07. Regression.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predicting Graduation Rates for Public High Schools\n",
    "**This notebook explores predicting graduation rates for public high schools in North Carolina.** \n",
    "* Public school student body racial compositions are also considered when making predictions in this notebook.\n",
    "* The North Carolina Educational Attainment Data Repository for Machine Learning is located on Github at: https://github.com/jakemdrew/EducationDataNC \n",
    "\n",
    "**For documentation on various Generalized Linear Models in Sklearn see:**\n",
    "* http://scikit-learn.org/stable/modules/linear_model.html\n",
    "* https://stackoverflow.com/questions/33845539/modelling-probabilities-in-a-regularized-logistic-regression-model-in-python"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 603 entries, 0 to 602\n",
      "Columns: 639 entries, AsianPct to agency_code\n",
      "dtypes: float64(377), int64(261), object(1)\n",
      "memory usage: 2.9+ MB\n"
     ]
    }
   ],
   "source": [
    "#import required Libraries\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "\n",
    "#Read in the high schools dataset - NC EA Public High Schools Dataset Feb 2018\n",
    "\n",
    "#The 'path' variable may also be a URL pointing to the proper raw file path on github: \n",
    "#path=\"https://raw.githubusercontent.com/jakemdrew/EducationDataNC/master/2018/Machine%20Learning%20Datasets/PublicHighSchools2018_ML.csv\"\n",
    "\n",
    "path=\"C:/Users/Jake/Documents/GitHub/EducationDataNC/2018/Machine Learning Datasets/PublicHighSchools2018_ML.csv\"\n",
    "schData=pd.read_csv(path, low_memory=False, dtype={'agency_code': object})\n",
    "\n",
    "#inspect data \n",
    "schData.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Check ML file after adding pct_STD_ALL_CGR:\n",
      "*******************************************\n",
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 603 entries, 0 to 602\n",
      "Columns: 640 entries, AsianPct to pct_STD_ALL_CGR\n",
      "dtypes: float64(378), int64(261), object(1)\n",
      "memory usage: 2.9+ MB\n"
     ]
    }
   ],
   "source": [
    "#Get 4 year graduation rates from the public school file. \n",
    "#4yr was removed from the ML file for correlation > 95% with other graduation rate fields\n",
    "path=\"C:/Users/Jake/Documents/GitHub/EducationDataNC/2018/School Datasets/PublicHighSchools2018.csv\"\n",
    "schDataALL=pd.read_csv(path, low_memory=False, dtype={'agency_code': object})\n",
    "\n",
    "schDataALL=schDataALL[['agency_code','pct_STD_ALL_CGR']]\n",
    "\n",
    "schData=schData.merge(schDataALL,how='left',on='agency_code', suffixes=('', '_Drop'))\n",
    "\n",
    "print(\"Check ML file after adding pct_STD_ALL_CGR:\")\n",
    "print(\"*******************************************\")\n",
    "schData.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['pct_EXT_AIG_CGR', 'pct_EXT_ALL_CGR', 'pct_EXT_BL7_CGR',\n",
       "       'pct_EXT_EDS_CGR', 'pct_EXT_FEM_CGR', 'pct_EXT_HI7_CGR',\n",
       "       'pct_EXT_MALE_CGR', 'pct_EXT_SWD_CGR', 'pct_EXT_WH7_CGR',\n",
       "       'pct_STD_AIG_CGR', 'pct_STD_BL7_CGR', 'pct_STD_EDS_CGR',\n",
       "       'pct_STD_FEM_CGR', 'pct_STD_HI7_CGR', 'pct_STD_MALE_CGR',\n",
       "       'pct_STD_SWD_CGR', 'pct_STD_WH7_CGR',\n",
       "       'pct_met_4 Year Graduation Rate_LTG', 'cgrs_score',\n",
       "       'cgrs_score_ALL_SPG2', 'cgrs_score_BL7_SPG2', 'cgrs_score_EDS_SPG2',\n",
       "       'cgrs_score_WH7_SPG2',\n",
       "       'target_list_4 Year Graduation Rate,EL Progress,Math Grade 11,Math Grades 3-8,Reading Grade 10,Reading Grades 3-8',\n",
       "       'target_list_4 Year Graduation Rate,EL Progress,Math Grade 11,Reading Grade 10',\n",
       "       'target_list_4 Year Graduation Rate,Math Grade 11',\n",
       "       'target_list_4 Year Graduation Rate,Math Grade 11,Math Grades 3-8,Reading Grade 10,Reading Grades 3-8',\n",
       "       'target_list_4 Year Graduation Rate,Math Grade 11,Math Grades 3-8,Reading Grades 3-8',\n",
       "       'target_list_4 Year Graduation Rate,Math Grade 11,Reading Grade 10',\n",
       "       'target_list_4 Year Graduation Rate,Math Grades 3-8,Reading Grades 3-8',\n",
       "       'target_list_4 Year Graduation Rate,Reading Grade 10',\n",
       "       'pct_STD_ALL_CGR'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Remove any columns from the model's training data that are directly related to the graduation rate. \n",
    "GradRateCols = schData.columns[schData.columns.str.lower().str.contains('graduation') |\n",
    "                               schData.columns.str.lower().str.contains('_cgr') |\n",
    "                               schData.columns.str.lower().str.contains('cgr_') |\n",
    "                               schData.columns.str.lower().str.contains('_cgrs') |\n",
    "                               schData.columns.str.lower().str.contains('cgrs_') ]\n",
    "GradRateCols"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### This model only considers public high schools which are reporting both a 4 and 5 year graduation rate. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 536 entries, 0 to 602\n",
      "Columns: 640 entries, AsianPct to pct_STD_ALL_CGR\n",
      "dtypes: float64(378), int64(261), object(1)\n",
      "memory usage: 2.6+ MB\n"
     ]
    }
   ],
   "source": [
    "#Remove schools with graduation rates == 0 from the model\n",
    "schData = schData[schData['pct_STD_ALL_CGR'] > 0]\n",
    "schData = schData[schData['pct_EXT_ALL_CGR'] > 0]\n",
    "schData.info()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 536 entries, 0 to 602\n",
      "Columns: 608 entries, AsianPct to agency_code\n",
      "dtypes: float64(354), int64(253), object(1)\n",
      "memory usage: 2.5+ MB\n"
     ]
    }
   ],
   "source": [
    "# create x explanatory and y response variables for regression\n",
    "Y_4yr_Gr = schData['pct_STD_ALL_CGR']\n",
    "Y_5yr_Gr = schData['pct_EXT_ALL_CGR']\n",
    "\n",
    "#Delete the response variable and other graduation rate data columns from our training dataset\n",
    "X_highSchools = schData.drop(GradRateCols, axis=1)\n",
    "\n",
    "Y = Y_4yr_Gr\n",
    "\n",
    "#inspect data \n",
    "X_highSchools.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Cross Validation\n",
    "**Cross validation is performed using repeated holdout using ShuffleSplit()**\n",
    "* Ten folds are used\n",
    "* The split is: 90% training data and 10% test data\n",
    "* A random seed is set so the same random test and training splits are used each time cross validation is performed."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Divide data into test and training splits\n",
    "from sklearn.model_selection import ShuffleSplit\n",
    "cv = ShuffleSplit(n_splits=10, test_size=0.10, random_state=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Custom Scorers for Evaluating Regression Models \n",
    "\n",
    "**All regression models created in this notebook are validated using the following metrics:**\n",
    "* Mean Absolute Error (MAE)\n",
    "* Root Mean Squared Error (RMSE) - https://stackoverflow.com/questions/17197492/root-mean-square-error-in-python\n",
    "* Mean Absolute Percentage Error (MAPE) - https://stackoverflow.com/questions/42250958/how-to-optimize-mape-code-in-python\n",
    "\n",
    "**For details on making scorers to return multiple mean error scores see:**\n",
    "* http://scikit-learn.org/stable/auto_examples/model_selection/plot_multi_metric_evaluation.html\n",
    "* https://github.com/scikit-learn/scikit-learn/pull/7388\n",
    "* https://github.com/drorata/multiscorer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Use mean absolute error (MAE) to score the regression models created \n",
    "#(the scale of MAE is identical to the response variable)\n",
    "from sklearn.metrics import mean_absolute_error, make_scorer, mean_squared_error\n",
    "\n",
    "#Function for Root mean squared error\n",
    "#https://stackoverflow.com/questions/17197492/root-mean-square-error-in-python\n",
