quora-nlp-text-classification-DL.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['sample_submission.csv', 'train.csv', 'test.csv', 'embeddings']\n"
     ]
    }
   ],
   "source": [
    "# Import Libraries\n",
    "import numpy as np # linear algebra\n",
    "import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n",
    "import time\n",
    "from tqdm import tqdm\n",
    "import math\n",
    "import gc\n",
    "\n",
    "import os\n",
    "print(os.listdir(\"../input\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [],
   "source": [
    "# Get automatic reloading and inline plotting\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "_uuid": "c007da7203d55256e8ba3e713e49f8401caac746"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "True\n",
      "True\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "### Checking GPU Set up\n",
    "print(torch.cuda.is_available())\n",
    "print(torch.backends.cudnn.enabled)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "_uuid": "634e83d34a09d7ad24178adb2b18a10479adffa6"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train shape :  (1306122, 3)\n",
      "Test shape :  (375806, 2)\n"
     ]
    }
   ],
   "source": [
    "train_df = pd.read_csv(\"../input/train.csv\")\n",
    "test_df = pd.read_csv(\"../input/test.csv\")\n",
    "print(\"Train shape : \",train_df.shape)\n",
    "print(\"Test shape : \",test_df.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "_uuid": "31c1333133a5dd7309facdb29305bc8e99472f05"
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "train_df, val_df = train_test_split(train_df, test_size=0.1, random_state=2018)\n",
    "## Get the target values\n",
    "train_y = train_df['target'].values\n",
    "val_y = val_df['target'].values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "_uuid": "8a79914305abf2aa86734ea132a91536379bf84c"
   },
   "outputs": [],
   "source": [
    "def null_treatment(df):\n",
    "    return df[\"question_text\"].fillna(\"__NA__\").values\n",
    "\n",
    "train_texts = null_treatment(train_df)\n",
    "val_texts   = null_treatment(val_df)\n",
    "test_texts  = null_treatment(test_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "_uuid": "539a481554cc76aee3f4e89d186af436336e9afc"
   },
   "outputs": [],
   "source": [
    "del train_df\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "_uuid": "cf9c4e44230e4d395830f231f55f6120315d64f2"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.preprocessing.text import Tokenizer\n",
    "max_features = 95000\n",
    "## Tokenize the sentences\n",
    "tokenizer = Tokenizer(num_words=max_features)\n",
    "tokenizer.fit_on_texts(np.concatenate([train_texts, val_texts, test_texts]))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "_uuid": "61d9f4a7554829d4e6a283266e44415d8f3749f4"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.preprocessing import sequence\n",
    "train_sequences = tokenizer.texts_to_sequences(train_texts)\n",
    "valid_sequences = tokenizer.texts_to_sequences(val_texts)\n",
    "test_sequences = tokenizer.texts_to_sequences(test_texts)\n",
    "\n",
    "del train_texts, val_texts, test_texts\n",
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "_uuid": "ed054b712a259eb31f010bafb027bfdc567986d9"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "max_len = 100\n",
    "train_X = sequence.pad_sequences(train_sequences, maxlen = max_len)\n",
    "val_X = sequence.pad_sequences(valid_sequences, maxlen = max_len)\n",
    "test_X = sequence.pad_sequences(test_sequences, maxlen = max_len)\n",
    "\n",
    "del train_sequences, valid_sequences, test_sequences\n",
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "_uuid": "4cc911f8e5b8fba8bf0608f66d11c6b67d1d919c"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.6/site-packages/ipykernel_launcher.py:5: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n",
      "  \"\"\"\n",
      "/opt/conda/lib/python3.6/site-packages/ipykernel_launcher.py:25: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "EMBEDDING_FILE = '../input/embeddings/glove.840B.300d/glove.840B.300d.txt'\n",
    "def get_coefs(word,*arr): return word, np.asarray(arr, dtype='float32')\n",
    "embeddings_index = dict(get_coefs(*o.split(\" \")) for o in open(EMBEDDING_FILE))\n",
    "\n",
    "all_embs = np.stack(embeddings_index.values())\n",
    "emb_mean,emb_std = all_embs.mean(), all_embs.std()\n",
    "embed_size = all_embs.shape[1]\n",
    "\n",
    "word_index = tokenizer.word_index\n",
    "nb_words = min(max_features, len(word_index))\n",
    "embedding_matrix = np.random.normal(emb_mean, emb_std, (nb_words, embed_size))\n",
    "for word, i in word_index.items():\n",
    "    if i >= max_features: continue\n",
    "    embedding_vector = embeddings_index.get(word)\n",
