Universal_Sentence_Encoder.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# import sys\n",
    "# !{sys.executable} -m pip install tensorflow_hub"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import tensorflow_hub as hub\n",
    "import pandas as pd\n",
    "from sklearn import preprocessing\n",
    "import keras\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# read training data\n",
    "train_data = pd.read_csv('train_data.csv',\n",
    "                            sep='\\t',\n",
    "                        encoding='utf-8',\n",
    "                        index_col=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x_train = train_data['textbody'].values\n",
    "y_train = train_data['hyperpartisan'].values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# prepare label encoder\n",
    "le = preprocessing.LabelEncoder()\n",
    "le.fit(y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def encode(le, labels):\n",
    "    enc = le.transform(labels)\n",
    "    return keras.utils.to_categorical(enc)\n",
    "\n",
    "def decode(le, one_hot):\n",
    "    dec = np.argmax(one_hot, axis=1)\n",
    "    return le.inverse_transform(dec)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# encode training labels\n",
    "y_train_encoded = encode(le, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from keras.layers import Input, Lambda, Dense, Dropout\n",
    "from keras.models import Model\n",
    "import keras.backend as K\n",
    "from keras.regularizers import l2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get the universal sentence encoder\n",
    "url = \"https://tfhub.dev/google/universal-sentence-encoder-large/3\"\n",
    "embed = hub.Module(url)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def UniversalEmbedding(x):\n",
    "    return embed(tf.squeeze(tf.cast(x, tf.string)))\n",
    "\n",
    "# create model\n",
    "input_text = Input(shape=(1,), dtype=tf.string)\n",
    "embedding = Lambda(UniversalEmbedding, output_shape=(512, ))(input_text)\n",
    "dropout_dense1 = Dropout(0.3)(embedding)\n",
    "dense = Dense(256, activation='relu', activity_regularizer=l2(0.01))(dropout_dense1)\n",
    "dropout_dense2 = Dropout(0.3)(dropout_dense1)\n",
    "pred = Dense(2, activation='softmax')(dropout_dense2)\n",
    "model = Model(inputs=[input_text], outputs=pred)\n",
    "model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# run training\n",
    "with tf.Session() as session:\n",
    "    K.set_session(session)\n",
    "    session.run(tf.global_variables_initializer())  \n",
    "    session.run(tf.tables_initializer())\n",
    "    model.load_weights('./model_hyperp.h5') \n",
    "    history = model.fit(x_train, y_train_encoded, epochs=2, batch_size=32, shuffle=True, verbose=1)\n",
    "    model.save_weights('./model_hyperp.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# read test data\n",
    "test_data = pd.read_csv('test_data.csv',\n",
    "                            sep='\\t',\n",
    "                        encoding='utf-8',\n",
    "                        index_col=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x_test = test_data['textbody'].values\n",
    "y_test = test_data['hyperpartisan'].values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# encode test labels\n",
    "y_test_encoded = encode(le, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# run validation\n",
    "with tf.Session() as session:\n",
    "    K.set_session(session)\n",
    "    session.run(tf.global_variables_initializer())\n",
    "    session.run(tf.tables_initializer())\n",
    "    model.load_weights('./model_hyperp.h5')  \n",
    "    predicts = model.predict(x_test, batch_size=512)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# calculate classification report (precision, recall, f1)\n",
    "y_test2 = decode(le, y_test_encoded)\n",
    "y_preds = decode(le, predicts)\n",
    "\n",
    "from sklearn import metrics\n",
    "\n",
    "print(metrics.classification_report(y_test2, y_preds))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# calculate accuracy\n",
    "metrics.accuracy_score(y_test2, y_preds)"
   ]
  }
 ],
 "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": 2
}