EdXLogisticModel.ipynb

{
 "worksheets": [
  {
   "cells": [
    {
     "metadata": {},
     "cell_type": "code",
     "input": "import pandas as pd, numpy as np, datetime, random, cPickle as pickle\nfrom sklearn import linear_model\nfrom sklearn.metrics import roc_auc_score, roc_curve, confusion_matrix\nfrom sklearn.feature_selection import RFE\nfrom __future__ import division\npd.set_option('max_colwidth', 200)\n%matplotlib inline",
     "prompt_number": 1,
     "outputs": [],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "markdown",
     "source": "### Load data"
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "activity = pickle.load(open('activity.pickle','rb'))",
     "prompt_number": 2,
     "outputs": [],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "markdown",
     "source": "### Building models"
    },
    {
     "metadata": {},
     "cell_type": "markdown",
     "source": "Logistic regression"
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "cutoff = int(len(activity.index) * 0.75)\nrows = random.sample(activity.index, cutoff)\ntrainset = activity.ix[rows]\ntestset = activity.drop(rows)",
     "prompt_number": 3,
     "outputs": [],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "for t in ['6hours','1hour','30min','6min']:\n    logit = linear_model.LogisticRegression()\n    cols = [c for c in activity.columns if t in c or '_correct_' in c]\n    logit.fit(trainset[cols], trainset['success'])\n    pred = logit.predict(testset[cols])\n    roc = roc_auc_score(testset['success'], pred)\n    ls = logit.score(testset[cols], testset['success'])\n    # print sorted(zip(cols, logit.coef_[0]), key=lambda x: abs(x[1]), reverse=True)\n    print '%s\\t%g\\t%g' % (t,roc,ls)",
     "prompt_number": 4,
     "outputs": [
      {
       "output_type": "stream",
       "text": "6hours\t0.827618\t0.832291\n1hour\t0.828469\t0.832993",
       "stream": "stdout"
      },
      {
       "output_type": "stream",
       "text": "\n30min\t0.828691\t0.833184",
       "stream": "stdout"
      },
      {
       "output_type": "stream",
       "text": "\n6min\t0.828863\t0.833504",
       "stream": "stdout"
      },
      {
       "output_type": "stream",
       "text": "\n",
       "stream": "stdout"
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "cols = activity.columns[4:]\nfeat = RFE(estimator=logit, n_features_to_select=1, step=1)\nfeat.fit(trainset[cols], trainset['success'])",
     "prompt_number": 5,
     "outputs": [
      {
       "text": "RFE(estimator=LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,\n          intercept_scaling=1, penalty='l2', random_state=None, tol=0.0001),\n  estimator_params={}, n_features_to_select=1, step=1, verbose=0)",
       "output_type": "pyout",
       "metadata": {},
       "prompt_number": 5
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "sorted(zip(cols,feat.ranking_), key=lambda x: x[1])",
     "prompt_number": 6,
     "outputs": [
      {
       "text": "[('person_correct_share', 1),\n ('quiz_correct_share', 2),\n ('resource_6min_quiz_any', 3),\n ('resource_30min_quiz_any', 4),\n ('resource_1hour_quiz_any', 5),\n ('resource_6hours_quiz_any', 6),\n ('resource_6min_quiz', 7),\n ('resource_6min_grades', 8),\n ('resource_6min_wiki', 9),\n ('resource_6hours_wiki', 10),\n ('resource_6hours_course_any', 11),\n ('resource_6hours_discuss', 12),\n ('resource_1hour_discuss', 13),\n ('resource_1hour_wiki', 14),\n ('resource_6hours_grades', 15),\n ('resource_30min_grades', 16),\n ('resource_1hour_video', 17),\n ('resource_6min_discuss', 18),\n ('resource_1hour_quiz', 19),\n ('resource_6hours_quiz', 20),\n ('resource_30min_wiki', 21),\n ('resource_30min_discuss', 22),\n ('resource_30min_quiz', 23),\n ('resource_6min_wiki_any', 24),\n ('resource_6hours_video', 25),\n ('resource_6min_video', 26),\n ('resource_6hours_video_any', 27),\n ('resource_6hours_grades_any', 28),\n ('resource_6min_course', 29),\n ('resource_30min_video', 30),\n ('resource_6min_course_any', 31),\n ('resource_1hour_course_any', 32),\n ('resource_1hour_course', 33),\n ('resource_1hour_wiki_any', 34),\n ('resource_30min_wiki_any', 35),\n ('resource_30min_discuss_any', 36),\n ('resource_6hours_discuss_any', 37),\n ('resource_1hour_discuss_any', 38),\n ('resource_1hour_grades_any', 39),\n ('resource_30min_grades_any', 40),\n ('resource_6hours_course', 41),\n ('resource_30min_course_any', 42),\n ('resource_1hour_grades', 43),\n ('resource_30min_course', 44),\n ('resource_6min_video_any', 45),\n ('resource_6hours_wiki_any', 46),\n ('resource_30min_video_any', 47),\n ('resource_6min_grades_any', 48),\n ('resource_1hour_video_any', 49),\n ('resource_6min_discuss_any', 50)]",
       "output_type": "pyout",
       "metadata": {},
       "prompt_number": 6
