0.2028 - acc: 0.9552 - val_loss: 0.5496 - val_acc: 0.8448\n",
      "Epoch 7/30\n",
      "15076/15076 [==============================] - 6s 391us/step - loss: 0.1290 - acc: 0.9745 - val_loss: 0.5485 - val_acc: 0.8427\n",
      "Epoch 8/30\n",
      "15076/15076 [==============================] - 6s 406us/step - loss: 0.0907 - acc: 0.9836 - val_loss: 0.5451 - val_acc: 0.8462\n",
      "Epoch 9/30\n",
      "15076/15076 [==============================] - 6s 399us/step - loss: 0.0641 - acc: 0.9897 - val_loss: 0.5477 - val_acc: 0.8501\n",
      "Epoch 10/30\n",
      "15076/15076 [==============================] - 6s 398us/step - loss: 0.0492 - acc: 0.9916 - val_loss: 0.5538 - val_acc: 0.8499\n",
      "Epoch 11/30\n",
      "15076/15076 [==============================] - 6s 391us/step - loss: 0.0390 - acc: 0.9942 - val_loss: 0.5762 - val_acc: 0.8509\n",
      "Epoch 12/30\n",
      "15076/15076 [==============================] - 6s 394us/step - loss: 0.0325 - acc: 0.9952 - val_loss: 0.5737 - val_acc: 0.8517\n",
      "Epoch 13/30\n",
      "15076/15076 [==============================] - 6s 394us/step - loss: 0.0264 - acc: 0.9957 - val_loss: 0.5878 - val_acc: 0.8493\n"
     ]
    }
   ],
   "source": [
    "earlystop = EarlyStopping(monitor='val_loss', patience=5)\n",
    "history = model.fit(x_train, y_train, validation_data=(x_test, y_test), epochs=30, batch_size=BATCH_SIZE, callbacks=[earlystop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best validation accuracy is  0.851724138025896\n"
     ]
    }
   ],
   "source": [
    "print ('Best validation accuracy is ', max(history.history['val_acc']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Overfitting occurs if trained too much \n",
    "\n",
    "Note, early stopping criteria kicked in at Epoch 13"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "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"
    },
    {
     "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": [
    "plot_history(history)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"images/WORD2VEC DNN Accuracy.png\" alt=\"Drawing\" style=\"width: 600px;\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Increase in the vocabulary size allows for increase in the network size without deterioration of accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ConvNet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [],
   "source": [
    "EMBEDDING_DIM = 64\n",
    "filters = [32,32]\n",
    "kernels = [2,3]\n",
    "pool_sizes = [3,5]\n",
    "\n",
    "sequence_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32', name = 'sequence_input')\n",
    "embedding_layer = Embedding(len(word_index) + 1, EMBEDDING_DIM, input_length=MAX_SEQUENCE_LENGTH, name = 'word_embedding')\n",
    "embedded_sequences = embedding_layer(sequence_input)\n",
    "\n",
    "x = Conv1D(filters[0], kernels[0], activation='relu', name = 'conv1d_first')(embedded_sequences)\n",
    "x = MaxPooling1D(pool_sizes[0], name ='max_pooling1d_first')(x)\n",
    "x = Conv1D(filters[1], kernels[1], activation='relu', name = 'conv1d_second')(x)\n",
    "x = MaxPooling1D(pool_sizes[1], name ='max_pooling1d_second')(x)\n",
    "x = GlobalMaxPooling1D(name = 'global_max_pooling1d')(x)\n",
    "preds = Dense(num_labels, activation='softmax', name = 'output_conv1d')(x)\n",
    "model = Model(sequence_input, preds)\n",
    "\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam',metrics=['acc'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "sequence_input (InputLayer)  (None, 1100)              0         \n",
      "_________________________________________________________________\n",
      "word_embedding (Embedding)   (None, 1100, 64)          320064    \n",
      "_________________________________________________________________\n",
      "conv1d_first (Conv1D)        (None, 1099, 32)          4128      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_first (MaxPool (None, 366, 32)           0         \n",
      "_________________________________________________________________\n",
      "conv1d_second (Conv1D)       (None, 364, 32)           3104      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_second (MaxPoo (None, 72, 32)            0         \n",
      "_________________________________________________________________\n",
      "global_max_pooling1d (Global (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "output_conv1d (Dense)        (None, 20)                660       \n",
