project_bikesharing.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# This tells matplotlib not to try opening a new window for each plot.\n",
    "%matplotlib inline\n",
    "\n",
    "# General libraries.\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "# SK-learn libraries for learning.\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.neighbors import KNeighborsRegressor\n",
    "from sklearn.tree import DecisionTreeRegressor\n",
    "from sklearn.ensemble import RandomForestRegressor\n",
    "from sklearn.ensemble import AdaBoostRegressor\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from sklearn.svm import SVR\n",
    "\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from sklearn.linear_model import Ridge\n",
    "from sklearn.linear_model import Lasso\n",
    "\n",
    "from sklearn.grid_search import GridSearchCV\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "# SK-learn libraries for evaluation.\n",
    "from sklearn import metrics\n",
    "from sklearn.metrics import make_scorer\n",
    "\n",
    "# Set randomizer seed\n",
    "np.random.seed(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Feature names: \n",
      "['year' 'dayofyear' 'hour' 'season' 'holiday' 'workingday' 'weather' 'temp'\n",
      " 'atemp' 'humidity' 'windspeed']\n",
      "\n",
      "First 5 examples: \n",
      "             datetime  year  dayofyear  hour  season  holiday  workingday  \\\n",
      "0 2011-01-01 00:00:00  2011          1     0       1        0           0   \n",
      "1 2011-01-01 01:00:00  2011          1     1       1        0           0   \n",
      "2 2011-01-01 02:00:00  2011          1     2       1        0           0   \n",
      "3 2011-01-01 03:00:00  2011          1     3       1        0           0   \n",
      "4 2011-01-01 04:00:00  2011          1     4       1        0           0   \n",
      "\n",
      "   weather  temp   atemp  humidity  windspeed  casual  registered  count  \n",
      "0        1  9.84  14.395        81     6.0032       3          13     16  \n",
      "1        1  9.02  13.635        80     6.0032       8          32     40  \n",
      "2        1  9.02  13.635        80     6.0032       5          27     32  \n",
      "3        1  9.84  14.395        75     6.0032       3          10     13  \n",
      "4        1  9.84  14.395        75     6.0032       0           1      1  \n",
      "All train data: (10886, 11)\n",
      "Train data: (9500, 11)\n",
      "Dev data: (1386, 11)\n",
      "Test data: (6493, 11)\n"
     ]
    }
   ],
   "source": [
    "## Prepare data for model fitting\n",
    "\n",
    "# Train and test sets on local machines\n",
    "file_paths = ['~/datasets/bike/train.csv', '~/datasets/bike/test.csv']\n",
    "\n",
    "# Define function to clean train and test data\n",
    "def clean_data(path):\n",
    "    \"\"\"\n",
    "    Input: path of csv files (train or test)\n",
    "    Output: cleaned dataframe\n",
    "    \"\"\"\n",
    "    df = pd.read_csv(path, parse_dates=[0])\n",
    "    # Split datetime up into 3 components (year, dayofyear, hour)\n",
    "    df.insert(1, \"year\", df[\"datetime\"].dt.year)\n",
    "    df.insert(2, \"dayofyear\", df[\"datetime\"].dt.dayofyear)\n",
    "    df.insert(3, \"hour\", df[\"datetime\"].dt.hour)    \n",
    "    # Fill in missing values in humidity and windspeed\n",
    "    df.loc[df[\"humidity\"]==0, \"humidity\"] = np.nan\n",
    "    df.loc[df[\"windspeed\"]==0, \"windspeed\"] = np.nan    \n",
    "    # Fill forward first, then, backwards (for first few missing values)\n",
    "    df = df.fillna(method=\"ffill\").fillna(method=\"bfill\")\n",
    "    return df\n",
    "\n",
    "# Clean train and test csv files\n",
    "df_train = clean_data(file_paths[0])\n",
    "df_test = clean_data(file_paths[1])\n",
    "\n",
    "# Get train feature names (excluding datetime, casual, registered, count)\n",
    "print (\"Feature names: \")\n",
