nlp_benchmark.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import json\n",
    "from sklearn.model_selection import StratifiedShuffleSplit\n",
    "from WordEmbeding.word_embeding_classifier import WordEmebdingClassifier\n",
    "from InferSent.classifier import InferSentClassifier,XGBoostClassifier\n",
    "from TFIDFVector.tfidf_classifier import TFIDFClassifer\n",
    "from SemHash.semhash_classifier import SemHashClassfier\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "nlu_data ='AskUbuntuCorpus.json'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp_data = None\n",
    "with open(nlu_data,'r') as f:\n",
    "    temp_data = json.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "texts = []\n",
    "labels = []\n",
    "for s in temp_data['sentences']:\n",
    "    texts.append(s['text'].lower())\n",
    "    labels.append(s['intent'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "sss = StratifiedShuffleSplit(n_splits=1,test_size=0.2) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_index, text_index = list(sss.split(texts,labels))[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_texts = [texts[index] for index in train_index]\n",
    "train_labels = [labels[index] for index in train_index]\n",
    "test_texts = [texts[index] for index in text_index]\n",
    "test_labels = [labels[index] for index in text_index]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "classifiers =[\"WordEmebdingClassifier\",\"InferSentClassifier\",\"TFIDFClassifer\",\"SemHashClassfier\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "result = []\n",
    "for cls_name in classifiers:\n",
    "    cls = globals()[cls_name]()\n",
    "    cls.train(train_texts,train_labels)\n",
    "    acc = cls.eval(test_texts,test_labels)\n",
    "    result.append([cls_name,acc])\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.DataFrame(result, columns = ['Name', 'Accuracy']) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Name</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>WordEmebdingClassifier</td>\n",
       "      <td>0.878788</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>InferSentClassifier</td>\n",
       "      <td>0.848485</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>TFIDFClassifer</td>\n",
       "      <td>0.727273</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>SemHashClassfier</td>\n",
       "      <td>0.848485</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     Name  Accuracy\n",
       "0  WordEmebdingClassifier  0.878788\n",
       "1     InferSentClassifier  0.848485\n",
       "2          TFIDFClassifer  0.727273\n",
       "3        SemHashClassfier  0.848485"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5,1,'Intent Rec Accuracy')"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "fig, ax = plt.subplots()\n",
    "ax.barh(df['Name'], df['Accuracy'], align='center')\n",
    "ax.set_yticklabels(df['Name'])\n",
    "ax.invert_yaxis()  # labels read top-to-bottom\n",
    "ax.set_xlabel('Performance')\n",
    "ax.set_title('Intent Rec Accuracy')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Conclusion"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We compare several different sentence embeding for intent classification.\n",
    "1. Simple sum of wordembeding, \n",
    "2. Facebook InferSent.\n",
    "3. TF-IDF\n",
    "4. Semhash, https://arxiv.org/abs/1810.07150\n",
    "\n",
    "### Simple sum the word embeding achieve the best accuray in intention classification.\n",
    "This is quiet interesting. Why could simple sum of wording embeding represent the sentence very well?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}