      "=================================================================\n",
      "Total params: 327,956\n",
      "Trainable params: 327,956\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Explanation of Output Shape and Number of Parameters \n",
    "\n",
    "```\n",
    "Layer (type)                 Output Shape              Param #   \n",
    "=================================================================\n",
    "sequence_input (InputLayer)  (None, 1100)              0       \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1100: Sequence Length\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "word_embedding (Embedding)   (None, 1100, 64)          320064      \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1100: Sequence Length\n",
    "- 64: Word Embedding Dimensions\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- (Vocabulary Size + 1)  x Word Embedding Dimensions =  $5001 \\times 64 = 320,064$  \n",
    "    - 1 is added for the 0's of the padding\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "conv1d_first (Conv1D)        (None, 1099, 32)          4128          \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1099: Sequence Length - 1. The 1 is lost due to convolution with a Filter Size = 2\n",
    "- 32: Number of Filters\n",
    "#### Number of Parameters \n",
    "\n",
    "- Word Embedding Dimensions  x Number of Filters x Filter Size + Number of Filters=  $64 \\times 32 \\times 2 + 32= 4,128$ \n",
    "    - Number of Filters is added for the bias terms\n",
    "<img src=\"images/Convnet Cube.png\" alt=\"Drawing\" style=\"width: 300px;\" />\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "max_pooling1d_first (MaxPool (None, 366, 32)           0       \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 366: (Sequence Length - 1) / Pool Size = 1099 / 3 = 366 (rounded down to)\n",
    "- 32: Number of Channels in the Previous layer, which is the Number of Filters \n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "conv1d_second (Conv1D)       (None, 364, 32)           3104          \n",
    "```\n",
    "#### Output Shape\n",
    "- None - batch_size will be determined run-time\n",
    "- 364 - 366 - 2. The 2 is lost due to convolution with a Filter Size = 3\n",
    "- 32 - Number of Filters\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- Number of Channels in the Previous layer  x Number of Filters x Filter Size + Number of Filters=  $32 \\times 32 \\times 3 + 32= 3,104$ \n",
    "    - Number of Filters is added for the bias terms\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "max_pooling1d_second (MaxPoo (None, 72, 32)            0       \n",
    "```\n",
    "#### Output Shape\n",
    "- None - batch_size will be determined run-time\n",
    "- 72: 364 / Pool Size = 364 / 5 = 72 (rounded down to)\n",
    "- 32: Number of Channels in the Previous layer, which is the Number of Filters \n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "global_max_pooling1d (Global (None, 32)                0 \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 32: Number of Channels in the Previous layer\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "output_conv1d (Dense)        (None, 20)                660   \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 20: Output Layer Size\n",
    "\n",
    "#### Number of Parameters\n",
    "- (Global Max Pooling Size + 1) x  Output Layer Size = $33 \\times 20 = 660$\n",
    "    - 1 is added for the bias term\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15016 samples, validate on 3830 samples\n",
      "Epoch 1/30\n",
      "15016/15016 [==============================] - 25s 2ms/step - loss: 2.9024 - acc: 0.1422 - val_loss: 2.5699 - val_acc: 0.3008\n",
      "Epoch 2/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 1.7743 - acc: 0.5314 - val_loss: 1.2166 - val_acc: 0.6666\n",
      "Epoch 3/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 0.8704 - acc: 0.7662 - val_loss: 0.8624 - val_acc: 0.7535\n",
      "Epoch 4/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 0.5406 - acc: 0.8552 - val_loss: 0.7415 - val_acc: 0.7843\n",
      "Epoch 5/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 0.3661 - acc: 0.9060 - val_loss: 0.7176 - val_acc: 0.7914\n",
      "Epoch 6/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 0.2484 - acc: 0.9432 - val_loss: 0.7060 - val_acc: 0.8000\n",