    "    if embedding_vector is not None: embedding_matrix[i] = embedding_vector\n",
    "\n",
    "embedding_matrix_1 = embedding_matrix\n",
    "del embeddings_index , word_index, all_embs,embedding_matrix\n",
    "gc.collect()\n",
    "\n",
    "EMBEDDING_FILE = '../input/embeddings/paragram_300_sl999/paragram_300_sl999.txt'\n",
    "def get_coefs(word,*arr): return word, np.asarray(arr, dtype='float32')\n",
    "embeddings_index = dict(get_coefs(*o.split(\" \")) for o in open(EMBEDDING_FILE, encoding=\"utf8\", errors='ignore') if len(o)>100)\n",
    "\n",
    "all_embs = np.stack(embeddings_index.values())\n",
    "emb_mean,emb_std = all_embs.mean(), all_embs.std()\n",
    "embed_size = all_embs.shape[1]\n",
    "word_index = tokenizer.word_index\n",
    "nb_words = min(max_features, len(word_index))\n",
    "embedding_matrix = np.random.normal(emb_mean, emb_std, (nb_words, embed_size))\n",
    "for word, i in word_index.items():\n",
    "    if i >= max_features: continue\n",
    "    embedding_vector = embeddings_index.get(word)\n",
    "    if embedding_vector is not None: embedding_matrix[i] = embedding_vector\n",
    "\n",
    "\n",
    "embedding_matrix = np.mean((embedding_matrix, embedding_matrix_1), axis=0)\n",
    "del embedding_matrix_1,embeddings_index , word_index, all_embs\n",
    "gc.collect()\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "_uuid": "3f6e836999ebb51cccf5fcc4119faa7a704d9adb"
   },
   "outputs": [],
   "source": [
    "import keras\n",
    "\n",
    "class SWA(keras.callbacks.Callback):\n",
    "    \n",
    "    def __init__(self, filepath, swa_epoch):\n",
    "        super(SWA, self).__init__()\n",
    "        self.filepath = filepath\n",
    "        self.swa_epoch = swa_epoch \n",
    "    \n",
    "    def on_train_begin(self, logs=None):\n",
    "        self.nb_epoch = self.params['epochs']\n",
    "        print('Stochastic weight averaging selected for last {} epochs.'\n",
    "              .format(self.nb_epoch - self.swa_epoch))\n",
    "        \n",
    "    def on_epoch_end(self, epoch, logs=None):\n",
    "        \n",
    "        if epoch == self.swa_epoch:\n",
    "            self.swa_weights = self.model.get_weights()\n",
    "            \n",
    "        elif epoch > self.swa_epoch:    \n",
    "            for i, layer in enumerate(self.model.layers):\n",
    "                self.swa_weights[i] = (self.swa_weights[i] * \n",
    "                    (epoch - self.swa_epoch) + self.model.get_weights()[i])/((epoch - self.swa_epoch)  + 1)  \n",
    "\n",
    "        else:\n",
    "            pass\n",
    "        \n",
    "    def on_train_end(self, logs=None):\n",
    "        self.model.set_weights(self.swa_weights)\n",
    "        print('Final model parameters set to stochastic weight average.')\n",
    "        self.model.save_weights(self.filepath)\n",
    "        print('Final stochastic averaged weights saved to file.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "_uuid": "53ba0eb1d395f5b07c81e4a1a2f3be2d790728a2"
   },
   "outputs": [],
   "source": [
    "from keras import backend as K\n",
    "from keras.engine.topology import Layer\n",
    "from keras import initializers, regularizers, constraints\n",
    "\n",
    "\n",
    "class Attention(Layer):\n",
    "    def __init__(self, step_dim,\n",
    "                 W_regularizer=None, b_regularizer=None,\n",
    "                 W_constraint=None, b_constraint=None,\n",
    "                 bias=True, **kwargs):\n",
    "        self.supports_masking = True\n",
    "        self.init = initializers.get('glorot_uniform')\n",
    "\n",
    "        self.W_regularizer = regularizers.get(W_regularizer)\n",
    "        self.b_regularizer = regularizers.get(b_regularizer)\n",
    "\n",
    "        self.W_constraint = constraints.get(W_constraint)\n",
    "        self.b_constraint = constraints.get(b_constraint)\n",
    "\n",
    "        self.bias = bias\n",
    "        self.step_dim = step_dim\n",
    "        self.features_dim = 0\n",
    "        super(Attention, self).__init__(**kwargs)\n",
    "\n",
    "    def build(self, input_shape):\n",
    "        assert len(input_shape) == 3\n",
    "\n",
    "        self.W = self.add_weight((input_shape[-1],),\n",
    "                                 initializer=self.init,\n",
    "                                 name='{}_W'.format(self.name),\n",
    "                                 regularizer=self.W_regularizer,\n",
    "                                 constraint=self.W_constraint)\n",
    "        self.features_dim = input_shape[-1]\n",
    "\n",
    "        if self.bias:\n",
    "            self.b = self.add_weight((input_shape[1],),\n",
    "                                     initializer='zero',\n",