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "cols = ['quiz_correct_share',\n        'person_correct_share',\n        'resource_30min_course',\n        'resource_6hours_discuss',\n        'resource_6hours_grades',\n        'resource_6hours_quiz',\n        'resource_30min_wiki',\n        'resource_30min_video']\nlogit = linear_model.LogisticRegression(C=1e5)\nlogit.fit(trainset[cols], trainset['success'])\npred = logit.predict(testset[cols])\nroc = roc_auc_score(testset['success'], pred)\nls = logit.score(testset[cols], testset['success'])",
     "prompt_number": 25,
     "outputs": [],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "roc",
     "prompt_number": 26,
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 26,
       "metadata": {},
       "text": "0.83271090175215978"
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "zip(cols,logit.coef_[0])",
     "prompt_number": 13,
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 13,
       "metadata": {},
       "text": "[('quiz_correct_share', 1.9588739831540602),\n ('person_correct_share', 1.4164212229647777),\n ('resource_30min_course', -0.0053761447780298725),\n ('resource_6hours_discuss', 0.002423759621709912),\n ('resource_6hours_grades', -0.0092037533471761752),\n ('resource_6hours_quiz', 0.012233299020303021),\n ('resource_30min_wiki', -0.0051592065705068184),\n ('resource_30min_video', 0.00015785408217884936)]"
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "markdown",
     "source": "#### 10-fold cross validation to get average coefficient"
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "cols = ['quiz_correct_share',\n        'person_correct_share',\n        'resource_30min_course',\n        'resource_1hour_discuss',\n        'resource_6hours_grades',\n        'resource_6hours_quiz',\n        'resource_30min_wiki',\n        'resource_30min_video']\ncoef_dict = {k: [] for k in cols}\ncutoffs = np.linspace(0,len(activity.index),10).astype(int)\n\nfor i in range(len(cutoffs)-1):\n    trainset = pd.concat([activity[:cutoffs[i]],activity[cutoffs[i+1]:]])\n    testset = activity[cutoffs[i]:cutoffs[i+1]]\n    \n    logit = linear_model.LogisticRegression(C=1e5)\n    logit.fit(trainset[cols], trainset['success'])\n    pred = logit.predict(testset[cols])\n    roc = roc_auc_score(testset['success'], pred)\n    print roc\n    ls = logit.score(testset[cols], testset['success'])\n    cur_coef = dict(zip(cols,logit.coef_[0]))\n    for c in cur_coef.keys():\n        coef_dict[c].append(cur_coef[c])",
     "prompt_number": 24,
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "0.822335766423\n0.826636644675"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n0.830296227347"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n0.820790076235"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n0.826902885421"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n0.834583815565"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n0.827650411056"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n0.830916805743"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n0.832217852166"
      },
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "\n"
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "sorted([(k, sum(c) / len(c)) for k, c in coef_dict.items()], key=lambda x: abs(x[1]), reverse=True)",
     "prompt_number": 25,
     "outputs": [
      {
       "output_type": "pyout",
       "prompt_number": 25,
       "metadata": {},
       "text": "[('person_correct_share', 7.7775643794348861),\n ('quiz_correct_share', 6.6159103178916485),\n ('resource_6hours_quiz', 0.44101550329900574),\n ('resource_6hours_grades', -0.33686739046138942),\n ('resource_30min_video', 0.23319519041209835),\n ('resource_1hour_discuss', -0.12726112503349668),\n ('resource_30min_course', -0.10627057826122169),\n ('resource_30min_wiki', -0.00054306281933279917)]"
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "for v, c in sorted([(k, sum(c) / len(c)) for k, c in coef_dict.items()], key=lambda x: abs(x[1]), reverse=True):\n    print '%s\\t%g' % (v, c)",
     "prompt_number": 19,
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": "person_correct_share\t7.77582\nquiz_correct_share\t6.61585\nresource_6hours_quiz\t0.443215\nresource_6hours_grades\t-0.346701\nresource_30min_video\t0.225609\nresource_30min_course\t-0.110822\nresource_6hours_discuss\t0.0774493\nresource_30min_wiki\t-0.00638102\n"
      }
     ],
     "language": "python",
     "trusted": true,
     "collapsed": false
    },
    {
     "metadata": {},
     "cell_type": "code",
     "input": "",
     "outputs": [],
     "language": "python",
     "trusted": true,
     "collapsed": false
    }
   ],
   "metadata": {}
  }
 ],
 "metadata": {
  "name": "",
  "signature": "sha256:eb47e2e310cdf37c09ea9c66be2746bc0d72d0b41f39ef8f1cb07d1cb0bfe6b6"
 },
 "nbformat": 3
}