      "Epoch 7/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 0.1674 - acc: 0.9669 - val_loss: 0.7199 - val_acc: 0.8060\n",
      "Epoch 8/30\n",
      "15016/15016 [==============================] - 23s 2ms/step - loss: 0.1103 - acc: 0.9824 - val_loss: 0.7421 - val_acc: 0.8037\n",
      "Epoch 9/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 0.0717 - acc: 0.9921 - val_loss: 0.7674 - val_acc: 0.8057\n",
      "Epoch 10/30\n",
      "15016/15016 [==============================] - 25s 2ms/step - loss: 0.0468 - acc: 0.9960 - val_loss: 0.7855 - val_acc: 0.8078\n",
      "Epoch 11/30\n",
      "15016/15016 [==============================] - 24s 2ms/step - loss: 0.0310 - acc: 0.9981 - val_loss: 0.8099 - val_acc: 0.8086\n",
      "Wall time: 4min 24s\n"
     ]
    }
   ],
   "source": [
    "earlystop = EarlyStopping(monitor='val_loss', patience=5)\n",
    "history = model.fit(x_train, y_train, validation_data=(x_test, y_test), epochs=30, batch_size=BATCH_SIZE, callbacks=[earlystop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best validation accuracy is  0.8086161882696824\n"
     ]
    }
   ],
   "source": [
    "print ('Best validation accuracy is ', max(history.history['val_acc']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Overfitting occurs if trained too much \n",
    "\n",
    "Note, early stopping criteria kicked in at Epoch 11"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_history(history)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Perform grid search for hyperparameter optimization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 890,
   "metadata": {},
   "outputs": [],
   "source": [
    "def run_cv(input_size, embedding_dim, data, labels, number_folds, n_layers, filters, kernels, pool_sizes,\n",
    "           n_epochs = 30, patience= 5, verbose =1):\n",
    "    results = []\n",
    "    batch_size = 128\n",
    "    for x_train, y_train, x_test, y_test in create_all_folds(data, labels, number_folds):\n",
    "        model = conv1D_sequence_model_def(input_size, embedding_dim, n_layers, \n",
    "                                             filters, kernels, pool_sizes, verbose = 0)\n",
    "        earlystop = EarlyStopping(monitor='val_loss', patience=patience)\n",
    "        history = model.fit(x_train, y_train, validation_data=(x_test, y_test), epochs=n_epochs, \n",
    "                        batch_size=batch_size, verbose = 0, callbacks=[earlystop])\n",
    "        accuracy = max(history.history['val_acc'])\n",
    "        del x_train, x_test, y_train, y_test\n",
    "        #gc.collect()\n",
    "        if verbose == 1:\n",
    "            print ('Best validation accuracy is ', accuracy)\n",
    "        results.append(accuracy)\n",
    "    mean_acc = np.mean(results)\n",
    "    std_acc = np.std(results)\n",
    "    return (mean_acc, std_acc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 891,
   "metadata": {},
   "outputs": [],
   "source": [
    "def conv1D_sequence_model_def(input_size, embedding_dim, n_layers, number_filters, kernels, pool_sizes, verbose = False):\n",
    "    sequence_input = Input(shape=(input_size,), dtype='int32', name = 'sequence_input')\n",
    "    embedding_layer = Embedding(len(word_index) + 1, embedding_dim, input_length=input_size, name = 'word_embedding')\n",
    "    x = embedding_layer(sequence_input)\n",
    "\n",
    "    for i in range(n_layers):\n",
    "        x = Conv1D(filters[i], kernels[i], activation='relu', name = 'conv1d_first')(x)\n",
    "        x = MaxPooling1D(pool_sizes[i], name ='max_pooling1d_first')(x)\n",
    "    x = GlobalMaxPooling1D(name = 'global_max_pooling1d')(x)\n",
    "    preds = Dense(num_labels, activation='softmax', name = 'output_conv1d')(x)\n",
    "    model = Model(sequence_input, preds)\n",
    "    \n",
    "    if verbose: \n",
    "        model.summary()\n",
    "    model.compile(loss='categorical_crossentropy', optimizer='adam',metrics=['acc'])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "results = {'embedding_dim':[], 'mean_acc':[], 'std_acc':[]}\n",
    "filters = [32,32]\n",
    "kernels = [2,3]\n",
    "pool_sizes = [3,5]\n",
    "number_folds = 5\n",
    "n_layers = 2\n",
    "input_size = MAX_SEQUENCE_LENGTH\n",
    "embedding_dim_set = [8,16,32,64,128,256,512,1028]\n",
    "for embedding_dim in embedding_dim_set:\n",