    "def rmse(y_actual, y_predicted):\n",
    "    return np.sqrt(mean_squared_error(y_actual, y_predicted))\n",
    "\n",
    "#Function for Mean Absolute Percentage Error (MAPE) - Untested\n",
    "#Adapted from - https://stackoverflow.com/questions/42250958/how-to-optimize-mape-code-in-python\n",
    "def mape(y_actual, y_predicted): \n",
    "    mask = y_actual != 0\n",
    "    return (np.fabs(y_actual - y_predicted)/y_actual)[mask].mean() * 100\n",
    "\n",
    "#Create scorers for rmse and mape functions\n",
    "mae_scorer = make_scorer(score_func=mean_absolute_error, greater_is_better=False)\n",
    "rmse_scorer = make_scorer(score_func=rmse, greater_is_better=False)\n",
    "mape_scorer = make_scorer(score_func=mape, greater_is_better=False)\n",
    "\n",
    "#Make scorer array to pass into cross_validate() function for producing mutiple scores for each cv fold.\n",
    "errorScoring = {'MAE':  mae_scorer, \n",
    "                'RMSE': rmse_scorer,\n",
    "                'MAPE': mape_scorer\n",
    "               } "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Regression Model Evaluation\n",
    "** All regression models are evaluated using the regression model evaluation function below: ** \n",
    "* The following regression evaluation function uses the cross validation object and the custom scorers in the two cells above in combination with sklearn.model_selection's cross_validate function to perform cross validation for regression estimators.\n",
    "* The cross validation object above uses a random seed to ensure that all regression estimators are tested on the same randomly selected records for each cross validation fold.\n",
    "* Custom scorers are created using the three chosen mean error scores and passed into cross_validate(), so all three scores are calcualted using a single call to cross_validate().\n",
    "* All of this functionality is wrapped within the custom EvaluateRegressionEstimator() function below so multiple regression models may be tested using the same test / train cv data and evaluation scores producing a consistent output for each model without the need to re-write the same code over and over. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import cross_validate\n",
    "\n",
    "def EvaluateRegressionEstimator(regEstimator, X, y, cv):\n",
    "    \n",
    "    scores = cross_validate(regEstimator, X, y, scoring=errorScoring, cv=cv, return_train_score=True)\n",
    "\n",
    "    #cross val score sign-flips the outputs of MAE\n",
    "    # https://github.com/scikit-learn/scikit-learn/issues/2439\n",
    "    scores['test_MAE'] = scores['test_MAE'] * -1\n",
    "    scores['test_MAPE'] = scores['test_MAPE'] * -1\n",
    "    scores['test_RMSE'] = scores['test_RMSE'] * -1\n",
    "\n",
    "    #print mean MAE for all folds \n",
    "    maeAvg = scores['test_MAE'].mean()\n",
    "    print_str = \"The average MAE for all cv folds is: \\t\\t\\t {maeAvg:.5}\"\n",
    "    print(print_str.format(maeAvg=maeAvg))\n",
    "\n",
    "    #print mean test_MAPE for all folds\n",
    "    scores['test_MAPE'] = scores['test_MAPE']\n",
    "    mape_avg = scores['test_MAPE'].mean()\n",
    "    print_str = \"The average MAE percentage (MAPE) for all cv folds is: \\t {mape_avg:.5}\"\n",
    "    print(print_str.format(mape_avg=mape_avg))\n",
    "\n",
    "    #print mean MAE for all folds \n",
    "    RMSEavg = scores['test_RMSE'].mean()\n",
    "    print_str = \"The average RMSE for all cv folds is: \\t\\t\\t {RMSEavg:.5}\"\n",
    "    print(print_str.format(RMSEavg=RMSEavg))\n",
    "    print('*********************************************************')\n",
    "\n",
    "    print('Cross Validation Fold Mean Error Scores')\n",
    "    scoresResults = pd.DataFrame()\n",
    "    scoresResults['MAE'] = scores['test_MAE']\n",
    "    scoresResults['MAPE'] = scores['test_MAPE']\n",
    "    scoresResults['RMSE'] = scores['test_RMSE']\n",
    "    return scoresResults\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Modifying or Making Custom Estimators\n",
    "\n",
    "**Linear Regression is used to create a baseline model here.  Since linear regression may predict response variable values outside the range of the training data's response variable, we create a custom linear regression estimator with graduation rate predictions clipped between 0% and 100%.** \n",
    "\n",
    "**For more details see:**\n",
    "* http://scikit-learn.org/stable/developers/contributing.html#rolling-your-own-estimator \n",
    "* https://github.com/scikit-learn/scikit-learn/issues/6950\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LinearRegression.html\n",
    "* https://github.com/scikit-learn-contrib/project-template/blob/master/skltemplate/template.py\n",
    "* https://stackoverflow.com/questions/44234682/how-to-use-sklearn-when-target-variable-is-a-proportion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Make new estimator compatible for use with GridSearchCV() and cross_validate()\n",
    "# -  Cap predict function for LinearRegression between 0 and 100\n",
    "# -  See: Roll your own estimator links above for details. \n",
    "from sklearn.base import BaseEstimator, RegressorMixin\n",
    "from sklearn.linear_model import LinearRegression\n",
    "\n",
    "class CappedLinearRegression(LinearRegression):\n",
    "\n",
    "    def predict(self, X):\n",
    "        return np.clip(super(CappedLinearRegression, self).predict(X), 0, 100) "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Linear Regression Grid Search\n",
    "** Here we perform a grid search testing 40 models to find the best parameters for our Linear Regression model based on Mean Absolute Error.  See more on parameter tuning with grid search here:**\n",
    "* http://scikit-learn.org/stable/modules/grid_search.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 4 candidates, totalling 40 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=1)]: Using backend SequentialBackend with 1 concurrent workers.\n",
      "[Parallel(n_jobs=1)]: Done  40 out of  40 | elapsed:    3.4s finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=CappedLinearRegression(copy_X=True, fit_intercept=True, n_jobs=None,\n",
       "            normalize=False),\n",
       "       fit_params=None, iid='warn', n_jobs=None,\n",
       "       param_grid={'normalize': (True, False), 'fit_intercept': (True, False)},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a Linear Regression object and perform a grid search to find the best parameters\n",
    "linreg = CappedLinearRegression()\n",
    "parameters = {'normalize':(True,False), 'fit_intercept':(True,False)}\n",
    "\n",
    "#Create a grid search object using the  \n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=linreg\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "CappedLinearRegression(copy_X=True, fit_intercept=True, n_jobs=None,\n",
       "            normalize=True)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Print the parameterization of the best estimator\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Baseline Regression Model - Cross Validation\n",
    "**Perform tenfold cross validation using the grid search \"best\" parameters and our Capped Linear Regression estimator**\n",
    "* 10-fold cross-validation using the parameters for the top performing model \n",
    "* CAP predictions between 0 and 100% \n",
    "* Evaluate cross-validation results using MAE, MAPE, and RMSE"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 11.864\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 19.877\n",
      "The average RMSE for all cv folds is: \t\t\t 17.365\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>11.160201</td>\n",
       "      <td>13.715467</td>\n",
       "      <td>16.940213</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>13.252552</td>\n",
       "      <td>22.070585</td>\n",
       "      <td>20.201376</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>13.280939</td>\n",
       "      <td>17.776753</td>\n",
       "      <td>19.542977</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>9.435834</td>\n",