    "                                     name='{}_b'.format(self.name),\n",
    "                                     regularizer=self.b_regularizer,\n",
    "                                     constraint=self.b_constraint)\n",
    "        else:\n",
    "            self.b = None\n",
    "\n",
    "        self.built = True\n",
    "\n",
    "    def compute_mask(self, input, input_mask=None):\n",
    "        return None\n",
    "\n",
    "    def call(self, x, mask=None):\n",
    "        features_dim = self.features_dim\n",
    "        step_dim = self.step_dim\n",
    "\n",
    "        eij = K.reshape(K.dot(K.reshape(x, (-1, features_dim)),\n",
    "                        K.reshape(self.W, (features_dim, 1))), (-1, step_dim))\n",
    "\n",
    "        if self.bias:\n",
    "            eij += self.b\n",
    "\n",
    "        eij = K.tanh(eij)\n",
    "\n",
    "        a = K.exp(eij)\n",
    "\n",
    "        if mask is not None:\n",
    "            a *= K.cast(mask, K.floatx())\n",
    "\n",
    "        a /= K.cast(K.sum(a, axis=1, keepdims=True) + K.epsilon(), K.floatx())\n",
    "\n",
    "        a = K.expand_dims(a)\n",
    "        weighted_input = x * a\n",
    "        return K.sum(weighted_input, axis=1)\n",
    "\n",
    "    def compute_output_shape(self, input_shape):\n",
    "        return input_shape[0],  self.features_dim"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "_uuid": "4e8efa2cf7821d41abd9c056038c56017464bc69"
   },
   "outputs": [],
   "source": [
    "from keras import backend as K\n",
    "from keras.callbacks import *\n",
    "\n",
    "class CyclicLR(Callback):\n",
    "    \"\"\"This callback implements a cyclical learning rate policy (CLR).\n",
    "    The method cycles the learning rate between two boundaries with\n",
    "    some constant frequency, as detailed in this paper (https://arxiv.org/abs/1506.01186).\n",
    "    The amplitude of the cycle can be scaled on a per-iteration or \n",
    "    per-cycle basis.\n",
    "    This class has three built-in policies, as put forth in the paper.\n",
    "    \"triangular\":\n",
    "        A basic triangular cycle w/ no amplitude scaling.\n",
    "    \"triangular2\":\n",
    "        A basic triangular cycle that scales initial amplitude by half each cycle.\n",
    "    \"exp_range\":\n",
    "        A cycle that scales initial amplitude by gamma**(cycle iterations) at each \n",
    "        cycle iteration.\n",
    "    For more detail, please see paper.\n",
    "    \n",
    "    # Example\n",
    "        ```python\n",
    "            clr = CyclicLR(base_lr=0.001, max_lr=0.006,\n",
    "                                step_size=2000., mode='triangular')\n",
    "            model.fit(X_train, Y_train, callbacks=[clr])\n",
    "        ```\n",
    "    \n",
    "    Class also supports custom scaling functions:\n",
    "        ```python\n",
    "            clr_fn = lambda x: 0.5*(1+np.sin(x*np.pi/2.))\n",
    "            clr = CyclicLR(base_lr=0.001, max_lr=0.006,\n",
    "                                step_size=2000., scale_fn=clr_fn,\n",
    "                                scale_mode='cycle')\n",
    "            model.fit(X_train, Y_train, callbacks=[clr])\n",
    "        ```    \n",
    "    # Arguments\n",
    "        base_lr: initial learning rate which is the\n",
    "            lower boundary in the cycle.\n",
    "        max_lr: upper boundary in the cycle. Functionally,\n",
    "            it defines the cycle amplitude (max_lr - base_lr).\n",
    "            The lr at any cycle is the sum of base_lr\n",
    "            and some scaling of the amplitude; therefore \n",
    "            max_lr may not actually be reached depending on\n",
    "            scaling function.\n",
    "        step_size: number of training iterations per\n",
    "            half cycle. Authors suggest setting step_size\n",
    "            2-8 x training iterations in epoch.\n",
    "        mode: one of {triangular, triangular2, exp_range}.\n",
    "            Default 'triangular'.\n",
    "            Values correspond to policies detailed above.\n",
    "            If scale_fn is not None, this argument is ignored.\n",
    "        gamma: constant in 'exp_range' scaling function:\n",
    "            gamma**(cycle iterations)\n",
    "        scale_fn: Custom scaling policy defined by a single\n",
    "            argument lambda function, where \n",
    "            0 <= scale_fn(x) <= 1 for all x >= 0.\n",
    "            mode paramater is ignored \n",
    "        scale_mode: {'cycle', 'iterations'}.\n",
    "            Defines whether scale_fn is evaluated on \n",
    "            cycle number or cycle iterations (training\n",
