SVM Regression Approach.ipynb

{
 "metadata": {
  "name": "",
  "signature": "sha256:adfe6e9250909379b0625a1b2147b631a8708d6062e24ac3c8abdbd033b1d41d"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "heading",
     "level": 1,
     "metadata": {},
     "source": [
      "Restaurant Revenue Prediction - Kaggle Competiton"
     ]
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Author: Patrick Boueri"
     ]
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Approach: SVM Regression"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "from sklearn import svm \n",
      "from sklearn import cross_validation\n",
      "from sklearn import preprocessing\n",
      "import numpy as np\n",
      "import pandas as pd\n",
      "from matplotlib import pyplot as plt\n",
      "import os\n",
      "import sys\n",
      "import random\n",
      "%matplotlib inline"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 1
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "os.path.dirname(os.path.abspath(sys.argv[0]))"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 2,
       "text": [
        "'C:\\\\Users\\\\Patrick\\\\Columbia\\\\Kaggle Competition\\\\Restaurant Revenue Prediction'"
       ]
      }
     ],
     "prompt_number": 2
    },
    {
     "cell_type": "heading",
     "level": 3,
     "metadata": {},
     "source": [
      "Load in the training data and display some summary stats"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "train = pd.read_csv('train.csv',sep=',')\n",
      "train.set_index('Id',inplace=True)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 3
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "plt.hist(train['revenue']) ## The distribution is log normal so correct for it"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 4,
       "text": [
        "(array([ 36.,  60.,  23.,  10.,   5.,   0.,   1.,   0.,   1.,   1.]),\n",
        " array([  1149870. ,   3004576.9,   4859283.8,   6713990.7,   8568697.6,\n",
        "         10423404.5,  12278111.4,  14132818.3,  15987525.2,  17842232.1,\n",
        "         19696939. ]),\n",
        " <a list of 10 Patch objects>)"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAENCAYAAAAfTp5aAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRU5ErkJggg==\n",
       "text": [
        "<matplotlib.figure.Figure at 0x44246d8>"
       ]
      }
     ],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "plt.hist(np.log(train['revenue'])) ## Really good looking normal distribution, we can work with it"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 5,
       "text": [
        "(array([  2.,  11.,  15.,  26.,  40.,  22.,  13.,   5.,   1.,   2.]),\n",
        " array([ 13.95515945,  14.23924089,  14.52332232,  14.80740376,\n",
        "         15.09148519,  15.37556663,  15.65964806,  15.9437295 ,\n",
        "         16.22781093,  16.51189237,  16.7959738 ]),\n",
        " <a list of 10 Patch objects>)"
       ]
      },
      {
       "metadata": {},
       "output_type": "display_data",
       "png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEACAYAAABI5zaHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nXZIa9P+ayUGQYUFFbQAAAABJRU5ErkJggg==\n",
       "text": [
        "<matplotlib.figure.Figure at 0x9058e48>"
       ]
      }
     ],
     "prompt_number": 5
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "yTrain = np.log(train['revenue'])\n",
      "xTrain = train.drop('revenue',axis=1)\n",
      "now = pd.datetime.now() #Get the current datetime to figure out how old the shops are\n",
      "xTrain['Open Date'] = pd.to_datetime(xTrain['Open Date']) #\n",
      "xTrain['Years Since Open'] = ((now - xTrain['Open Date'])/np.timedelta64(1,'Y')).astype('int') ##How many years ago the shops opened\n",
      "xTrain = pd.merge(xTrain,pd.get_dummies(xTrain['Type']),left_index=True,right_index=True,how='left') ##Get the indicator vars for Type\n",
      "xTrain = pd.merge(xTrain,pd.get_dummies(xTrain['City Group']),left_index=True,right_index=True,how='left')##Get the indicator vars for City Group\n",
      "xTrain.drop(['Type', 'City Group','City','Open Date'],inplace=True,axis=1) ##Remove unnecessary columns"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 6
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "k=10\n",
      "C = np.logspace(-5,5,num=20)\n",
      "Eps = np.logspace(-5,5,num=20)\n",
      "allRows = list(xTrain.index)\n",
      "remainingRows = allRows\n",
      "kIndex = {}\n",
      "for i in range(0,k):\n",
      "    sampleRows = random.sample(remainingRows,int(len(xTrain)/k))\n",
      "    remainingRows = list(set(remainingRows)-set(sampleRows))\n",
      "    kIndex[i] = sampleRows\n",
      "kIndex[k] = remainingRows\n",
      "kIndex"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 21,
       "text": [
        "{0: [104, 69, 67, 123, 33, 135, 116, 24, 94, 41, 85, 21, 105],\n",
        " 1: [113, 83, 22, 87, 106, 103, 107, 54, 84, 128, 121, 48, 114],\n",
        " 2: [55, 98, 75, 50, 57, 11, 0, 16, 80, 43, 118, 4, 90],\n",
        " 3: [119, 72, 10, 12, 77, 23, 117, 82, 17, 111, 81, 46, 26],\n",