       "      <td>13.218761</td>\n",
       "      <td>13.872872</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>13.143821</td>\n",
       "      <td>16.108767</td>\n",
       "      <td>17.833189</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>12.056523</td>\n",
       "      <td>16.482314</td>\n",
       "      <td>17.415171</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>8.857411</td>\n",
       "      <td>21.138153</td>\n",
       "      <td>12.067560</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>12.185528</td>\n",
       "      <td>14.843477</td>\n",
       "      <td>17.838612</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>13.346722</td>\n",
       "      <td>28.294134</td>\n",
       "      <td>19.384689</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>11.920470</td>\n",
       "      <td>35.120694</td>\n",
       "      <td>18.555962</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         MAE       MAPE       RMSE\n",
       "0  11.160201  13.715467  16.940213\n",
       "1  13.252552  22.070585  20.201376\n",
       "2  13.280939  17.776753  19.542977\n",
       "3   9.435834  13.218761  13.872872\n",
       "4  13.143821  16.108767  17.833189\n",
       "5  12.056523  16.482314  17.415171\n",
       "6   8.857411  21.138153  12.067560\n",
       "7  12.185528  14.843477  17.838612\n",
       "8  13.346722  28.294134  19.384689\n",
       "9  11.920470  35.120694  18.555962"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create CappedLinearRegression predictions between 0 and 100% using the best parameters for our Linear Regression object\n",
    "regEstimator = regGridSearch.best_estimator_\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics. \n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## Support Vector Regression\n",
    "**This model uses Support Vector Machines for regression of continuous variables (SVR). Please see documentation here:\"**\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.svm.SVR.html\n",
    "* http://scikit-learn.org/stable/modules/svm.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 8 candidates, totalling 80 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed: 87.2min\n",
      "[Parallel(n_jobs=8)]: Done  80 out of  80 | elapsed: 306.8min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=SVR(C=1.0, cache_size=200, coef0=0.0, degree=3, epsilon=0.1,\n",
       "  gamma='auto_deprecated', kernel='rbf', max_iter=-1, shrinking=True,\n",
       "  tol=0.001, verbose=False),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'C': [0.001, 0.1], 'gamma': [0.0016313213703099511, 0.1], 'kernel': ['rbf', 'linear']},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a Linear regression object and perform a grid search to find the best parameters\n",
    "from sklearn.svm import SVR\n",
    "reg = SVR()\n",
    "\n",
    "#Set up SVR parameters to test (WARNING: Creates 80 models!!!) \n",
    "costs = [0.001, 0.1]\n",
    "defGamma = 1 / X_highSchools.shape[1]  #This is the default value for the gamma parameter\n",
    "gammas = [defGamma, 0.1]\n",
    "kernels = ['rbf','linear']\n",
    "parameters = {'C': costs, 'gamma' : gammas, 'kernel': kernels}\n",
    "\n",
    "#Create a grid search object using the parameters above\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=reg\n",
    "                   , n_jobs=8 # jobs to run in parallel\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SVR(C=0.001, cache_size=200, coef0=0.0, degree=3, epsilon=0.1,\n",
       "  gamma=0.0016313213703099511, kernel='rbf', max_iter=-1, shrinking=True,\n",
       "  tol=0.001, verbose=False)"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 8.4487\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 24.946\n",
      "The average RMSE for all cv folds is: \t\t\t 13.804\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7.675926</td>\n",
       "      <td>10.012086</td>\n",
       "      <td>10.242068</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>8.474074</td>\n",
       "      <td>41.117028</td>\n",
       "      <td>15.886227</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>8.546315</td>\n",
       "      <td>12.578250</td>\n",
       "      <td>12.771193</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>8.255389</td>\n",
       "      <td>13.680975</td>\n",
       "      <td>13.413115</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>7.600019</td>\n",
       "      <td>10.128867</td>\n",
       "      <td>10.502447</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>7.970370</td>\n",
       "      <td>12.612069</td>\n",
       "      <td>12.548454</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>10.244519</td>\n",
       "      <td>73.994533</td>\n",
       "      <td>20.315463</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>6.898056</td>\n",
       "      <td>9.083452</td>\n",
       "      <td>9.821015</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>11.322296</td>\n",
       "      <td>27.515239</td>\n",
       "      <td>18.209379</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>7.499981</td>\n",
       "      <td>38.738319</td>\n",
       "      <td>14.330399</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         MAE       MAPE       RMSE\n",
       "0   7.675926  10.012086  10.242068\n",
       "1   8.474074  41.117028  15.886227\n",
       "2   8.546315  12.578250  12.771193\n",
       "3   8.255389  13.680975  13.413115\n",
       "4   7.600019  10.128867  10.502447\n",
       "5   7.970370  12.612069  12.548454\n",
       "6  10.244519  73.994533  20.315463\n",
       "7   6.898056   9.083452   9.821015\n",
       "8  11.322296  27.515239  18.209379\n",
       "9   7.499981  38.738319  14.330399"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.svm import SVR\n",
    "\n",
    "#Create a regression estimator with best parameters for cross validation\n",
    "regEstimator = SVR(C=0.001, cache_size=200, coef0=0.0, degree=3, epsilon=0.1,\n",
    "                   gamma=0.0016313213703099511, kernel='rbf', max_iter=-1, shrinking=True,\n",
    "                   tol=0.001, verbose=False)\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics.\n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Yhat Max:  88.50150000000001\n"
     ]
    }
   ],
   "source": [
    "#Do we predict graduation rates greater than 100%?\n",
    "regEstimator = SVR(C=0.001, cache_size=200, coef0=0.0, degree=3, epsilon=0.1,\n",
    "                   gamma=0.0016313213703099511, kernel='rbf', max_iter=-1, shrinking=True,\n",
    "                   tol=0.001, verbose=False)\n",
    "\n",
    "regEstimator.fit(X_highSchools, Y)\n",
    "yhat = regEstimator.predict(X_highSchools)\n",
    "print(\"Yhat Max: \", yhat.max())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Graduation Rate - Lasso Regression\n",
    "**This model uses Lasso regression (L1 Norm). Please see documentation here:\"**\n",
    "* **Caution!** - See documentation for fit_intercept, normalize, and copy_X. Lasso can over-write your X data!\n",
    "* Lasso may also perform scaling as well.  Please see docs!\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.Lasso.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 20 candidates, totalling 200 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed:    5.6s\n",
      "[Parallel(n_jobs=8)]: Done 184 tasks      | elapsed:   10.6s\n",
      "[Parallel(n_jobs=8)]: Done 200 out of 200 | elapsed:   10.9s finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=Lasso(alpha=1.0, copy_X=True, fit_intercept=True, max_iter=10000,\n",
       "   normalize=True, positive=False, precompute=True, random_state=0,\n",
       "   selection='cyclic', tol=0.0001, warm_start=False),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'alpha': [0.001, 0.1, 1, 10, 20], 'selection': ['cyclic', 'random'], 'warm_start': [True, False]},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a regression object and perform a grid search to find the best parameters\n",
    "from sklearn.linear_model import Lasso\n",
    "\n",
    "reg = Lasso(fit_intercept=True, normalize=True,copy_X=True\n",
