Experimenting with Naive Bayes for neurosynth, just the beginning steps

Naive_bayes_starter.json

{
 "metadata": {
  "name": "Naive Bayes Starter"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "import naive_bayes as nb # the python script which is a work in progress\n# get the numpy feature and target array, replicating the criteria of the Large Scale... paper\nX, y = nb.get_X_y()",
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 1
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "# check that X and y are of the expected dimensions and view them\nprint X.ndim\nprint y.ndim\nprint X.size\nprint y.size\nprint X\nprint y",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "2\n1\n1393659176\n1544\n[[ 0.  0.  0. ...,  0.  0.  0.]\n [ 0.  0.  0. ...,  0.  0.  0.]\n [ 0.  0.  0. ...,  0.  0.  0.]\n ..., \n [ 0.  0.  0. ...,  0.  0.  0.]\n [ 0.  0.  0. ...,  0.  0.  0.]\n [ 0.  0.  0. ...,  0.  0.  0.]]\n['visual' 'social' 'visual' ..., 'attention' 'reward' 'executive']\n"
      }
     ],
     "prompt_number": 5
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "# Since y has string labels encode them to numerical values\nfrom sklearn import preprocessing\nle = preprocessing.LabelEncoder()\nle.fit(y)\n# see all the classes\nle.classes_",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 6,
       "text": "array(['action', 'attention', 'auditory', 'conflict', 'emotion',\n       'encoding', 'episodic', 'executive', 'imagery', 'language', 'pain',\n       'perception', 'phonological', 'recognition', 'retrieval', 'reward',\n       'semantic', 'sensory', 'social', 'spatial', 'verbal', 'visual'], dtype=object)"
      }
     ],
     "prompt_number": 6
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "# now encode y so that it has numerical classes rather than string\ny_enc = le.transform(y)\ny_enc",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 7,
       "text": "array([21, 18, 21, ...,  1, 15,  7])"
      }
     ],
     "prompt_number": 7
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "# see the inverse transform just to check\nle.inverse_transform(y_enc)",
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 8,
       "text": "array(['visual', 'social', 'visual', ..., 'attention', 'reward',\n       'executive'], dtype=object)"
      }
     ],
     "prompt_number": 8
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "# Now set up the naive bayes classifier\nfrom sklearn.naive_bayes import MultinomialNB\n# since study assumes uniform prior for each term, set fit_prior to false\nclf = MultinomialNB(fit_prior=False)\n# import cross validation module\nfrom sklearn import cross_validation\n# also the One vs One \nfrom sklearn.multiclass import OneVsOneClassifier",
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 10
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": "# Roughly this should be the way to proceed\n# however takes too long on my mac, so diverging to learn about deploying it via lonestar\nscores = cross_validation.cross_val_score(OneVsOneClassifier(clf), X, y_enc, cv=10)",
     "language": "python",
     "metadata": {},
     "outputs": []
    }
   ],
   "metadata": {}
  }
 ]
}