    "    results['embedding_dim'].append(embedding_dim)\n",
    "    m, s = run_cv(input_size, embedding_dim, sequence_data, cat_data_labels, number_folds, \n",
    "                  n_layers, filters, kernels,pool_sizes, patience= 5)\n",
    "    results['mean_acc'].append(m)\n",
    "    results['std_acc'].append(s)"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "Wall time: 7h 41min 23s"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 896,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'embedding_dim': [8, 16, 32, 64, 128, 256, 512, 1028],\n",
       " 'mean_acc': [0.8439647740925136,\n",
       "  0.8648821757761999,\n",
       "  0.8738042886660653,\n",
       "  0.881992831758055,\n",
       "  0.8838656279510712,\n",
       "  0.8855289527078389,\n",
       "  0.8820515607166763,\n",
       "  0.884695699262865],\n",
       " 'std_acc': [0.004022124370563573,\n",
       "  0.003092485895523163,\n",
       "  0.006446151036361878,\n",
       "  0.008472682380729327,\n",
       "  0.004316526345908498,\n",
       "  0.008106580349402035,\n",
       "  0.007074357808043372,\n",
       "  0.004118346768984175]}"
      ]
     },
     "execution_count": 896,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"images/WORD2VEC CONVNET Sequence Accuracy.png\" alt=\"Drawing\" style=\"width: 600px;\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Increase in the vocabulary size allows for increase in the network size without deterioration of accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ConvNet in Parallel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 305,
   "metadata": {},
   "outputs": [],
   "source": [
    "EMBEDDING_DIM = 64\n",
    "filters = [100,75,50]\n",
    "kernels = [2,3,4]\n",
    "\n",
    "sequence_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32', name = 'sequence_input')\n",
    "embedding_layer = Embedding(len(word_index) + 1, EMBEDDING_DIM, input_length=MAX_SEQUENCE_LENGTH, name = 'word_embedding')\n",
    "embedded_sequences = embedding_layer(sequence_input)\n",
    "\n",
    "branch1 = Conv1D(filters[0], kernels[0], name = 'conv1d_branch1')(embedded_sequences)\n",
    "branch1 = GlobalMaxPooling1D(name = 'global_max_pooling1d_branch1')(branch1)\n",
    "\n",
    "branch2 = Conv1D(filters[1], kernels[1], name = 'conv1d_branch2')(embedded_sequences)\n",
    "branch2 = GlobalMaxPooling1D(name = 'global_max_pooling1d_branch2')(branch2)\n",
    "\n",
    "branch3 = Conv1D(filters[2], kernels[2], name = 'conv1d_branch3')(embedded_sequences)\n",
    "branch3 = GlobalMaxPooling1D(name = 'global_max_pooling1d_branch3')(branch3)\n",
    "\n",
    "x = concatenate([branch1, branch2, branch3], name = 'concatenate')\n",
    "x = Activation('relu', name='activation')(x)\n",
    "preds = Dense(num_labels, activation='softmax', name = 'output_conv1d')(x)\n",
    "\n",
    "model = Model(sequence_input, preds)\n",
    "\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['acc'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 306,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "sequence_input (InputLayer)     (None, 1100)         0                                            \n",
      "__________________________________________________________________________________________________\n",
      "word_embedding (Embedding)      (None, 1100, 64)     320064      sequence_input[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_branch1 (Conv1D)         (None, 1099, 100)    12900       word_embedding[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_branch2 (Conv1D)         (None, 1098, 75)     14475       word_embedding[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_branch3 (Conv1D)         (None, 1097, 50)     12850       word_embedding[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "global_max_pooling1d_branch1 (G (None, 100)          0           conv1d_branch1[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "global_max_pooling1d_branch2 (G (None, 75)           0           conv1d_branch2[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "global_max_pooling1d_branch3 (G (None, 50)           0           conv1d_branch3[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "concatenate (Concatenate)       (None, 225)          0           global_max_pooling1d_branch1[0][0\n",
      "                                                                 global_max_pooling1d_branch2[0][0\n",