    "          , max_iter=10000, precompute=True, tol=0.0001, random_state=0)\n",
    "\n",
    "#Test parameters \n",
    "alpha = [0.001, 0.1, 1, 10, 20]\n",
    "selection = ['cyclic','random']\n",
    "warm_start = [True, False]\n",
    "parameters = {'alpha': alpha, 'selection': selection, 'warm_start': warm_start}\n",
    "\n",
    "#Create a grid search object using the parameters above\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=reg\n",
    "                   , n_jobs=8 # jobs to run in parallel\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Lasso(alpha=0.1, copy_X=True, fit_intercept=True, max_iter=10000,\n",
       "   normalize=True, positive=False, precompute=True, random_state=0,\n",
       "   selection='cyclic', tol=0.0001, warm_start=True)"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 4.8915\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 11.125\n",
      "The average RMSE for all cv folds is: \t\t\t 7.7657\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3.938741</td>\n",
       "      <td>5.074130</td>\n",
       "      <td>5.673482</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.443190</td>\n",
       "      <td>16.353963</td>\n",
       "      <td>10.024857</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.954757</td>\n",
       "      <td>6.610556</td>\n",
       "      <td>7.827504</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3.919176</td>\n",
       "      <td>5.908358</td>\n",
       "      <td>6.034645</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4.243233</td>\n",
       "      <td>5.373732</td>\n",
       "      <td>5.434744</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5.449490</td>\n",
       "      <td>7.692975</td>\n",
       "      <td>8.450277</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>5.223369</td>\n",
       "      <td>28.373057</td>\n",
       "      <td>10.150536</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>5.032247</td>\n",
       "      <td>6.064938</td>\n",
       "      <td>7.464032</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>5.196559</td>\n",
       "      <td>13.742380</td>\n",
       "      <td>9.193858</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>4.514683</td>\n",
       "      <td>16.058614</td>\n",
       "      <td>7.403092</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        MAE       MAPE       RMSE\n",
       "0  3.938741   5.074130   5.673482\n",
       "1  6.443190  16.353963  10.024857\n",
       "2  4.954757   6.610556   7.827504\n",
       "3  3.919176   5.908358   6.034645\n",
       "4  4.243233   5.373732   5.434744\n",
       "5  5.449490   7.692975   8.450277\n",
       "6  5.223369  28.373057  10.150536\n",
       "7  5.032247   6.064938   7.464032\n",
       "8  5.196559  13.742380   9.193858\n",
       "9  4.514683  16.058614   7.403092"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.linear_model import Lasso\n",
    "\n",
    "#Create a regression estimator with best parameters for cross validation\n",
    "regEstimator = Lasso(alpha=0.1, copy_X=True, fit_intercept=True, max_iter=10000,\n",
    "                     normalize=True, positive=False, precompute=True, random_state=0,\n",
    "                     selection='cyclic', tol=0.0001, warm_start=True)\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics.\n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Yhat Max:  95.5050078743451\n"
     ]
    }
   ],
   "source": [
    "#Do we predict graduation rates greater than 100%?\n",
    "regEstimator = Lasso(alpha=0.1, copy_X=True, fit_intercept=True, max_iter=10000,\n",
    "                     normalize=True, positive=False, precompute=True, random_state=0,\n",
    "                     selection='cyclic', tol=0.0001, warm_start=True)\n",
    "\n",
    "regEstimator.fit(X_highSchools, Y)\n",
    "yhat = regEstimator.predict(X_highSchools)\n",
    "print(\"Yhat Max: \", yhat.max())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Ridge Regression\n",
    "**This model uses Ridge regression (L2 Norm). Please see documentation here:\"**\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.Ridge.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 36 candidates, totalling 360 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed:    4.4s\n",
      "[Parallel(n_jobs=8)]: Done 184 tasks      | elapsed:   28.4s\n",
      "[Parallel(n_jobs=8)]: Done 360 out of 360 | elapsed:   33.4s finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=Ridge(alpha=1.0, copy_X=True, fit_intercept=True, max_iter=1000,\n",
       "   normalize=True, random_state=0, solver='auto', tol=0.0001),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'alpha': [0.001, 0.1, 1, 5, 10, 20], 'solver': ['svd', 'cholesky', 'lsqr', 'sparse_cg', 'sag', 'saga']},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a regression object and perform a grid search to find the best parameters\n",
    "from sklearn.linear_model import Ridge\n",
    "\n",
    "reg = Ridge(fit_intercept=True, normalize=True,copy_X=True\n",
    "          , max_iter=1000, tol=0.0001, random_state=0)\n",
    "\n",
    "#Test parameters \n",
    "alpha = [0.001, 0.1, 1, 5, 10, 20]\n",
    "solver = [ 'svd', 'cholesky', 'lsqr', 'sparse_cg', 'sag', 'saga']\n",
    "parameters = {'alpha': alpha, 'solver': solver}\n",
    "\n",
    "#Create a grid search object using the parameters above\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=reg\n",
    "                   , n_jobs=8 # jobs to run in parallel\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Ridge(alpha=1, copy_X=True, fit_intercept=True, max_iter=1000, normalize=True,\n",
       "   random_state=0, solver='svd', tol=0.0001)"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 5.3858\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 11.828\n",
      "The average RMSE for all cv folds is: \t\t\t 8.4452\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>4.559282</td>\n",
       "      <td>5.722364</td>\n",
       "      <td>6.562002</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.821794</td>\n",
       "      <td>16.732035</td>\n",
       "      <td>10.308633</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.933653</td>\n",
       "      <td>6.415271</td>\n",
       "      <td>7.775474</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.563335</td>\n",
       "      <td>6.131716</td>\n",
       "      <td>6.683638</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4.155343</td>\n",
       "      <td>5.007036</td>\n",
       "      <td>5.594813</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>6.951969</td>\n",
       "      <td>10.103231</td>\n",
       "      <td>11.089799</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>5.100118</td>\n",
       "      <td>30.957213</td>\n",
       "      <td>9.404295</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>5.620008</td>\n",
       "      <td>6.657565</td>\n",
       "      <td>9.139963</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>5.484358</td>\n",
       "      <td>12.895808</td>\n",
       "      <td>8.915608</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>5.668196</td>\n",
       "      <td>17.655651</td>\n",
       "      <td>8.977631</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        MAE       MAPE       RMSE\n",
       "0  4.559282   5.722364   6.562002\n",
       "1  6.821794  16.732035  10.308633\n",
       "2  4.933653   6.415271   7.775474\n",
       "3  4.563335   6.131716   6.683638\n",
       "4  4.155343   5.007036   5.594813\n",
       "5  6.951969  10.103231  11.089799\n",
       "6  5.100118  30.957213   9.404295\n",
       "7  5.620008   6.657565   9.139963\n",
       "8  5.484358  12.895808   8.915608\n",
       "9  5.668196  17.655651   8.977631"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a regression estimator with best parameters for cross validation\n",
    "regEstimator = regGridSearch.best_estimator_\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics.\n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Yhat Max:  100.82452049849329\n"
     ]