    "            iterations since start of cycle). Default is 'cycle'.\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(self, base_lr=0.001, max_lr=0.006, step_size=2000., mode='triangular',\n",
    "                 gamma=1., scale_fn=None, scale_mode='cycle'):\n",
    "        super(CyclicLR, self).__init__()\n",
    "\n",
    "        self.base_lr = base_lr\n",
    "        self.max_lr = max_lr\n",
    "        self.step_size = step_size\n",
    "        self.mode = mode\n",
    "        self.gamma = gamma\n",
    "        if scale_fn == None:\n",
    "            if self.mode == 'triangular':\n",
    "                self.scale_fn = lambda x: 1.\n",
    "                self.scale_mode = 'cycle'\n",
    "            elif self.mode == 'triangular2':\n",
    "                self.scale_fn = lambda x: 1/(2.**(x-1))\n",
    "                self.scale_mode = 'cycle'\n",
    "            elif self.mode == 'exp_range':\n",
    "                self.scale_fn = lambda x: gamma**(x)\n",
    "                self.scale_mode = 'iterations'\n",
    "        else:\n",
    "            self.scale_fn = scale_fn\n",
    "            self.scale_mode = scale_mode\n",
    "        self.clr_iterations = 0.\n",
    "        self.trn_iterations = 0.\n",
    "        self.history = {}\n",
    "\n",
    "        self._reset()\n",
    "\n",
    "    def _reset(self, new_base_lr=None, new_max_lr=None,\n",
    "               new_step_size=None):\n",
    "        \"\"\"Resets cycle iterations.\n",
    "        Optional boundary/step size adjustment.\n",
    "        \"\"\"\n",
    "        if new_base_lr != None:\n",
    "            self.base_lr = new_base_lr\n",
    "        if new_max_lr != None:\n",
    "            self.max_lr = new_max_lr\n",
    "        if new_step_size != None:\n",
    "            self.step_size = new_step_size\n",
    "        self.clr_iterations = 0.\n",
    "        \n",
    "    def clr(self):\n",
    "        cycle = np.floor(1+self.clr_iterations/(2*self.step_size))\n",
    "        x = np.abs(self.clr_iterations/self.step_size - 2*cycle + 1)\n",
    "        if self.scale_mode == 'cycle':\n",
    "            return self.base_lr + (self.max_lr-self.base_lr)*np.maximum(0, (1-x))*self.scale_fn(cycle)\n",
    "        else:\n",
    "            return self.base_lr + (self.max_lr-self.base_lr)*np.maximum(0, (1-x))*self.scale_fn(self.clr_iterations)\n",
    "        \n",
    "    def on_train_begin(self, logs={}):\n",
    "        logs = logs or {}\n",
    "\n",
    "        if self.clr_iterations == 0:\n",
    "            K.set_value(self.model.optimizer.lr, self.base_lr)\n",
    "        else:\n",
    "            K.set_value(self.model.optimizer.lr, self.clr())        \n",
    "            \n",
    "    def on_batch_end(self, epoch, logs=None):\n",
    "        \n",
    "        logs = logs or {}\n",
    "        self.trn_iterations += 1\n",
    "        self.clr_iterations += 1\n",
    "\n",
    "        self.history.setdefault('lr', []).append(K.get_value(self.model.optimizer.lr))\n",
    "        self.history.setdefault('iterations', []).append(self.trn_iterations)\n",
    "\n",
    "        for k, v in logs.items():\n",
    "            self.history.setdefault(k, []).append(v)\n",
    "        \n",
    "        K.set_value(self.model.optimizer.lr, self.clr())\n",
    "    \n",
    "\n",
    "def f1_calc(y_true, y_pred):\n",
    "    '''\n",
    "    metric from here \n",
    "    https://stackoverflow.com/questions/43547402/how-to-calculate-f1-macro-in-keras\n",
    "    '''\n",
    "    def recall(y_true, y_pred):\n",
    "        \"\"\"Recall metric.\n",
    "\n",
    "        Only computes a batch-wise average of recall.\n",
    "\n",
    "        Computes the recall, a metric for multi-label classification of\n",
    "        how many relevant items are selected.\n",
    "        \"\"\"\n",
    "        true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "        possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))\n",
    "        recall = true_positives / (possible_positives + K.epsilon())\n",
    "        return recall\n",
    "\n",
    "    def precision(y_true, y_pred):\n",
    "        \"\"\"Precision metric.\n",
    "\n",
    "        Only computes a batch-wise average of precision.\n",
    "\n",
    "        Computes the precision, a metric for multi-label classification of\n",
    "        how many selected items are relevant.\n",
    "        \"\"\"\n",
    "        true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "        predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))\n",
    "        precision = true_positives / (predicted_positives + K.epsilon())\n",
    "        return precision\n",
    "    precision = precision(y_true, y_pred)\n",
    "    recall = recall(y_true, y_pred)\n",