        " 4: [125, 30, 39, 1, 7, 59, 86, 136, 124, 132, 5, 73, 35],\n",
        " 5: [134, 61, 102, 25, 78, 63, 100, 97, 8, 126, 3, 13, 127],\n",
        " 6: [109, 6, 18, 20, 110, 96, 40, 91, 131, 74, 53, 52, 42],\n",
        " 7: [37, 112, 44, 93, 60, 34, 89, 62, 88, 14, 38, 32, 29],\n",
        " 8: [56, 47, 133, 76, 120, 19, 108, 2, 101, 71, 64, 122, 70],\n",
        " 9: [15, 28, 99, 95, 49, 65, 79, 51, 45, 115, 68, 66, 27],\n",
        " 10: [129, 130, 36, 9, 58, 92, 31]}"
       ]
      }
     ],
     "prompt_number": 21
    },
    {
     "cell_type": "heading",
     "level": 2,
     "metadata": {},
     "source": [
      "Cross Validation - Coarse parameter grid search"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "RMSEresults = [];\n",
      "nonIndicatorCols = xTrain.columns[:-5]\n",
      "for i in range(0,k+1):\n",
      "    kList = []\n",
      "    remainingRows = list(set(allRows)-set(kIndex[i]))\n",
      "    cvXTrain = xTrain.ix[remainingRows]\n",
      "    cvXTest = xTrain.ix[kIndex[k]]\n",
      "    cvYTrain = yTrain.ix[remainingRows]\n",
      "    cvYTest = yTrain.ix[kIndex[k]]\n",
      "    for c in C:\n",
      "        tempList = []\n",
      "        for eps in Eps:\n",
      "            scaler = preprocessing.StandardScaler().fit(cvXTrain[nonIndicatorCols])\n",
      "            cvXTrainNorm = cvXTrain\n",
      "            cvXTrainNorm[nonIndicatorCols] = scaler.transform(cvXTrain[nonIndicatorCols])\n",
      "            cvXTestNorm = cvXTest\n",
      "            cvXTestNorm[nonIndicatorCols] = scaler.transform(cvXTestNorm[nonIndicatorCols])\n",
      "            svmR = svm.SVR(C = c,epsilon=eps)\n",
      "            svmR.fit(cvXTrainNorm,cvYTrain)\n",
      "            yTestPred = svmR.predict(cvXTest)\n",
      "            rmse = (np.mean((yTestPred-cvYTest)**2))**.5\n",
      "            tempList.append(rmse)\n",
      "        kList.append(tempList)\n",
      "    RMSEresults.append(kList)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": []
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "grid = []\n",
      "absMin ={}\n",
      "absMin['RMSE'] =1\n",
      "for i in range(0,len(C)):\n",
      "    tempList = [];\n",
      "    for j in  range(0,len(Eps)):\n",
      "        sumMSE = 0\n",
      "        for kI in range(0,k+1):\n",
      "            sumMSE = sumMSE + RMSEresults[kI][i][j]\n",
      "        meanMSE = sumMSE/(k+1)\n",
      "        if absMin['RMSE'] > meanMSE:\n",
      "            absMin['RMSE'] = meanMSE\n",
      "            absMin['C'] = C[i]\n",
      "            absMin['Eps'] = Eps[j]\n",
      "        tempList.append(meanMSE)\n",
      "    grid.append(tempList)\n",
      "grid = np.matrix(data=grid)\n",
      "absMin"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 49,
       "text": [
        "{'RMSE': 0.035635234294532366,\n",
        " 'C': 20.691380811147901,\n",
        " 'Eps': 3.3598182862837813e-05}"
       ]
      }
     ],
     "prompt_number": 49
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "xTrainNorm = xTrain\n",
      "scaler = preprocessing.StandardScaler().fit(xTrain[nonIndicatorCols])\n",
      "xTrainNorm[nonIndicatorCols] = scaler.transform(xTrain[nonIndicatorCols])\n",
      "SVMR = svm.SVR(C = absMin['C'],epsilon = absMin['Eps'])\n",
      "SVMR.fit(xTrainNorm,yTrain)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 53,
       "text": [
        "SVR(C=20.691380811147901, cache_size=200, coef0=0.0, degree=3,\n",
        "  epsilon=3.3598182862837813e-05, gamma=0.0, kernel='rbf', max_iter=-1,\n",
        "  probability=False, random_state=None, shrinking=True, tol=0.001,\n",
        "  verbose=False)"
       ]
      }
     ],
     "prompt_number": 53
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "test = pd.read_csv('test.csv',sep=',')\n",
      "test.set_index('Id',inplace=True)\n",
      "now = pd.datetime.now() #Get the current datetime to figure out how old the shops are\n",
      "test['Open Date'] = pd.to_datetime(test['Open Date']) #\n",
      "test['Years Since Open'] = ((now - test['Open Date'])/np.timedelta64(1,'Y')).astype('int') ##How many years ago the shops opened\n",
      "test = pd.merge(test,pd.get_dummies(test['Type']),left_index=True,right_index=True,how='left') ##Get the indicator vars for Type\n",
      "test = pd.merge(test,pd.get_dummies(test['City Group']),left_index=True,right_index=True,how='left')##Get the indicator vars for City Group\n",
      "test.drop(['Type', 'City Group','City','Open Date','MB'],inplace=True,axis=1) ##Remove unnecessary columns, MB some unseen variable\n",
      "testNorm = test\n",
      "testNorm[nonIndicatorCols] = (test[nonIndicatorCols] - test[nonIndicatorCols].mean())/test[nonIndicatorCols].std() ##dunno about this, ask Francois\n",
      "testPred = SVMR.predict(testNorm)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 54
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "testPred = np.exp(np.reshape(testPred,(len(testPred),1)))\n",
      "testPred.tofile('testSVMPred.csv',sep='\\n')"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 55
    }
   ],
   "metadata": {}
  }
 ]
}