    }
   ],
   "source": [
    "#Do we predict graduation rates greater than 100%?\n",
    "regEstimator = regGridSearch.best_estimator_\n",
    "\n",
    "regEstimator.fit(X_highSchools, Y)\n",
    "yhat = regEstimator.predict(X_highSchools)\n",
    "print(\"Yhat Max: \", yhat.max())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Elastic Net Regression\n",
    "**This model uses Elastic Net Regression (L1 and L2 Norm mixing). Please see documentation here:\"**\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.ElasticNet.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 96 candidates, totalling 960 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed:    5.0s\n",
      "[Parallel(n_jobs=8)]: Done 184 tasks      | elapsed:   21.4s\n",
      "[Parallel(n_jobs=8)]: Done 434 tasks      | elapsed:   30.5s\n",
      "[Parallel(n_jobs=8)]: Done 784 tasks      | elapsed:   38.9s\n",
      "[Parallel(n_jobs=8)]: Done 960 out of 960 | elapsed:   42.9s finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=ElasticNet(alpha=1.0, copy_X=True, fit_intercept=True, l1_ratio=0.5,\n",
       "      max_iter=10000, normalize=True, positive=False, precompute=True,\n",
       "      random_state=0, selection='cyclic', tol=0.0001, warm_start=False),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'l1_ratio': [0.001, 0.01, 0.1, 0.5, 0.75, 1], 'alpha': [0.001, 0.1, 1, 10], 'selection': ['cyclic', 'random'], 'warm_start': [True, False]},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a regression object and perform a grid search to find the best parameters\n",
    "from sklearn.linear_model import ElasticNet\n",
    "\n",
    "reg = ElasticNet(fit_intercept=True, normalize=True, precompute=True, copy_X=True\n",
    "          , max_iter=10000, tol=0.0001, random_state=0)\n",
    " \n",
    "#Test parameters\n",
    "l1_ratio = [0.001, 0.01, 0.1, 0.5, 0.75, 1]\n",
    "alpha = [0.001, 0.1, 1, 10]\n",
    "selection = ['cyclic','random']\n",
    "warm_start = [True, False]\n",
    "parameters = {'l1_ratio': l1_ratio, 'alpha': alpha, 'selection': selection, 'warm_start': warm_start}\n",
    "\n",
    "#Create a grid search object using the parameters above\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=reg\n",
    "                   , n_jobs=8 # jobs to run in parallel\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ElasticNet(alpha=0.1, copy_X=True, fit_intercept=True, l1_ratio=1,\n",
       "      max_iter=10000, normalize=True, positive=False, precompute=True,\n",
       "      random_state=0, selection='random', tol=0.0001, warm_start=True)"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 4.8908\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 11.124\n",
      "The average RMSE for all cv folds is: \t\t\t 7.7642\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3.937957</td>\n",
       "      <td>5.073328</td>\n",
       "      <td>5.673693</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.442637</td>\n",
       "      <td>16.352083</td>\n",
       "      <td>10.023126</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.952152</td>\n",
       "      <td>6.605203</td>\n",
       "      <td>7.821276</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3.920452</td>\n",
       "      <td>5.909554</td>\n",
       "      <td>6.032423</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4.241561</td>\n",
       "      <td>5.372037</td>\n",
       "      <td>5.432706</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5.449684</td>\n",
       "      <td>7.692945</td>\n",
       "      <td>8.449574</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>5.222069</td>\n",
       "      <td>28.373343</td>\n",
       "      <td>10.149818</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>5.031768</td>\n",
       "      <td>6.063995</td>\n",
       "      <td>7.462546</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>5.195722</td>\n",
       "      <td>13.740614</td>\n",
       "      <td>9.193469</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>4.514205</td>\n",
       "      <td>16.059373</td>\n",
       "      <td>7.402930</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        MAE       MAPE       RMSE\n",
       "0  3.937957   5.073328   5.673693\n",
       "1  6.442637  16.352083  10.023126\n",
       "2  4.952152   6.605203   7.821276\n",
       "3  3.920452   5.909554   6.032423\n",
       "4  4.241561   5.372037   5.432706\n",
       "5  5.449684   7.692945   8.449574\n",
       "6  5.222069  28.373343  10.149818\n",
       "7  5.031768   6.063995   7.462546\n",
       "8  5.195722  13.740614   9.193469\n",
       "9  4.514205  16.059373   7.402930"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a regression estimator with best parameters for cross validation\n",
    "regEstimator = regGridSearch.best_estimator_\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics.\n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Yhat Max:  95.50569622560306\n"
     ]
    }
   ],
   "source": [
    "#Do we predict graduation rates greater than 100%?\n",
    "regEstimator =  regGridSearch.best_estimator_\n",
    "\n",
    "regEstimator.fit(X_highSchools, Y)\n",
    "yhat = regEstimator.predict(X_highSchools)\n",
    "print(\"Yhat Max: \", yhat.max())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Regression using the Random Forest Regressor \n",
    "**This model uses a custom Logit model using the Random Forest Regressor. Please see documentation here: **\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestRegressor.html#sklearn.ensemble.RandomForestRegressor\n",
    "* https://www.analyticsvidhya.com/blog/2015/06/tuning-random-forest-model/\n",
    "* https://www.kaggle.com/general/4092"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 12 candidates, totalling 120 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed: 16.6min\n",
      "[Parallel(n_jobs=8)]: Done 120 out of 120 | elapsed: 35.4min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=RandomForestRegressor(bootstrap=True, criterion='mse', max_depth=None,\n",
       "           max_features='auto', max_leaf_nodes=None,\n",
       "           min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "           min_samples_leaf=1, min_samples_split=2,\n",
       "           min_weight_fraction_leaf=0.0, n_estimators='warn', n_jobs=None,\n",
       "           oob_score=False, random_state=None, verbose=0, warm_start=False),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'min_samples_split': [2, 3, 4, 5], 'n_estimators': [500], 'min_samples_leaf': [10, 25, 50], 'criterion': ['mae'], 'n_jobs': [8], 'random_state': [0]},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a Linear Regression object and perform a grid search to find the best parameters\n",
    "from sklearn.ensemble import RandomForestRegressor\n",
    "\n",
    "linreg = RandomForestRegressor()\n",
    "parameters = { 'min_samples_split':[2,3,4,5]\n",
    "              ,'n_estimators' : [500]\n",
    "              ,'min_samples_leaf': [10, 25, 50]\n",
    "              ,'criterion': ['mae']\n",
    "              ,'n_jobs':[8] \n",
    "              ,'random_state': [0]\n",
    "             }\n",
    "\n",
    "#Create a grid search object using the  \n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=linreg\n",
    "                   , n_jobs=8 \n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "RandomForestRegressor(bootstrap=True, criterion='mae', max_depth=None,\n",
       "           max_features='auto', max_leaf_nodes=None,\n",
       "           min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "           min_samples_leaf=25, min_samples_split=2,\n",
       "           min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
       "           oob_score=False, random_state=0, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 4.5672\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 14.218\n",
      "The average RMSE for all cv folds is: \t\t\t 8.2166\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3.259081</td>\n",
       "      <td>4.216215</td>\n",