      "                                                                 global_max_pooling1d_branch3[0][0\n",
      "__________________________________________________________________________________________________\n",
      "activation (Activation)         (None, 225)          0           concatenate[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "output_conv1d (Dense)           (None, 20)           4520        activation[0][0]                 \n",
      "==================================================================================================\n",
      "Total params: 364,809\n",
      "Trainable params: 364,809\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Explanation of Output Shape and Number of Parameters \n",
    "\n",
    "```\n",
    "Layer (type)                 Output Shape              Param #   \n",
    "=================================================================\n",
    "sequence_input (InputLayer)  (None, 1100)              0       \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1100: Sequence Length\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "word_embedding (Embedding)   (None, 1100, 64)          320064      \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1100: Sequence Length\n",
    "- 64: Word Embedding Dimensions\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- (Vocabulary Size + 1)  x Word Embedding Dimensions =  $5001 \\times 64 = 320,064$ \n",
    "    - 1 is added for the unknown token, i.e. all the out-of-vocabulary words\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "conv1d_branch1 (Conv1D)        (None, 1099, 100)          12900          \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1099: Sequence Length - 1. The 1 is lost due to convolution with a Filter Size = 2\n",
    "- 100: Number of Filters\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- Word Embedding Dimensions  x Number of Filters x Filter Size + Number of Filters=$64 \\times 100 \\times 2 + 100 = 12,900$ \n",
    "    - Number of Filters is added for the bias terms\n",
    "<img src=\"images/Convnet Cube.png\" alt=\"Drawing\" style=\"width: 300px;\" />\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "conv1d_branch2 (Conv1D)        (None, 1098, 75)          14475          \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1098: Sequence Length - 2. The 1 is lost due to convolution with a Filter Size = 3\n",
    "- 75: Number of Filters\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- Word Embedding Dimensions  x Number of Filters x Filter Size + Number of Filters= $64 \\times 75 \\times 3 + 75 = 14,475$ \n",
    "    - Number of Filters is added for the bias terms\n",
    "\n",
    "```\n",
    "_________________________________________________________________\n",
    "conv1d_branch3 (Conv1D)        (None, 1097, 50)          12850          \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 1097: Sequence Length - 3. The 1 is lost due to convolution with a Filter Size = 4\n",
    "- 50: Number of Filters\n",
    "\n",
    "#### Number of Parameters \n",
    "\n",
    "- Word Embedding Dimensions  x Number of Filters x Filter Size + Number of Filters= $64 \\times 50 \\times 4 + 50 = 12,850$ \n",
    "    - Number of Filters is added for the bias terms\n",
    "    \n",
    "```\n",
    "_________________________________________________________________\n",
    "concatenate (Concatenate)       (None, 225)          0 \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 225: Sum of the Number of Channels in the Concatenated layers $100 + 75 + 50 = 225$\n",
    "\n",
    "\n",
    "```\n",
    "activation (Activation)         (None, 225)          0                          \n",
    "__________________________________________________________________________________________________\n",
    "output_conv1d (Dense)           (None, 20)           4520 \n",
    "```\n",
    "#### Output Shape\n",
    "- None: batch_size will be determined run-time\n",
    "- 20: Output Layer Size\n",
    "\n",
    "#### Number of Parameters\n",
    "- (Concatenated Size + 1) x  Output Layer Size = $226 \\times 20 = 4520$\n",
    "    - 1 is added for the bias term\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 307,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 15105 samples, validate on 3741 samples\n",
      "Epoch 1/50\n",