    "    return 2*((precision*recall)/(precision+recall+K.epsilon()))\n",
    " \n",
    "\n",
    "clr = CyclicLR(base_lr=0.001, max_lr=0.007,\n",
    "                        step_size=300., mode='exp_range',\n",
    "                        gamma=0.99994)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "_uuid": "13dc7bf9d2fccd63687a18dbc753676aed0909e7"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 100)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "embedding_1 (Embedding)         (None, 100, 300)     28500000    input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "spatial_dropout1d_1 (SpatialDro (None, 100, 300)     0           embedding_1[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_1 (Bidirectional) (None, 100, 80)      109440      spatial_dropout1d_1[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_2 (Bidirectional) (None, 100, 80)      39040       bidirectional_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "attention_1 (Attention)         (None, 80)           180         bidirectional_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "attention_2 (Attention)         (None, 80)           180         bidirectional_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "global_average_pooling1d_1 (Glo (None, 80)           0           bidirectional_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "global_max_pooling1d_1 (GlobalM (None, 80)           0           bidirectional_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 320)          0           attention_1[0][0]                \n",
      "                                                                 attention_2[0][0]                \n",
      "                                                                 global_average_pooling1d_1[0][0] \n",
      "                                                                 global_max_pooling1d_1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 32)           10272       concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_1 (BatchNor (None, 32)           128         dense_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)             (None, 32)           0           batch_normalization_1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "dense_2 (Dense)                 (None, 1)            33          dropout_1[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 28,659,273\n",
      "Trainable params: 159,209\n",
      "Non-trainable params: 28,500,064\n",
      "__________________________________________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "from keras.layers import Dense, Input, LSTM, Embedding, Dropout, Activation, CuDNNLSTM,CuDNNGRU, Conv1D\n",
    "from keras.layers import Bidirectional, GlobalMaxPool1D\n",
    "from keras.models import Model\n",
    "from keras import initializers, regularizers, constraints, optimizers, layers\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.layers import SpatialDropout1D,GlobalAveragePooling1D,GlobalMaxPooling1D,concatenate\n",
    "\n",
    "maxlen = max_len\n",
    "inp = Input(shape=(max_len,))\n",
    "x = Embedding(max_features, embedding_matrix.shape[1], weights=[embedding_matrix], trainable=False)(inp)\n",
    "\n",
    "x = SpatialDropout1D(0.2)(x)\n",
    "x = Bidirectional(CuDNNLSTM(40, return_sequences=True))(x)\n",
    "y = Bidirectional(CuDNNLSTM(40, return_sequences=True))(x)\n",
    "    \n",
    "atten_1 = Attention(max_len)(x) # skip connect\n",
    "atten_2 = Attention(max_len)(y)\n",
    "avg_pool = GlobalAveragePooling1D()(y)\n",
    "max_pool = GlobalMaxPooling1D()(y)\n",
    "    \n",
    "conc = concatenate([atten_1, atten_2, avg_pool, max_pool])\n",
    "conc = Dense(32, activation=\"relu\")(conc)\n",
    "conc = BatchNormalization(axis =1 )(conc)\n",
    "conc = Dropout(0.1)(conc)\n",
    "outp = Dense(1, activation=\"sigmoid\")(conc)    \n",
    "\n",
    "model = Model(inputs=inp, outputs=outp)\n",
    "model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=[f1_calc])\n",
    "print(model.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "_uuid": "414e72437916d8494ba7ddb022c0aacc577c7694"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 1175509 samples, validate on 130613 samples\n",
      "Stochastic weight averaging selected for last 6 epochs.\n",
      "Epoch 1/10\n",
      "1175509/1175509 [==============================] - 312s 265us/step - loss: 0.1419 - f1_calc: 0.5320 - val_loss: 0.1382 - val_f1_calc: 0.6415\n",
      "Epoch 2/10\n",
      "1175509/1175509 [==============================] - 305s 260us/step - loss: 0.1084 - f1_calc: 0.5995 - val_loss: 0.1094 - val_f1_calc: 0.6628\n",
      "Epoch 3/10\n",
      " 689664/1175509 [================>.............] - ETA: 2:01 - loss: 0.1033 - f1_calc: 0.6264"
     ]
    }
   ],