    "print (df_train.columns.values[1:-3])\n",
    "# Print out first 5 training examples\n",
    "print (\"\\nFirst 5 examples: \")\n",
    "print (df_train.head())\n",
    "\n",
    "# Extract numerical features\n",
    "num_data = df_train.iloc[:, 1:].values\n",
    "X, Y = num_data[:, :-3], num_data[:, -1]\n",
    "test_data = df_test.iloc[:, 1:].values\n",
    "\n",
    "# Split all training data into train set and dev set (about 7:1 ratio)\n",
    "shuffle = np.random.permutation(np.arange(X.shape[0]))\n",
    "X, Y = X[shuffle], Y[shuffle]\n",
    "print (\"All train data: {}\".format(X.shape))\n",
    "train_data, train_labels = X[:9500], Y[:9500]\n",
    "dev_data, dev_labels = X[9500:], Y[9500:]\n",
    "\n",
    "# Array sizes\n",
    "print (\"Train data: {}\".format(train_data.shape))\n",
    "print (\"Dev data: {}\".format(dev_data.shape))\n",
    "print (\"Test data: {}\".format(test_data.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x10af38b50>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmMAAAJQCAYAAADCJmN+AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAACAjWjGAAAAbPT/ASFoqrl1SDdoAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x109d24990>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Histograms for cetain variables\n",
    "plt.figure(figsize=(10,10))\n",
    "plt.subplot(321)\n",
    "df_train[\"weather\"].hist()\n",
    "plt.subplot(322)\n",
    "df_train[\"temp\"].hist(bins=20)\n",
    "plt.subplot(323)\n",
    "df_train[\"atemp\"].hist(bins=20)\n",
    "plt.subplot(324)\n",
    "df_train[\"humidity\"].hist(bins=20) # missing values\n",
    "plt.subplot(325)\n",
    "df_train[\"windspeed\"].hist(bins=20) # missing values\n",
    "plt.subplot(326)\n",
    "df_train[\"count\"].hist(bins=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# Evaluation: Root Mean Squared Logarithmic Error (RMSLE)\n",
    "def RMSLE(actual, predicted):\n",
    "    \"\"\"\n",
    "    Inputs: actual counts, predicted counts\n",
    "    Outputs: Root Mean Squared Logarithmic Error\n",
    "    \"\"\"\n",
    "    return np.sqrt(np.mean(np.square(np.log(predicted+1) - np.log(actual+1))))\n",
    "\n",
    "# Make a scoring object for GridSearch to use\n",
    "RMSLE_scorer = make_scorer(RMSLE, greater_is_better=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE(baseline): 1.5629246036\n",
      "RMSLE: 1.28782961803\n",
      "R-square on dev: 0.410914367472\n"
     ]
    }
   ],
   "source": [
    "# Linear Regression\n",
    "# Fit a simple regression model using all training data\n",
    "lr_model = LinearRegression()\n",
    "lr_model.fit(train_data, train_labels)\n",
    "preds = lr_model.predict(dev_data)\n",
    "\n",
    "# Compute baseline score (using the mean as the base model)\n",
    "base_preds = np.repeat(np.mean(train_labels), dev_labels.shape)\n",
    "print (\"RMSLE(baseline): {}\".format(RMSLE(dev_labels, base_preds)))\n",
    "\n",
    "# Now try linear regression model\n",
    "# Predictions cannot be negative, convert negative predictions to zero\n",
    "preds[preds<0] = 0\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, preds)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 1.23488945278\n",
      "R-square on dev: 0.396515305504\n"
     ]
    }
   ],
   "source": [
    "# Linear Regression with L2 Regularization\n",
    "# Fit ridge regression model\n",
    "lr_ridge_model = Ridge(alpha=2400)\n",
    "lr_ridge_model.fit(train_data, train_labels)\n",
    "preds = lr_ridge_model.predict(dev_data)\n",
    "\n",
    "# Predictions cannot be negative, convert negative predictions to zero\n",
    "preds[preds<0] = 0\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, preds)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 1.24022166193\n",
      "R-square on dev: 0.400560310343\n"
     ]
    }
   ],
   "source": [
    "# Linear Regression with L1 Regularization\n",
    "# Fit lasso regression model\n",
    "lr_lasso_model = Lasso(alpha=8)\n",
