Appendix01_UserTypeClassification_SMOTE.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<a id='sec0'></a>\n",
    "# Evaluating RandomForest using SMOTE oversampled dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Use RandomForestClassifier with max_depth=40\n",
    "- Train the model on oversample set and test it on the imbalanced set with the original class ratio\n",
    "- Do the above with multiple rounds of oversampling to make sure the results are similar\n",
    "- Only 2016 data was used to keep the sample size manageable for computing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.model_selection import train_test_split, GridSearchCV\n",
    "from sklearn.metrics import precision_recall_curve\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from imblearn.over_sampling import SMOTE\n",
    "from collections import Counter\n",
    "\n",
    "import my_scoring_functions as msf\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Setting up Data (This time tried to label subscriber as 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data = pd.read_csv('../data/for_predictions/dataset_2016_membership_pred.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>tripduration</th>\n",
       "      <th>start station latitude</th>\n",
       "      <th>start station longitude</th>\n",
       "      <th>end station latitude</th>\n",
       "      <th>end station longitude</th>\n",
       "      <th>start_year</th>\n",
       "      <th>start_month</th>\n",
       "      <th>start_day</th>\n",
       "      <th>start_hour</th>\n",
       "      <th>start_dayofweek</th>\n",
       "      <th>gender</th>\n",
       "      <th>birth year</th>\n",
       "      <th>usertype</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>923</td>\n",
       "      <td>40.719105</td>\n",
       "      <td>-73.999733</td>\n",
       "      <td>40.711512</td>\n",
       "      <td>-74.015756</td>\n",
       "      <td>2016</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>1958.0</td>\n",
       "      <td>Subscriber</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>379</td>\n",
       "      <td>40.743943</td>\n",
       "      <td>-73.979661</td>\n",
       "      <td>40.748549</td>\n",
       "      <td>-73.988084</td>\n",
       "      <td>2016</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>1969.0</td>\n",
       "      <td>Subscriber</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>589</td>\n",
       "      <td>40.750664</td>\n",
       "      <td>-74.001768</td>\n",
       "      <td>40.739017</td>\n",
       "      <td>-74.002638</td>\n",
       "      <td>2016</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>2</td>\n",
       "      <td>1982.0</td>\n",
       "      <td>Subscriber</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>889</td>\n",
       "      <td>40.719105</td>\n",
       "      <td>-73.999733</td>\n",
       "      <td>40.711512</td>\n",
       "      <td>-74.015756</td>\n",
       "      <td>2016</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>2</td>\n",
       "      <td>1961.0</td>\n",
       "      <td>Subscriber</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1480</td>\n",
       "      <td>40.765909</td>\n",
       "      <td>-73.976342</td>\n",
       "      <td>40.765909</td>\n",
       "      <td>-73.976342</td>\n",
       "      <td>2016</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>1952.0</td>\n",
       "      <td>Subscriber</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   tripduration  start station latitude  start station longitude  \\\n",
       "0           923               40.719105               -73.999733   \n",
       "1           379               40.743943               -73.979661   \n",
       "2           589               40.750664               -74.001768   \n",
       "3           889               40.719105               -73.999733   \n",
       "4          1480               40.765909               -73.976342   \n",
       "\n",
       "   end station latitude  end station longitude  start_year  start_month  \\\n",
       "0             40.711512             -74.015756        2016            1   \n",
       "1             40.748549             -73.988084        2016            1   \n",
       "2             40.739017             -74.002638        2016            1   \n",
       "3             40.711512             -74.015756        2016            1   \n",
       "4             40.765909             -73.976342        2016            1   \n",
       "\n",
       "   start_day  start_hour  start_dayofweek  gender  birth year    usertype  \n",
       "0          1           0                4       1      1958.0  Subscriber  \n",
       "1          1           0                4       1      1969.0  Subscriber  \n",
       "2          1           0                4       2      1982.0  Subscriber  \n",
       "3          1           0                4       2      1961.0  Subscriber  \n",
       "4          1           0                4       1      1952.0  Subscriber  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>tripduration</th>\n",
       "      <th>start station latitude</th>\n",
       "      <th>start station longitude</th>\n",
       "      <th>end station latitude</th>\n",
       "      <th>end station longitude</th>\n",
       "      <th>start_year</th>\n",
       "      <th>start_month</th>\n",
       "      <th>start_day</th>\n",
       "      <th>start_hour</th>\n",
       "      <th>start_dayofweek</th>\n",
       "      <th>gender</th>\n",