   "source": [
    "tmp_path = \"/tmp/tmp\" \n",
    "model_path = \"/tmp/model/\"\n",
    "\n",
    "swa = SWA('./keras_swa.model',4)\n",
    "\n",
    "model.fit(train_X, train_y, batch_size=512, epochs= 10 ,callbacks = [swa,], validation_data=(val_X, val_y))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "_uuid": "5aaac3bf024fac81c6afe0a5ad826aeb4eee041f"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "using swa weight model\n"
     ]
    }
   ],
   "source": [
    "# Load best model\n",
    "try:\n",
    "    print('using swa weight model')\n",
    "    model.load_weights('./keras_swa.model')\n",
    "except:\n",
    "    model.load_weights('./keras.model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "_uuid": "5823bf0da31cb3d86c810f67361ac45dd2eabfd7"
   },
   "outputs": [],
   "source": [
    "## Model design\n",
    "from IPython.display import SVG\n",
    "from keras.utils.vis_utils import model_to_dot\n",
    " \n",
    "SVG(model_to_dot(model).create(prog = \"dot\", format = \"svg\") )\n",
    " \n",
    "from keras.utils import plot_model\n",
    "plot_model(model, to_file = \"../working/model.png\") ##, show_shapes=True, show_layer_names=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "_uuid": "9f0fbc07e212b4416298b67ec3677d2eb5d67c48"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import Image\n",
    "Image(\"../working/model.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "_uuid": "e066d7a6c2b76df92f4f14c2b78cd89df9166d55"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1175509/1175509 [==============================] - 78s 67us/step\n",
      "0.16\n",
      "0.7358368889033645\n"
     ]
    }
   ],
   "source": [
    "from sklearn import metrics\n",
    "\n",
    "pred_train_glLSTM_y = model.predict([train_X], batch_size=1024, verbose=1)\n",
    "train_target = train_y\n",
    "\n",
    "thresh_glLSTM = 0.5\n",
    "f1_opt = 0\n",
    "for thresh in np.arange(0.1, 0.91, 0.01):\n",
    "    thresh = np.round(thresh, 2)\n",
    "    f1_value = metrics.f1_score(train_target, (pred_train_glLSTM_y.astype(float) > thresh).astype(int))\n",
    "    #print(\"F1 score at threshold {0} is {1}\".format(thresh, f1))\n",
    "    if f1_opt < f1_value:\n",
    "        f1_opt = f1_value\n",
    "        thresh_glLSTM = thresh\n",
    "print(thresh_glLSTM)\n",
    "pred_train_glLSTM_y_final = (pred_train_glLSTM_y > thresh_glLSTM).astype(np.int)\n",
    "print(f1_score(train_target, pred_train_glLSTM_y_final))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "_uuid": "0c225ddbd2aa1eaff90f747583a3ab3093028157"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "130613/130613 [==============================] - 9s 67us/step\n",
      "0.16\n",
      "0.6889561975768872\n"
     ]
    }
   ],
   "source": [
    "pred_val_glLSTM_y = model.predict([val_X], batch_size=1024, verbose=1)\n",
    "val_target = val_y\n",
    "\n",
    "thresh_glLSTM = 0.5\n",
    "f1_opt = 0\n",
    "for thresh in np.arange(0.1, 0.91, 0.01):\n",
    "    thresh = np.round(thresh, 2)\n",
    "    f1_value = metrics.f1_score(val_target, (pred_val_glLSTM_y.astype(float) > thresh).astype(int))\n",
    "    #print(\"F1 score at threshold {0} is {1}\".format(thresh, f1))\n",
    "    if f1_opt < f1_value:\n",
    "        f1_opt = f1_value\n",
    "        thresh_glLSTM = thresh\n",
    "print(thresh_glLSTM)\n",
    "#pred_val_bilstm_y_final = (pred_val_bigru_y > thresh_glLSTM).astype(np.int)\n",
    "pred_val_bilstm_y_final = (pred_val_glLSTM_y > thresh_glLSTM).astype(np.int)\n",
    "print(f1_score(val_target, pred_val_bilstm_y_final))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "_uuid": "e7b8c10071f5cce3dc19d986a43b298a5063b0c6"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.7358368889033645\n"
     ]
    }
   ],
   "source": [
    "print(f1_score(train_target, (pred_train_glLSTM_y > thresh_glLSTM).astype(np.int)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "_uuid": "9fce1248806b0516d0d985d19a9afc8fe1e6f48e"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "375806/375806 [==============================] - 25s 66us/step\n"
     ]
    }
   ],
   "source": [
    "pred_test_bilstm_y = model.predict([test_X], batch_size=1024, verbose=1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "_uuid": "24cecad273977e31efd40e28856d0d4b01c9fca2"
   },
   "outputs": [],
   "source": [
    "def most_false_negative(pred_y,true_y,thresh, N ):\n",
    "    pred_y_round = (pred_y>thresh)\n",
    "    \n",
    "    flag = (pred_y_round == 0) & (true_y == 1)[:,np.newaxis]\n",
    "    \n",
    "    idxs = np.where(flag)[0] # ignore the np.newaxis dimension\n",
    "    newidx = np.argsort(pred_y[idxs,0])\n",
    "\n",
    "    return idxs[newidx[:N]], pred_y[idxs[newidx[:N]]]\n",
    "\n",
