JTM experiments

jtm_experiment.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from jtm import JTM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "vocab = ['apple', 'banana', 'computer', 'mac', 'burger', 'ipad']\n",
    "categories = ['PC', 'FOOD']\n",
    "\n",
    "X = [[],[]]\n",
    "X[0].append([0,0,2,3,3,5]) # apple, apple, computer, mac, mac, ipad\n",
    "X[1].append([0]) # PC\n",
    "X[0].append([0,0,1,3,3,4]) # apple, apple, banana, mac, mac, burger\n",
    "X[1].append([1]) # FOOD\n",
    "X[0].append([2,2,5,5]) # computer, computer, ipad, ipad\n",
    "X[1].append([0]) # PC\n",
    "X[0].append([1,1,4,4]) # banana, banana, burger, burger\n",
    "X[1].append([1]) # FOOD\n",
    "X[0].append([0,3]) # apple, mac\n",
    "X[1].append([0]) # PC\n",
    "X[0].append([0,3]) # apple, mac\n",
    "X[1].append([1]) # FOOD\n",
    "\n",
    "V = [len(vocab), len(categories)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# モデルの定義\n",
    "K = 2\n",
    "alpha=0.01\n",
    "beta=0.01\n",
    "max_iter=1000\n",
    "model = JTM(K=K, alpha=alpha, beta=beta, max_iter=max_iter)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<jtm.JTM instance at 0x10cd1c638>"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# JTMのフィッティング\n",
    "model.fit(X,V)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[array([1, 1, 1, 1, 1, 1]), array([0, 0, 0, 0, 0, 0]), array([1, 1, 1, 1]), array([0, 0, 0, 0]), array([1, 1]), array([0, 0])]\n",
      "[array([1]), array([0]), array([1]), array([0]), array([1]), array([0])]\n"
     ]
    }
   ],
   "source": [
    "# トピック割り当ての結果\n",
    "# 最後の２つのドキュメントは出現単語は同じだけどカテゴリの情報をうまく使って正しいトピックに割り当てられている\n",
    "print model.Z[0]\n",
    "print model.Z[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<jtm.JTM instance at 0x10cd1c638>"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# LDAのフィッティング\n",
    "# カテゴリの情報（side information）を与えなければJTMはLDAと等価\n",
    "X = [X[0]]\n",
    "V = [V[0]]\n",
    "model.fit(X,V)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[array([1, 1, 0, 1, 1, 0]), array([1, 1, 1, 1, 1, 1]), array([0, 0, 0, 0]), array([1, 1, 1, 1]), array([1, 1]), array([1, 1])]\n"
     ]
    }
   ],
   "source": [
    "# トピック割り当ての結果\n",
    "# カテゴリ情報が与えられていないので、普通のLDAでは当然最後の２つのドキュメントを正しくトピック割り当て出来ない\n",
    "print model.Z[0]"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}