{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import pandas\n",
    "\n",
    "bike_rentals = pandas.read_csv(\"bike_rental_hour.csv\")\n",
    "bike_rentals.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.hist(bike_rentals[\"cnt\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "bike_rentals.corr()[\"cnt\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def assign_label(hour):\n",
    "    if hour >=0 and hour < 6:\n",
    "        return 4\n",
    "    elif hour >=6 and hour < 12:\n",
    "        return 1\n",
    "    elif hour >= 12 and hour < 18:\n",
    "        return 2\n",
    "    elif hour >= 18 and hour <=24:\n",
    "        return 3\n",
    "\n",
    "bike_rentals[\"time_label\"] = bike_rentals[\"hr\"].apply(assign_label)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Error metric\n",
    "\n",
    "The mean squared error metric makes the most sense to evaluate our error.  MSE works on continuous numeric data, which fits our data quite well."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "train = bike_rentals.sample(frac=.8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "test = bike_rentals.loc[~bike_rentals.index.isin(train.index)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LinearRegression\n",
    "\n",
    "predictors = list(train.columns)\n",
    "predictors.remove(\"cnt\")\n",
    "predictors.remove(\"casual\")\n",
    "predictors.remove(\"registered\")\n",
    "predictors.remove(\"dteday\")\n",
    "\n",
    "reg = LinearRegression()\n",
    "\n",
    "reg.fit(train[predictors], train[\"cnt\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import numpy\n",
    "predictions = reg.predict(test[predictors])\n",
    "\n",
    "numpy.mean((predictions - test[\"cnt\"]) ** 2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "actual"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "test[\"cnt\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Error\n",
    "\n",
    "The error is very high, which may be due to the fact that the data has a few extremely high rental counts, but otherwise mostly low counts.  Larger errors are penalized more with MSE, which leads to a higher total error."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from sklearn.tree import DecisionTreeRegressor\n",
    "\n",
    "reg = DecisionTreeRegressor(min_samples_leaf=5)\n",
    "\n",
    "reg.fit(train[predictors], train[\"cnt\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "predictions = reg.predict(test[predictors])\n",
    "\n",
    "numpy.mean((predictions - test[\"cnt\"]) ** 2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "reg = DecisionTreeRegressor(min_samples_leaf=2)\n",
    "\n",
    "reg.fit(train[predictors], train[\"cnt\"])\n",
    "\n",
    "predictions = reg.predict(test[predictors])\n",
    "\n",
    "numpy.mean((predictions - test[\"cnt\"]) ** 2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decision tree error\n",
    "\n",
    "By taking the nonlinear predictors into account, the decision tree regressor appears to have much higher accuracy than linear regression."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import RandomForestRegressor\n",
    "\n",
    "reg = RandomForestRegressor(min_samples_leaf=5)\n",
    "reg.fit(train[predictors], train[\"cnt\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "predictions = reg.predict(test[predictors])\n",
    "\n",
    "numpy.mean((predictions - test[\"cnt\"]) ** 2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random forest error\n",
    "\n",
    "By removing some of the sources of overfitting, the random forest accuracy is improved over the decision tree accuracy."
   ]
  }
 ],
 "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": "3.5.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}