      "15105/15105 [==============================] - 144s 10ms/step - loss: 2.7102 - acc: 0.3567 - val_loss: 1.9352 - val_acc: 0.6362\n",
      "Epoch 2/50\n",
      "15105/15105 [==============================] - 146s 10ms/step - loss: 1.2608 - acc: 0.7186 - val_loss: 0.8936 - val_acc: 0.7632\n",
      "Epoch 3/50\n",
      "15105/15105 [==============================] - 141s 9ms/step - loss: 0.6805 - acc: 0.8173 - val_loss: 0.6419 - val_acc: 0.8161\n",
      "Epoch 4/50\n",
      "15105/15105 [==============================] - 141s 9ms/step - loss: 0.4518 - acc: 0.8783 - val_loss: 0.5347 - val_acc: 0.8455\n",
      "Epoch 5/50\n",
      "15105/15105 [==============================] - 145s 10ms/step - loss: 0.3146 - acc: 0.9210 - val_loss: 0.4775 - val_acc: 0.8602\n",
      "Epoch 6/50\n",
      "15105/15105 [==============================] - 151s 10ms/step - loss: 0.2250 - acc: 0.9480 - val_loss: 0.4472 - val_acc: 0.8690\n",
      "Epoch 7/50\n",
      "15105/15105 [==============================] - 144s 10ms/step - loss: 0.1498 - acc: 0.9721 - val_loss: 0.4245 - val_acc: 0.8762\n",
      "Epoch 8/50\n",
      "15105/15105 [==============================] - 142s 9ms/step - loss: 0.1021 - acc: 0.9868 - val_loss: 0.4140 - val_acc: 0.8802\n",
      "Epoch 9/50\n",
      "15105/15105 [==============================] - 140s 9ms/step - loss: 0.0686 - acc: 0.9939 - val_loss: 0.4125 - val_acc: 0.8792\n",
      "Epoch 10/50\n",
      "15105/15105 [==============================] - 140s 9ms/step - loss: 0.0469 - acc: 0.9970 - val_loss: 0.4109 - val_acc: 0.8810\n",
      "Epoch 11/50\n",
      "15105/15105 [==============================] - 143s 9ms/step - loss: 0.0337 - acc: 0.9987 - val_loss: 0.4120 - val_acc: 0.8837\n",
      "Epoch 12/50\n",
      "15105/15105 [==============================] - 143s 9ms/step - loss: 0.0244 - acc: 0.9992 - val_loss: 0.4147 - val_acc: 0.8837\n",
      "Epoch 13/50\n",
      "15105/15105 [==============================] - 140s 9ms/step - loss: 0.0185 - acc: 0.9996 - val_loss: 0.4180 - val_acc: 0.8853\n",
      "Epoch 14/50\n",
      "15105/15105 [==============================] - 141s 9ms/step - loss: 0.0149 - acc: 0.9993 - val_loss: 0.4241 - val_acc: 0.8859\n",
      "Epoch 15/50\n",
      "15105/15105 [==============================] - 142s 9ms/step - loss: 0.0121 - acc: 0.9993 - val_loss: 0.4268 - val_acc: 0.8861\n",
      "Wall time: 35min 46s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "earlystop = EarlyStopping(monitor='val_loss', patience=5)\n",
    "history = model.fit(x_train, y_train, validation_data=(x_test, y_test), epochs=50, batch_size=BATCH_SIZE, callbacks=[earlystop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 308,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best validation accuracy is  0.8861267039464202\n"
     ]
    }
   ],
   "source": [
    "print ('Best validation accuracy is ', max(history.history['val_acc']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Overfitting occurs if trained too much \n",
    "\n",
    "Note, early stopping criteria kicked in at Epoch 15"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 302,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_history(history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Perform grid search for hyperparameter optimization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "def run_cv_parallel(input_size, embedding_dim, data, labels, number_folds, n_branches, filters, kernels, \n",
    "           n_epochs = 30, patience= 5, verbose =1):\n",
    "    results = []\n",
    "    batch_size = 128\n",
    "    for x_train, y_train, x_test, y_test in create_all_folds(data, labels, number_folds):\n",
    "        model = conv1D_parallel_model_def(input_size, embedding_dim, n_branches, \n",
    "                                             filters, kernels,  verbose = 0)\n",
    "        earlystop = EarlyStopping(monitor='val_loss', patience=patience)\n",
    "        history = model.fit(x_train, y_train, validation_data=(x_test, y_test), epochs=n_epochs, \n",
    "                        batch_size=batch_size, verbose = 0, callbacks=[earlystop])\n",
    "        accuracy = max(history.history['val_acc'])\n",
    "        del x_train, x_test, y_train, y_test\n",
    "        #gc.collect()\n",
    "        if verbose == 1:\n",
    "            print ('Best validation accuracy is ', accuracy)\n",
    "        results.append(accuracy)\n",