    "lr_lasso_model.fit(train_data, train_labels)\n",
    "preds = lr_lasso_model.predict(dev_data)\n",
    "\n",
    "# Predictions cannot be negative, convert negative predictions to zero\n",
    "preds[preds<0] = 0\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, preds)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## Feature Scaling\n",
    "scaler = StandardScaler()\n",
    "train_data_scaled = scaler.fit_transform(train_data)\n",
    "dev_data_scaled = scaler.transform(dev_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 0.713312406042\n",
      "R-square on dev: 0.680080041734\n"
     ]
    }
   ],
   "source": [
    "## kNN Regression\n",
    "k = 2\n",
    "knn_model = KNeighborsRegressor(n_neighbors = k, weights = \"distance\")\n",
    "knn_model.fit(train_data_scaled, train_labels)\n",
    "knn_preds = knn_model.predict(dev_data_scaled)\n",
    "\n",
    "# Predictions cannot be negative, convert negative predictions to zero\n",
    "knn_preds[knn_preds<0] = 0\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, knn_preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, knn_preds)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 0.370972433814\n",
      "R-square on dev: 0.917695434478\n",
      "[ 0.099  0.102  0.616  0.     0.001  0.109  0.011  0.014  0.024  0.022\n",
      "  0.001]\n"
     ]
    }
   ],
   "source": [
    "## Decision Trees\n",
    "# Ensure consistency of results\n",
    "np.random.seed(0)\n",
    "\n",
    "# Fit decision tree regressor\n",
    "dt_model = DecisionTreeRegressor(max_depth=10)\n",
    "dt_model.fit(train_data_scaled, train_labels)\n",
    "dt_preds = dt_model.predict(dev_data_scaled)\n",
    "\n",
    "# RMSLE\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, dt_preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, dt_preds)))\n",
    "print (np.round(dt_model.feature_importances_,3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-0.434502158435\n",
      "{'max_depth': 9}\n"
     ]
    }
   ],
   "source": [
    "# Decision Tree Grid Search\n",
    "dt_params = {\"max_depth\": np.arange(1,30)}\n",
    "dt_grid = GridSearchCV(DecisionTreeRegressor(), dt_params, scoring=RMSLE_scorer)\n",
    "dt_grid.fit(train_data_scaled, train_labels)\n",
    "\n",
    "print (dt_grid.best_score_)\n",
    "print (dt_grid.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA64AAAHpCAYAAACCxGGPAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nBFcAAAAm7f8B7oc08P8RBoIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1317104d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(16,8))\n",
    "ax = sns.barplot(x=df_train.columns.values[1:-3], y=dt_model.feature_importances_)\n",
    "title = plt.title(\"Feature Importances\")\n",
    "ylabel = plt.ylabel(\"Gini Importance\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 0.324199932671\n",
      "R-square on dev: 0.950496040476\n",
      "[ 0.089  0.088  0.593  0.002  0.003  0.086  0.015  0.055  0.029  0.032\n",
      "  0.009]\n"
     ]
    }
   ],
   "source": [
    "## Random Forest Regressor\n",
    "# Ensure consistency of results\n",
    "np.random.seed(0)\n",
    "\n",
    "# Fit random forest regressor\n",
    "rf_model = RandomForestRegressor(n_estimators = 50, max_features = 11)\n",
    "rf_model.fit(train_data_scaled, train_labels)\n",
    "rf_preds = rf_model.predict(dev_data_scaled)\n",
    "\n",
    "# RMSLE\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, rf_preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, rf_preds)))\n",
    "print (np.round(rf_model.feature_importances_,3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-0.363770401116\n",
      "{'n_estimators': 80}\n"
     ]
    }
   ],
   "source": [
    "# Random Forest Grid Search\n",
    "# Ensure consistency of results\n",
    "np.random.seed(0)\n",
    "rf_params = {\"n_estimators\": np.arange(1,101)}\n",
    "rf_grid = GridSearchCV(RandomForestRegressor(), rf_params, scoring=RMSLE_scorer)\n",