    "def most_uncertain_negative(pred_y,true_y,thresh, N ):\n",
    "    pred_y_round = (pred_y>thresh)\n",
    "    flag = (pred_y_round == 0) & (true_y == 1)[:,np.newaxis]\n",
    "    \n",
    "    idxs = np.where(flag)[0] # ignore the np.newaxis dimension\n",
    "    newidx = np.argsort(pred_y[idxs,0])\n",
    "    return idxs[newidx[-N:]], pred_y[idxs[newidx[-N:]]]\n",
    "\n",
    "def most_uncertain_positive(pred_y,true_y,thresh, N ):\n",
    "    pred_y_round = (pred_y>thresh)\n",
    "    \n",
    "    flag = (pred_y_round == 1) & (true_y == 0)[:,np.newaxis]\n",
    "    \n",
    "    idxs = np.where(flag)[0] # ignore the np.newaxis dimension\n",
    "    newidx = np.argsort(pred_y[idxs,0])\n",
    "\n",
    "    return idxs[newidx[:N]], pred_y[idxs[newidx[:N]]]\n",
    "\n",
    "def most_false_positive(pred_y,true_y,thresh, N ):\n",
    "    pred_y_round = (pred_y>thresh)\n",
    "    flag = (pred_y_round == 1) & (true_y == 0)[:,np.newaxis]\n",
    "    \n",
    "    idxs = np.where(flag)[0] # ignore the np.newaxis dimension\n",
    "    newidx = np.argsort(pred_y[idxs,0])\n",
    "    return idxs[newidx[-N:]], pred_y[idxs[newidx[-N:]]]\n",
    "\n",
    "def most_true_positive(pred_y,true_y,thresh, N ):\n",
    "    pred_y_round = (pred_y>thresh)\n",
    "    \n",
    "    flag = (pred_y_round == 1) & (true_y == 1)[:,np.newaxis]\n",
    "    \n",
    "    idxs = np.where(flag)[0] # ignore the np.newaxis dimension\n",
    "    newidx = np.argsort(pred_y[idxs,0])\n",
    "\n",
    "    return idxs[newidx[-N:]], pred_y[idxs[newidx[-N:]]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "_uuid": "3657694445db065d42f8d764c630381671951468"
   },
   "outputs": [],
   "source": [
    "# Example Count to explore for each question\n",
    "NN = 10\n",
    "## Validation prediction\n",
    "pred =  pred_val_glLSTM_y\n",
    "thresh_fin = thresh_glLSTM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "_uuid": "ebe903da77322130eaa881b976069159abf4a90f"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "class 0 [0.91142184] Why was it ok to use public funds to transport blacks to the polls to vote for Obama? Blacks who never voted in their life voted solely on color. Why isn't that racist? If whites proudly stated they voted for a white person it would never fly.\n",
      "class 0 [0.9182942] If Muslim men cut their penis in childhood, then how are they able to pee and produce so many children?\n",
      "class 0 [0.9206569] Why would a young, beautiful woman like Melania marry an old, fat bastard like Donald Trump?\n",
      "class 0 [0.9273666] Should we kick out Muslims from India?\n",
      "class 0 [0.92889756] Are Bangladeshi women ugly?\n",
      "class 0 [0.9325481] Why don’t the Christian and Muslim quarter be under Palestinian control and the Jewish quarter be under Israeli control?\n",
      "class 0 [0.93764615] Are Indians ruining Quora with their silly answers copied from Facebook and WhatsApp?\n",
      "class 0 [0.94907016] If Mr. Frodo had asked Samwise Gamgee to castrate himself, would he have? How would Rosie react?\n",
      "class 0 [0.96715164] Did Luke Skywalker perform self- castration during his Jedi training?\n",
      "class 0 [0.9754995] If I put a gun to your head and ordered you to castrate your dog without anaesthesia, would you complete the procedure?\n"
     ]
    }
   ],
   "source": [
    "##what are insincere topics where the network strongly believe to be sincere ?\n",
    "idx, prob = most_false_positive( pred, val_y,thresh_fin,NN)\n",
    "for i in range(NN):\n",
    "    print('class',val_y[idx[i]],prob[i],val_df[\"question_text\"].values[idx[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "_uuid": "9500339a00ac2583e862b920407a39e1a7252ed7"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "class 1 [1.0047527e-06] What if iam having bleeding after my regular periods on my ovulation day?\n",
      "class 1 [1.940261e-06] What astrological combinations are needed to obtain a scholarship for studies?\n",
      "class 1 [3.3018885e-06] How can I design a logo on PowerPoint?\n",
      "class 1 [4.559017e-06] Is there any spot round in COMEDK? If yes, how is it conducting?\n",
      "class 1 [1.332324e-05] What are some popular night spots in Chicago?\n",
      "class 1 [1.5242949e-05] How many SITA devices are present in Ramayana?\n",
      "class 1 [1.5345091e-05] How competitive is the hiring process at NVIDIA?\n",
      "class 1 [1.994286e-05] How do I choose the best payday loan deal in Calgary?\n",
      "class 1 [2.1206737e-05] How many karats needed for a karat farm?\n",
      "class 1 [2.1854998e-05] How often do your gums bleed?\n"
     ]
    }
   ],
   "source": [
    "###what are sincere topics where the network strongly believe to be insincere ?\n",