    "    mean_acc = np.mean(results)\n",
    "    std_acc = np.std(results)\n",
    "    return (mean_acc, std_acc)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "def conv1D_parallel_model_def(input_size, embedding_dim, n_branches, number_filters, kernels, verbose = False):\n",
    "    sequence_input = Input(shape=(input_size,), dtype='int32', name = 'sequence_input')\n",
    "    embedding_layer = Embedding(len(word_index) + 1, embedding_dim, input_length=input_size, name = 'word_embedding')\n",
    "    embedded_sequences = embedding_layer(sequence_input)\n",
    "\n",
    "    branches = []\n",
    "    for i in range(n_branches):\n",
    "        y = Conv1D(filters[i], kernels[i], activation='relu')(embedded_sequences)\n",
    "        y = GlobalMaxPooling1D()(y)\n",
    "        branches.append(y)\n",
    "    x = concatenate(branches)\n",
    "    x = Activation('relu')(x)\n",
    "    preds = Dense(num_labels, activation='softmax')(x)\n",
    "    model = Model(sequence_input, preds)\n",
    "    \n",
    "    if verbose: \n",
    "        model.summary()\n",
    "    model.compile(loss='categorical_crossentropy', optimizer='adam',metrics=['acc'])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "results = {'embedding_dim':[], 'vocab_size':[],'mean_acc':[], 'std_acc':[]}\n",
    "filters = [100,75,50]\n",
    "kernels = [2,3,4]\n",
    "number_folds = 5\n",
    "n_branches = 3\n",
    "embedding_dim_set = [8,16,32,64,128,256,512,1028]\n",
    "for vocab_size in [5000,15000]:\n",
    "    seq_data, word_index = make_sequences(data, vocab_size, percentile=99)\n",
    "    input_size = seq_data.shape[1]\n",
    "    for embedding_dim in embedding_dim_set:\n",
    "        results['embedding_dim'].append(embedding_dim)\n",
    "        results['vocab_size'].append(vocab_size)\n",
    "        m, s = run_cv_parallel(input_size, embedding_dim, seq_data, cat_data_labels, number_folds, \n",
    "                      n_branches, filters, kernels, patience= 5)\n",
    "        results['mean_acc'].append(m)\n",
    "        results['std_acc'].append(s)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It took almost 6 days to run the grid search for vocab_size = 1500:\n",
    "\n",
    "CPU times: user 10d 7h 47min 55s, sys: 16h 40min 37s, total: 11d 28min 33s\n",
    "\n",
    "Wall time: 5d 15h 24min 19s"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"images/WORD2VEC CONVNET Parallel Accuracy.png?modified=12345678\" alt=\"Drawing\" style=\"width: 600px;\" />"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Translational Invariance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "964"
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "word_index['cpu']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"images/Translational Invariance.png\" alt=\"Drawing\" style=\"width: 600px;\" />"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "sequence = np.zeros((1,1100))\n",
    "sequence[0][0] = 964\n",
    "#a sequence with the word 'cpu' in the beginning and the rest are 0's\n",
    "predict_0 = model.predict(sequence)[0]\n",
    "sequence = np.zeros((1,1100))\n",
    "sequence[0][1] = 964\n",
    "#a sequence with the word 'cpu' in the second position and the rest are 0's\n",
    "predict_1 = model.predict(sequence)[0]\n",
    "sequence = np.zeros((1,1100))\n",
    "sequence[0][10] = 964\n",
    "#a sequence with the word 'cpu' in the 11's position and the rest are 0's\n",
    "predict_10 = model.predict(sequence)[0]\n",
    "sequence = np.zeros((1,1100))\n",
    "sequence[0][100] = 964\n",
    "#a sequence with the word 'cpu' in the 101's position and the rest are 0's\n",
    "predict_100 = model.predict(sequence)[0]\n",
    "sequence = np.zeros((1,1100))\n",
    "sequence[0][1099] = 964\n",
    "#a sequence with the word 'cpu' in the last position and the rest are 0's\n",
    "predict_last = model.predict(sequence)[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### predict_0, etc. are predicted probabilities of the 20 topics, given the word sequence, which is really just the word 'cpu' at different positions in the sequence of 0's"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 259,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(15,10))\n",
    "plt.plot(predict_0, label='predict_0')\n",
    "plt.plot(predict_1, label='predict_1', linewidth=7.0)\n",
    "plt.plot(predict_10,  label='predict_10', linewidth=7.0)\n",