       "      <th>birth year</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "      <td>12222976</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>819</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>2016</td>\n",
       "      <td>7</td>\n",
       "      <td>15</td>\n",
       "      <td>13</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>1977</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>7586</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>8</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>61</td>\n",
       "      <td>0</td>\n",
       "      <td>-74</td>\n",
       "      <td>0</td>\n",
       "      <td>-74</td>\n",
       "      <td>2016</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1885</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>362</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>2016</td>\n",
       "      <td>5</td>\n",
       "      <td>8</td>\n",
       "      <td>9</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1970</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>582</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>2016</td>\n",
       "      <td>7</td>\n",
       "      <td>16</td>\n",
       "      <td>15</td>\n",
       "      <td>3</td>\n",
       "      <td>1</td>\n",
       "      <td>1980</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>956</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>40</td>\n",
       "      <td>-73</td>\n",
       "      <td>2016</td>\n",
       "      <td>10</td>\n",
       "      <td>23</td>\n",
       "      <td>18</td>\n",
       "      <td>4</td>\n",
       "      <td>1</td>\n",
       "      <td>1987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>8933552</td>\n",
       "      <td>40</td>\n",
       "      <td>0</td>\n",
       "      <td>40</td>\n",
       "      <td>0</td>\n",
       "      <td>2016</td>\n",
       "      <td>12</td>\n",
       "      <td>31</td>\n",
       "      <td>23</td>\n",
       "      <td>6</td>\n",
       "      <td>2</td>\n",
       "      <td>2000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       tripduration  start station latitude  start station longitude  \\\n",
       "count      12222976                12222976                 12222976   \n",
       "mean            819                      40                      -73   \n",
       "std            7586                       0                        0   \n",
       "min              61                       0                      -74   \n",
       "25%             362                      40                      -73   \n",
       "50%             582                      40                      -73   \n",
       "75%             956                      40                      -73   \n",
       "max         8933552                      40                        0   \n",
       "\n",
       "       end station latitude  end station longitude  start_year  start_month  \\\n",
       "count              12222976               12222976    12222976     12222976   \n",
       "mean                     40                    -73        2016            7   \n",
       "std                       0                      0           0            3   \n",
       "min                       0                    -74        2016            1   \n",
       "25%                      40                    -73        2016            5   \n",
       "50%                      40                    -73        2016            7   \n",
       "75%                      40                    -73        2016           10   \n",
       "max                      40                      0        2016           12   \n",
       "\n",
       "       start_day  start_hour  start_dayofweek    gender  birth year  \n",
       "count   12222976    12222976         12222976  12222976    12222976  \n",
       "mean          15          13                2         1        1977  \n",
       "std            8           4                1         0          11  \n",
       "min            1           0                0         0        1885  \n",
       "25%            8           9                1         1        1970  \n",
       "50%           16          15                3         1        1980  \n",
       "75%           23          18                4         1        1987  \n",
       "max           31          23                6         2        2000  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.describe().astype(int)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Create X and y for sklearn with non-subscriber as class=1.0\n",
    "X = data[data.columns[:-1]].astype(np.float)\n",
    "y = (data['usertype'] != 'Subscriber').values.astype(np.int)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(12222976, 12)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(12222976,)"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SMOTE Demonstration"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<b>Class frequency in the original dataset</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==== Before SMOTE ====\n",
      "Total number of Rides: 12222976\n",
      "Total number of Rides by Subscribers: 12188588 (99.72% of total rides)\n",
      "Total number of Rides by Non-subscribers: 34388 (0.28% of total rides)\n"
     ]
    }
   ],
   "source": [
    "print('==== Before SMOTE ====')\n",
    "total_number_rides = len(y)\n",
    "total_number_subscribers = Counter(y)[0]\n",
    "total_number_nonsubscribers = Counter(y)[1]\n",
    "subs_perc = 100*total_number_subscribers / total_number_rides\n",
    "nonsubs_perc = 100*total_number_nonsubscribers / total_number_rides\n",