       "      <td>5.040674</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.437228</td>\n",
       "      <td>26.553875</td>\n",
       "      <td>12.042380</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.016654</td>\n",
       "      <td>5.456181</td>\n",
       "      <td>6.503988</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.133141</td>\n",
       "      <td>6.430898</td>\n",
       "      <td>6.682765</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3.247300</td>\n",
       "      <td>4.078971</td>\n",
       "      <td>4.406723</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5.413031</td>\n",
       "      <td>7.934487</td>\n",
       "      <td>9.394775</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>5.721611</td>\n",
       "      <td>46.024083</td>\n",
       "      <td>12.202571</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>4.190207</td>\n",
       "      <td>5.152883</td>\n",
       "      <td>7.937173</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>5.002378</td>\n",
       "      <td>13.250361</td>\n",
       "      <td>9.211104</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>4.251294</td>\n",
       "      <td>23.078166</td>\n",
       "      <td>8.743986</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        MAE       MAPE       RMSE\n",
       "0  3.259081   4.216215   5.040674\n",
       "1  6.437228  26.553875  12.042380\n",
       "2  4.016654   5.456181   6.503988\n",
       "3  4.133141   6.430898   6.682765\n",
       "4  3.247300   4.078971   4.406723\n",
       "5  5.413031   7.934487   9.394775\n",
       "6  5.721611  46.024083  12.202571\n",
       "7  4.190207   5.152883   7.937173\n",
       "8  5.002378  13.250361   9.211104\n",
       "9  4.251294  23.078166   8.743986"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "regEstimator = RandomForestRegressor(bootstrap=True, criterion='mae', max_depth=None,\n",
    "           max_features='auto', max_leaf_nodes=None,\n",
    "           min_impurity_decrease=0.0, min_impurity_split=None,\n",
    "           min_samples_leaf=25, min_samples_split=5,\n",
    "           min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
    "           oob_score=False, random_state=0, verbose=0, warm_start=False)\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics. \n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Yhat Max:  95.0\n"
     ]
    }
   ],
   "source": [
    "#Do we predict graduation rates greater than 100%?\n",
    "regEstimator = RandomForestRegressor(bootstrap=True, criterion='mae', max_depth=None,\n",
    "           max_features='auto', max_leaf_nodes=None,\n",
    "           min_impurity_decrease=0.0, min_impurity_split=None,\n",
    "           min_samples_leaf=25, min_samples_split=5,\n",
    "           min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
    "           oob_score=False, random_state=0, verbose=0, warm_start=False)\n",
    "\n",
    "regEstimator.fit(X_highSchools, Y)\n",
    "yhat = regEstimator.predict(X_highSchools)\n",
    "print(\"Yhat Max: \", yhat.max())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Regression using the Extra Trees Regressor \n",
    "**This model uses a custom Logit model using the Random Forest Regressor. Please see documentation here: **\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.ensemble.ExtraTreesRegressor.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 12 candidates, totalling 120 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed:  9.7min\n",
      "[Parallel(n_jobs=8)]: Done 120 out of 120 | elapsed: 19.8min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=ExtraTreesRegressor(bootstrap=False, criterion='mse', max_depth=None,\n",
       "          max_features='auto', max_leaf_nodes=None,\n",
       "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "          min_samples_leaf=1, min_samples_split=2,\n",
       "          min_weight_fraction_leaf=0.0, n_estimators='warn', n_jobs=None,\n",
       "          oob_score=False, random_state=None, verbose=0, warm_start=False),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'min_samples_split': [2, 3, 4, 5], 'n_estimators': [500], 'min_samples_leaf': [10, 25, 50], 'criterion': ['mae'], 'n_jobs': [8], 'random_state': [0]},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a Linear Regression object and perform a grid search to find the best parameters\n",
    "from sklearn.ensemble import ExtraTreesRegressor\n",
    "\n",
    "linreg = ExtraTreesRegressor()\n",
    "parameters = { 'min_samples_split':[2,3,4,5]\n",
    "              ,'n_estimators' : [500]\n",
    "              ,'min_samples_leaf': [10, 25, 50]\n",
    "              ,'criterion': ['mae']\n",
    "              ,'n_jobs':[8] \n",
    "              ,'random_state': [0]\n",
    "             }\n",
    "\n",
    "#Create a grid search object using the  \n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=linreg\n",
    "                   , n_jobs=8 # jobs to run in parallel\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ExtraTreesRegressor(bootstrap=False, criterion='mae', max_depth=None,\n",
       "          max_features='auto', max_leaf_nodes=None,\n",
       "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "          min_samples_leaf=25, min_samples_split=2,\n",
       "          min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
       "          oob_score=False, random_state=0, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 4.7414\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 11.869\n",
      "The average RMSE for all cv folds is: \t\t\t 9.0968\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>4.533613</td>\n",
       "      <td>6.188370</td>\n",
       "      <td>9.820520</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.087533</td>\n",
       "      <td>13.866803</td>\n",
       "      <td>11.224118</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.910267</td>\n",
       "      <td>6.693758</td>\n",
       "      <td>9.418293</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.297085</td>\n",
       "      <td>6.530385</td>\n",
       "      <td>7.162960</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3.660500</td>\n",
       "      <td>4.758730</td>\n",
       "      <td>6.517192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5.767494</td>\n",
       "      <td>7.932863</td>\n",
       "      <td>10.835884</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>4.500280</td>\n",
       "      <td>37.113224</td>\n",
       "      <td>9.758154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>4.515439</td>\n",
       "      <td>5.557012</td>\n",
       "      <td>9.823689</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>4.343585</td>\n",
       "      <td>10.991885</td>\n",
       "      <td>7.708154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>4.798543</td>\n",
       "      <td>19.060000</td>\n",
       "      <td>8.699373</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        MAE       MAPE       RMSE\n",
       "0  4.533613   6.188370   9.820520\n",
       "1  6.087533  13.866803  11.224118\n",
       "2  4.910267   6.693758   9.418293\n",
       "3  4.297085   6.530385   7.162960\n",
       "4  3.660500   4.758730   6.517192\n",
       "5  5.767494   7.932863  10.835884\n",
       "6  4.500280  37.113224   9.758154\n",
       "7  4.515439   5.557012   9.823689\n",
       "8  4.343585  10.991885   7.708154\n",
       "9  4.798543  19.060000   8.699373"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble import ExtraTreesRegressor\n",
    "\n",
    "#Create CappedLinearRegression predictions between 0 and 100% using the best parameters for our Linear Regression object\n",
    "regEstimator = ExtraTreesRegressor(bootstrap=False, criterion='mae', max_depth=None,\n",
    "          max_features='auto', max_leaf_nodes=None,\n",
    "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
    "          min_samples_leaf=10, min_samples_split=2,\n",
    "          min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