    "rf_grid.fit(train_data_scaled, train_labels)\n",
    "\n",
    "print (rf_grid.best_score_)\n",
    "print (rf_grid.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA7sAAAH4CAYAAACLyfodAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x109c7b690>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Random forest plot\n",
    "accuracies = [-rf_grid.grid_scores_[i][1] for i in range(100)]\n",
    "plt.figure(figsize=(16,8))\n",
    "scatter = plt.plot(rf_params[\"n_estimators\"], accuracies, 'bo')\n",
    "title = plt.title (\"Random Forest (m=11)\")\n",
    "xlabel = plt.xlabel (\"Number of Estimators\")\n",
    "ylabel = plt.ylabel (\"RMSLE\")\n",
    "# Looks maxed out around n=30"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 0.314907966266\n",
      "R-square on dev: 0.950630350674\n",
      "[ 0.094  0.093  0.488  0.003  0.007  0.09   0.029  0.05   0.066  0.062\n",
      "  0.018]\n"
     ]
    }
   ],
   "source": [
    "## Adaboost with Decision Tree Regressor\n",
    "# Ensure consistency of results\n",
    "np.random.seed(0)\n",
    "\n",
    "# Fit Adaboost with decision tree regressor\n",
    "ada_model = AdaBoostRegressor(base_estimator=DecisionTreeRegressor(max_depth=20), n_estimators=30, loss=\"linear\")\n",
    "ada_model.fit(train_data_scaled, train_labels)\n",
    "ada_preds = ada_model.predict(dev_data_scaled)\n",
    "\n",
    "# RMSLE\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, ada_preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, ada_preds)))\n",
    "print (np.round(ada_model.feature_importances_,3))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 1.02011780053\n",
      "R-square on dev: 0.420535656001\n"
     ]
    }
   ],
   "source": [
    "## SVM Regression\n",
    "# Ensure consistency of results\n",
    "np.random.seed(0)\n",
    "\n",
    "# Fit SVM\n",
    "svm_model = SVR(C=1.5)\n",
    "svm_model.fit(train_data_scaled, train_labels)\n",
    "svm_preds = svm_model.predict(dev_data_scaled)\n",
    "\n",
    "# Predictions cannot be negative, convert negative predictions to zero\n",
    "svm_preds[svm_preds<0] = 0\n",
    "\n",
    "# RMSLE\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, svm_preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, svm_preds)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RMSLE: 0.318029615747\n",
      "R-square on dev: 0.951649235839\n"
     ]
    }
   ],
   "source": [
    "## Gradient Boosting with Decision Tree Regressor\n",
    "# Ensure consistency of results\n",
    "np.random.seed(0)\n",
    "\n",
    "# Fit Adaboost with decision tree regressor\n",
    "gb_model = GradientBoostingRegressor(loss='lad', max_depth=9, n_estimators=125)\n",
    "gb_model.fit(train_data_scaled, train_labels)\n",
    "gb_preds = gb_model.predict(dev_data_scaled)\n",
    "\n",
    "# Predictions cannot be negative, convert negative predictions to zero\n",
    "gb_preds[gb_preds<0] = 0\n",
    "\n",
    "# RMSLE\n",
    "print (\"RMSLE: {}\".format(RMSLE(dev_labels, gb_preds)))\n",
    "\n",
    "# R2 for dev set\n",
    "print (\"R-square on dev: {}\".format(metrics.r2_score(dev_labels, gb_preds)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "## Write results in Kaggle format\n",
    "# Scale ALL train and test data\n",
    "scaler = StandardScaler()\n",
    "X_scaled = scaler.fit_transform(X)\n",
    "test_data_scaled = scaler.transform(test_data)\n",
    "\n",
    "# Predict using Adaboost (RF: 0.49, ADA: 0.48)\n",
    "ada_model.fit(X_scaled, Y)\n",
    "kaggle_preds = ada_model.predict(test_data_scaled)\n",
    "\n",
    "# Write to csv\n",
    "out_path = '/Users/InfernoIX/datasets/bike/kaggle_submission.csv'\n",
    "df_submission = pd.DataFrame({\"datetime\": df_test[\"datetime\"], \"count\": kaggle_preds}, columns = [\"datetime\", \"count\"])\n",
    "df_submission.to_csv(out_path, index=False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}