    "idx, prob = most_false_negative( pred, val_y,thresh_fin,NN)\n",
    "for i in range(NN):\n",
    "    print('class',val_y[idx[i]],prob[i],val_df[\"question_text\"].values[idx[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "_uuid": "e0493c6b0a8780e59c7633101c9c40b094642514"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Keep in mind our chosen threshold is:\n",
      "0.16\n",
      "\n",
      "\n",
      "class 0 [0.16011325] Do all overweight and obese people binge?\n",
      "class 0 [0.16024299] Do preachers who keep affirming \"Biblical inerrancy\" have integrity or are they irresponsible to their congregation?\n",
      "class 0 [0.16028625] Why so many Indians in US prefer to live abridged versions of Indian life, starting from selecting spouses & culture, imstead of assimilating?\n",
      "class 0 [0.16035472] Why do taurus people love so hard?\n",
      "class 0 [0.16044523] How can change gay hormones in a straight man?\n",
      "\n",
      "\n",
      "class 1 [0.15925789] Israel is the cause of all the worlds problems why does it exist?\n",
      "class 1 [0.15939191] Do smart people pretend to believe in god in order to fit in with stupider people around them?\n",
      "class 1 [0.15946251] What will the people that insist Donald Trump colluded with Russia say when he is still president on Jan 19, 2021?\n",
      "class 1 [0.15958627] Was Hitler a furry?\n",
      "class 1 [0.15961424] How many times has Prince Harry had intimate relations with the Duchess of Cambridge?\n"
     ]
    }
   ],
   "source": [
    "###what are the topics where the network are most uncertain how to classify ?\n",
    "print('Keep in mind our chosen threshold is:')\n",
    "print(thresh_fin)\n",
    "print('\\n')\n",
    "NN_2 = np.int(NN/2)\n",
    "idx, prob = most_uncertain_positive( pred, val_y,thresh_fin,NN_2)\n",
    "for i in range(NN_2):\n",
    "    print('class',val_y[idx[i]],prob[i],val_df[\"question_text\"].values[idx[i]])\n",
    "print('\\n')\n",
    "\n",
    "idx, prob = most_uncertain_negative( pred, val_y,thresh_fin,NN_2)\n",
    "for i in range(NN_2):\n",
    "    print('class',val_y[idx[i]],prob[i],val_df[\"question_text\"].values[idx[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "_uuid": "fc36e4d9beb2a31449336382cb96b53ccb495baf"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "class 1 [0.98243356] Can a son have sexual intercourse with his mother?\n",
      "class 1 [0.9829297] I want to have sex with my elder sister in . how can I?\n",
      "class 1 [0.9832184] How can a human perform self-castration? God has requested I mutilate myself, but I was hoping there was a better way than cutting open my scrotum, pulling out my testicle and severing the spermatic cord.\n",
      "class 1 [0.9833286] Why cant I have sex with my mom?\n",
      "class 1 [0.984144] Mother's, would you be okay for trying to fuck your 30 yr old son if you were single, lived with him and his dick was big?\n",
      "class 1 [0.98487896] Why are the Quora moderators fucking assholes?\n",
      "class 1 [0.98634726] How many of you like to fuck your mom?\n",
      "class 1 [0.98866475] Which site has the best son mom incest videos for free?\n",
      "class 1 [0.99115294] Can I have intercourse with my mom?\n",
      "class 1 [0.99435556] I want to have sex with my sister. She has Milky large boobs and hot figure. How to proceed?\n"
     ]
    }
   ],
   "source": [
    "###what are insincere topics where the network strongly believes correctly ?\n",
    "idx, prob = most_true_positive( pred, val_y,thresh_fin,NN)\n",
    "for i in range(NN):\n",
    "    print('class',val_y[idx[i]],prob[i],val_df[\"question_text\"].values[idx[i]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "_uuid": "0248668dd048a158d90188c0f683b74a68e7edba"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7fed5e0d6470>"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Confusion Matrix\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn import metrics\n",
    "import matplotlib.pyplot as plt\n",
    "import scikitplot as skplt\n",
    "skplt.metrics.plot_confusion_matrix(val_y, pred_val_bilstm_y_final,labels = [1,0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "_uuid": "737de1ca164145be8de8e3352c813493337d1ca2"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.06866308680542621\n"
     ]
    }
   ],
   "source": [
    "pred_test_final = ( pred_test_bilstm_y > thresh_glLSTM).astype(np.int)\n",
    "submission = pd.DataFrame.from_dict({'qid': test_df['qid']})\n",
    "submission['prediction'] = pred_test_final\n",
    "print(np.sum(submission.prediction)/(test_df.shape[0]))\n",
    "submission.to_csv('submission.csv', index=False)\n"
   ]
  }
 ],
 "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}