    "print('Total number of Rides: %d' % total_number_rides)\n",
    "print('Total number of Rides by Subscribers: %d (%.2f%% of total rides)' % (total_number_subscribers, subs_perc))\n",
    "print('Total number of Rides by Non-subscribers: %d (%.2f%% of total rides)' % (total_number_nonsubscribers, nonsubs_perc))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<b>Class frequency in after SMOTE</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1min 46s, sys: 2.2 s, total: 1min 49s\n",
      "Wall time: 1min 49s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# run SMOTE algorithm\n",
    "sm = SMOTE()\n",
    "X_res, y_res = sm.fit_sample(X, y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==== After SMOTE ====\n",
      "Total number of Rides: 24377176\n",
      "Total number of Rides by Subscribers: 12188588 (50.00% of total rides)\n",
      "Total number of Rides by Non-subscribers: 12188588 (50.00% of total rides)\n"
     ]
    }
   ],
   "source": [
    "print('==== After SMOTE ====')\n",
    "total_number_rides = len(y_res)\n",
    "total_number_subscribers = Counter(y_res)[0]\n",
    "total_number_nonsubscribers = Counter(y_res)[1]\n",
    "subs_perc = 100*total_number_subscribers / total_number_rides\n",
    "nonsubs_perc = 100*total_number_nonsubscribers / total_number_rides\n",
    "print('Total number of Rides: %d' % total_number_rides)\n",
    "print('Total number of Rides by Subscribers: %d (%.2f%% of total rides)' % (total_number_subscribers, subs_perc))\n",
    "print('Total number of Rides by Non-subscribers: %d (%.2f%% of total rides)' % (total_number_nonsubscribers, nonsubs_perc))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train on oversampled set, test on regular set (multiple rounds)\n",
    "- First, split the dataset into train and test sets\n",
    "- Oversample the minority by SMOTE on the train set only\n",
    "- Train the classifier on the oversampled train set, test on the imbalanced test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Average accuracy: 0.996889 +/- 0.000033 (n=3)\n",
      "Average AP Score: 0.099403 +/- 0.002113 (n=3)\n",
      "CPU times: user 1h 4min 16s, sys: 56.4 s, total: 1h 5min 12s\n",
      "Wall time: 22min 57s\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmQAAAEyCAYAAABULszLAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nzMzMzHLmhszMzMwsZ/8PXoXd+17mP0cAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10519b6a0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "# Set up basic parameters and data containers\n",
    "n = 3\n",
    "test_size = 0.4\n",
    "accuracies = []\n",
    "aps = []\n",
    "\n",
    "fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(10, 4.5))\n",
    "\n",
    "# Split the sample first\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)\n",
    "\n",
    "for i in range(n):\n",
    "    \n",
    "    # Undersample the majority for train set only\n",
    "    sm = SMOTE()\n",
    "    X_res, y_res = sm.fit_sample(X_train, y_train)\n",
    "    \n",
    "    # Train RandomForest on Undersampled train set\n",
    "    clf = RandomForestClassifier(max_depth=40, n_jobs=4)\n",
    "    clf.fit(X_res, y_res)\n",
    "    \n",
    "    # Predict/score on X_test and y_test\n",
    "    result = msf.score_clf(clf, X_test, y_test, X_train=X_res, y_train=y_res)\n",
    "    accuracies.append(result['scores']['accuracy'])\n",
    "    aps.append(result['scores']['average_precision_score'])\n",
    "    \n",
    "    # Plot ROC curve\n",
    "    fpr = result['scores']['roc_curve'][0]\n",
    "    tpr = result['scores']['roc_curve'][1]\n",
    "    ax1.plot(fpr, tpr)\n",
    "    ax1.set_xlabel('False Positive Rate')\n",
    "    ax1.set_ylabel('True Positive Rate')\n",
    "    ax1.set_title('ROC Curve')\n",
    "    \n",
    "    # Plot Precision-Recall Curve\n",
    "    precision, recall, _ = precision_recall_curve(y_test, clf.predict_proba(X_test)[:, 1])\n",
    "\n",
    "    ax2.plot(recall, precision)\n",
    "    ax2.set_xlabel('Recall')\n",
    "    ax2.set_ylabel('Precision')\n",
    "    ax2.set_title('Precision-Recall')\n",
    "\n",
    "print('Average accuracy: %f +/- %f (n=%d)' % (np.mean(accuracies), np.std(accuracies), n))\n",
    "print('Average AP Score: %f +/- %f (n=%d)' % (np.mean(aps), np.std(aps), n))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "result = msf.score_clf2(clf, X_test, y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==== Classifier Metrics on Test Set ====\n",
      "Accuracy: 0.996935\n",
      "Precision: 0.302592\n",
      "Recall: 0.065088\n",
      "F-Score: 0.107132\n",
      "Brier Score: 0.003246\n",
      "Average Precision Score: 0.102387\n",
      "Confusion Matrix:  [[4873307    2072]\n",
      " [  12913     899]]\n",
      "--------------- Feature Importances ---------------\n",
      " 1) start_month          0.417815\n",
      " 2) start_dayofweek      0.143318\n",
      " 3) gender               0.124208\n",
      " 4) tripduration         0.109225\n",
      " 5) start_hour           0.100156\n",
      " 6) birth year           0.033727\n",
      " 7) start_day            0.031869\n",
      " 8) end station longitude 0.011647\n",
      " 9) start station longitude 0.009770\n",
      "10) end station latitude 0.009576\n",
      "11) start station latitude 0.008688\n",
      "12) start_year           0.000000\n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "feature_labels = data.columns[:-1]\n",
    "msf.print_clf_scores2(result, trainset=False, feature_labels=feature_labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}