    "plt.plot(predict_100, label='predict_100')\n",
    "plt.plot(predict_last, label='predict_last')\n",
    "plt.xticks(np.arange(num_labels))\n",
    "ax = plt.axes()\n",
    "\n",
    "ax.set_xticklabels(np.array(data_label_names), rotation=90)\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Try different words to see which topics they predict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 210,
   "metadata": {},
   "outputs": [],
   "source": [
    "pitching_index = word_index['pitching']\n",
    "baseball_index = word_index['baseball']\n",
    "christians_index = word_index['christians']\n",
    "cpu_index = word_index['cpu']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 232,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'comp.sys.ibm.pc.hardware'"
      ]
     },
     "execution_count": 232,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#try to comment some of the lines\n",
    "sequence = np.zeros((1,1100))\n",
    "sequence[0][410] = pitching_index \n",
    "sequence[0][511] = christians_index \n",
    "sentsequenceence[0][512] = cpu_index \n",
    "sequence[0][513] = cpu_index \n",
    "data_label_names[np.argmax(model.predict(sequence))]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## The word embeddings have learned some semantic similarity even though the only goal was text classification"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Copy the word embeddings into a gensim's Word2Vec model to find similar words efficiently"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 135,
   "metadata": {},
   "outputs": [],
   "source": [
    "embedding_layer_weights =  model.layers[1].get_weights()[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_tokens = [x.split() for x in data]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 137,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_w2v = models.Word2Vec(sg = 1, size=64) #sg=1 skip-a-gram\n",
    "model_w2v.build_vocab(all_tokens)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 138,
   "metadata": {},
   "outputs": [],
   "source": [
    "#to change the vectors without creating a new model\n",
    "#have to reset the vectors cashed by most_similar\n",
    "model_w2v.wv.vectors_norm = None"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_w2v.wv.vectors  = embedding_layer_weights[2:,:] # the first two are 0 padding and UNK"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Find similar words"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('catholic', 0.8506920337677002),\n",
       " ('luke', 0.7527848482131958),\n",
       " ('founded', 0.7421770095825195),\n",
       " ('ye', 0.678794264793396),\n",
       " ('commonly', 0.6145917773246765),\n",
       " ('ups', 0.5935655832290649),\n",
       " ('living', 0.5869908928871155),\n",
       " ('century', 0.5866395235061646),\n",
       " ('koresh', 0.563356876373291),\n",
       " ('bos', 0.5430530309677124)]"
      ]
     },
     "execution_count": 140,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_w2v.wv.most_similar(positive=['christians'], topn=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('config', 0.6700758337974548),\n",
       " ('busy', 0.6633105874061584),\n",
       " ('modem', 0.6468589901924133),\n",
       " ('pcs', 0.6393556594848633),\n",
       " ('linear', 0.6344971060752869),\n",
       " ('programmer', 0.6251541376113892),\n",
       " ('memory', 0.6202409267425537),\n",
       " ('cat', 0.6051343679428101),\n",
       " ('computers', 0.604193925857544),\n",
       " ('installed', 0.598760724067688)]"
      ]
     },
     "execution_count": 141,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_w2v.wv.most_similar(positive=['cpu'], topn=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('defensive', 0.7596086859703064),\n",
       " ('yankees', 0.753547191619873),\n",
       " ('league', 0.7222006320953369),\n",
       " ('pitching', 0.6910199522972107),\n",
       " ('ted', 0.689267098903656),\n",
       " ('agent', 0.6745107769966125),\n",
       " ('asd', 0.6468268632888794),\n",
       " ('batting', 0.6451112031936646),\n",
       " ('hit', 0.6194806098937988),\n",
       " ('career', 0.6016578674316406)]"
      ]
     },
     "execution_count": 142,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model_w2v.wv.most_similar(positive=['baseball'], topn=10) "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}