    "          oob_score=False, random_state=0, verbose=0, warm_start=False)\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics. \n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Yhat Max:  95.0\n"
     ]
    }
   ],
   "source": [
    "#Do we predict graduation rates greater than 100%?\n",
    "regEstimator = ExtraTreesRegressor(bootstrap=False, criterion='mae', max_depth=None,\n",
    "          max_features='auto', max_leaf_nodes=None,\n",
    "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
    "          min_samples_leaf=10, min_samples_split=2,\n",
    "          min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
    "          oob_score=False, random_state=0, verbose=0, warm_start=False)\n",
    "\n",
    "regEstimator.fit(X_highSchools, Y)\n",
    "yhat = regEstimator.predict(X_highSchools)\n",
    "print(\"Yhat Max: \", yhat.max())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## Multilayer Perceptron Regression \n",
    "**This model uses a neural network for regression. Please see documentation here: **\n",
    "* http://scikit-learn.org/stable/modules/generated/sklearn.neural_network.MLPRegressor.html\n",
    "* https://stackoverflow.com/questions/41069905/trouble-fitting-simple-data-with-mlpregressor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 6 candidates, totalling 60 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed:  1.1min\n",
      "[Parallel(n_jobs=8)]: Done  60 out of  60 | elapsed:  2.0min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=MLPRegressor(activation='relu', alpha=0.0001, batch_size='auto', beta_1=0.9,\n",
       "       beta_2=0.999, early_stopping=False, epsilon=1e-08,\n",
       "       hidden_layer_sizes=(100,), learning_rate='constant',\n",
       "       learning_rate_init=0.001, max_iter=200, momentum=0.9,\n",
       "       n_iter_no_change=10, nesterovs_momentum=True, power_t=0.5,\n",
       "       random_state=None, shuffle=True, solver='adam', tol=0.0001,\n",
       "       validation_fraction=0.1, verbose=False, warm_start=False),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'activation': ['logistic'], 'hidden_layer_sizes': [200], 'solver': ['sgd', 'adam'], 'alpha': [0.001], 'batch_size': ['auto'], 'random_state': [0], 'learning_rate': ['constant', 'invscaling', 'adaptive'], 'learning_rate_init': [0.001], 'power_t': [0.5], 'max_iter': [1000]},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a Linear Regression object and perform a grid search to find the best parameters\n",
    "from sklearn.neural_network import MLPRegressor\n",
    "\n",
    "linreg = MLPRegressor()\n",
    "parameters = { 'activation':['logistic']\n",
    "              ,'hidden_layer_sizes' : [200]\n",
    "              ,'solver': ['sgd','adam']\n",
    "              ,'alpha': [0.001]\n",
    "              ,'batch_size':['auto'] \n",
    "              ,'random_state': [0]\n",
    "              ,'learning_rate': ['constant', 'invscaling', 'adaptive']\n",
    "              ,'learning_rate_init':[0.001]\n",
    "              ,'power_t':[0.5]\n",
    "              ,'max_iter':[1000]\n",
    "             }\n",
    "\n",
    "#Create a grid search object using the  \n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=linreg\n",
    "                   , n_jobs=8 # jobs to run in parallel\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MLPRegressor(activation='logistic', alpha=0.001, batch_size='auto',\n",
       "       beta_1=0.9, beta_2=0.999, early_stopping=False, epsilon=1e-08,\n",
       "       hidden_layer_sizes=200, learning_rate='constant',\n",
       "       learning_rate_init=0.001, max_iter=1000, momentum=0.9,\n",
       "       n_iter_no_change=10, nesterovs_momentum=True, power_t=0.5,\n",
       "       random_state=0, shuffle=True, solver='sgd', tol=0.0001,\n",
       "       validation_fraction=0.1, verbose=False, warm_start=False)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 9.0619\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 24.546\n",
      "The average RMSE for all cv folds is: \t\t\t 13.33\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>8.575180</td>\n",
       "      <td>10.521021</td>\n",
       "      <td>10.056089</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>9.345976</td>\n",
       "      <td>40.298305</td>\n",
       "      <td>15.231470</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>9.272864</td>\n",
       "      <td>12.793694</td>\n",
       "      <td>12.333323</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>7.810063</td>\n",
       "      <td>12.561104</td>\n",
       "      <td>12.350277</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>8.066394</td>\n",
       "      <td>10.270659</td>\n",
       "      <td>9.937087</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>8.548315</td>\n",
       "      <td>12.785252</td>\n",
       "      <td>12.267600</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>10.504672</td>\n",
       "      <td>71.113033</td>\n",
       "      <td>19.408175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>8.493422</td>\n",
       "      <td>10.455705</td>\n",
       "      <td>10.471773</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>11.494230</td>\n",
       "      <td>26.509389</td>\n",
       "      <td>17.156747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>8.507687</td>\n",
       "      <td>38.148408</td>\n",
       "      <td>14.084503</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         MAE       MAPE       RMSE\n",
       "0   8.575180  10.521021  10.056089\n",
       "1   9.345976  40.298305  15.231470\n",
       "2   9.272864  12.793694  12.333323\n",
       "3   7.810063  12.561104  12.350277\n",
       "4   8.066394  10.270659   9.937087\n",
       "5   8.548315  12.785252  12.267600\n",
       "6  10.504672  71.113033  19.408175\n",
       "7   8.493422  10.455705  10.471773\n",
       "8  11.494230  26.509389  17.156747\n",
       "9   8.507687  38.148408  14.084503"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create CappedLinearRegression predictions between 0 and 100% using the best parameters for our Linear Regression object\n",
    "regEstimator = regGridSearch.best_estimator_\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics. \n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Yhat Max:  85.7039677069584\n"
     ]
    }
   ],
   "source": [
    "#Do we predict graduation rates greater than 100%?\n",
    "regEstimator = regGridSearch.best_estimator_\n",
    "\n",
    "regEstimator.fit(X_highSchools, Y)\n",
    "yhat = regEstimator.predict(X_highSchools)\n",
    "print(\"Yhat Max: \", yhat.max())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# Top Performing Model Analisys\n",
    "## Testing Scaled vs. Unscaled Training Data\n",
    "**Cross validation is performed using both scaled and unscaled training data.**\n",
    "* A grid search is also performed on the scaled training data to determine the best parameters to use with the scaled training data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 4.7414\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 11.869\n",
      "The average RMSE for all cv folds is: \t\t\t 9.0968\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>4.533613</td>\n",
       "      <td>6.188370</td>\n",
       "      <td>9.820520</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.087533</td>\n",
       "      <td>13.866803</td>\n",
       "      <td>11.224118</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4.910267</td>\n",
       "      <td>6.693758</td>\n",
       "      <td>9.418293</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4.297085</td>\n",
       "      <td>6.530385</td>\n",
       "      <td>7.162960</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3.660500</td>\n",
       "      <td>4.758730</td>\n",
       "      <td>6.517192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5.767494</td>\n",
       "      <td>7.932863</td>\n",
       "      <td>10.835884</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>4.500280</td>\n",
       "      <td>37.113224</td>\n",
       "      <td>9.758154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>4.515439</td>\n",
       "      <td>5.557012</td>\n",
       "      <td>9.823689</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>4.343585</td>\n",
       "      <td>10.991885</td>\n",
       "      <td>7.708154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>4.798543</td>\n",
       "      <td>19.060000</td>\n",
       "      <td>8.699373</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        MAE       MAPE       RMSE\n",
       "0  4.533613   6.188370   9.820520\n",
       "1  6.087533  13.866803  11.224118\n",
       "2  4.910267   6.693758   9.418293\n",
       "3  4.297085   6.530385   7.162960\n",
       "4  3.660500   4.758730   6.517192\n",
       "5  5.767494   7.932863  10.835884\n",
       "6  4.500280  37.113224   9.758154\n",
       "7  4.515439   5.557012   9.823689\n",
       "8  4.343585  10.991885   7.708154\n",
       "9  4.798543  19.060000   8.699373"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble import ExtraTreesRegressor\n",
    "\n",
    "#Create a regression estimator with best parameters for cross validation\n",
    "regEstimator = ExtraTreesRegressor(bootstrap=False, criterion='mae', max_depth=None,\n",
    "          max_features='auto', max_leaf_nodes=None,\n",
    "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
    "          min_samples_leaf=10, min_samples_split=2,\n",
    "          min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
    "          oob_score=False, random_state=0, verbose=0, warm_start=False)\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics.\n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools, Y, cv)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "* **We scale the data prior to looking at feature importances.  Since the raw features have different dynamic ranges, scaling the data first allows us to compare the model's coefficients accurately.** \n",
    "* **The scaled data makes more accurate predictions in many cases but not always!**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Jake\\Anaconda2\\envs\\example_env\\Lib\\site-packages\\sklearn\\preprocessing\\data.py:645: DataConversionWarning: Data with input dtype int64, float64, object were all converted to float64 by StandardScaler.\n",
      "  return self.partial_fit(X, y)\n",
      "C:\\Users\\Jake\\Anaconda2\\envs\\example_env\\Lib\\site-packages\\ipykernel_launcher.py:11: DataConversionWarning: Data with input dtype int64, float64, object were all converted to float64 by StandardScaler.\n",
      "  # This is added back by InteractiveShellApp.init_path()\n"
     ]
    }
   ],
   "source": [
    "#Scale dataset converting to standard normally distributed data \n",
    "# (e.g. Gaussian with 0 mean and unit variance).\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "#Fit to data for scaling\n",
    "scaler = StandardScaler()\n",
    "scaler.fit(X_highSchools)\n",
    "\n",
    "#Transform training data to z-scores\n",
    "#This makes our model's coefficients take on the same scale for accurate feature importance analisys \n",
    "X_highSchools_Scl = scaler.transform(X_highSchools)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**In this case, I tested scaling the data before cross validation as well.**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 12 candidates, totalling 120 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=8)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=8)]: Done  34 tasks      | elapsed:  9.8min\n",
      "[Parallel(n_jobs=8)]: Done 120 out of 120 | elapsed: 20.3min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "GridSearchCV(cv=ShuffleSplit(n_splits=10, random_state=0, test_size=0.1, train_size=None),\n",
       "       error_score='raise-deprecating',\n",
       "       estimator=ExtraTreesRegressor(bootstrap=False, criterion='mse', max_depth=None,\n",
       "          max_features='auto', max_leaf_nodes=None,\n",
       "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "          min_samples_leaf=1, min_samples_split=2,\n",
       "          min_weight_fraction_leaf=0.0, n_estimators='warn', n_jobs=None,\n",
       "          oob_score=False, random_state=None, verbose=0, warm_start=False),\n",
       "       fit_params=None, iid='warn', n_jobs=8,\n",
       "       param_grid={'min_samples_split': [2, 3, 4, 5], 'n_estimators': [500], 'min_samples_leaf': [10, 25, 50], 'criterion': ['mae'], 'n_jobs': [8], 'random_state': [0]},\n",
       "       pre_dispatch='2*n_jobs', refit=True, return_train_score='warn',\n",
       "       scoring=make_scorer(mean_absolute_error, greater_is_better=False),\n",
       "       verbose=1)"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble import ExtraTreesRegressor\n",
    "\n",
    "linreg = ExtraTreesRegressor()\n",
    "parameters = { 'min_samples_split':[2,3,4,5]\n",
    "              ,'n_estimators' : [500]\n",
    "              ,'min_samples_leaf': [10, 25, 50]\n",
    "              ,'criterion': ['mae']\n",
    "              ,'n_jobs':[8] \n",
    "              ,'random_state': [0]\n",
    "             }\n",
    "\n",
    "#Create a grid search object using the  \n",
    "from sklearn.model_selection import GridSearchCV\n",
    "regGridSearch = GridSearchCV(estimator=linreg\n",
    "                   , n_jobs=8 # jobs to run in parallel\n",
    "                   , verbose=1 # low verbosity\n",
    "                   , param_grid=parameters\n",
    "                   , cv=cv # KFolds = 10\n",
    "                   , scoring=mae_scorer)\n",
    "\n",
    "#Perform hyperparameter search to find the best combination of parameters for our data\n",
    "regGridSearch.fit(X_highSchools_Scl, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ExtraTreesRegressor(bootstrap=False, criterion='mae', max_depth=None,\n",
       "          max_features='auto', max_leaf_nodes=None,\n",
       "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "          min_samples_leaf=25, min_samples_split=2,\n",
       "          min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
       "          oob_score=False, random_state=0, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Display the best estimator parameters\n",
    "regGridSearch.best_estimator_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The average MAE for all cv folds is: \t\t\t 4.3614\n",
      "The average MAE percentage (MAPE) for all cv folds is: \t 15.012\n",
      "The average RMSE for all cv folds is: \t\t\t 7.981\n",
      "*********************************************************\n",
      "Cross Validation Fold Mean Error Scores\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>MAE</th>\n",
       "      <th>MAPE</th>\n",
       "      <th>RMSE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>3.150093</td>\n",
       "      <td>4.120482</td>\n",
       "      <td>4.774432</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>6.210463</td>\n",
       "      <td>25.366572</td>\n",
       "      <td>11.287395</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3.474954</td>\n",
       "      <td>4.506942</td>\n",
       "      <td>5.321563</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3.741624</td>\n",
       "      <td>5.733119</td>\n",
       "      <td>5.849354</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3.293494</td>\n",
       "      <td>4.199712</td>\n",
       "      <td>4.456315</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5.060748</td>\n",
       "      <td>7.294209</td>\n",
       "      <td>8.452276</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>5.981122</td>\n",
       "      <td>51.895294</td>\n",
       "      <td>13.494034</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>3.870546</td>\n",
       "      <td>4.708183</td>\n",
       "      <td>7.264682</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>4.425983</td>\n",
       "      <td>12.333225</td>\n",
       "      <td>8.390218</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>4.404517</td>\n",
       "      <td>29.966831</td>\n",
       "      <td>10.519402</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        MAE       MAPE       RMSE\n",
       "0  3.150093   4.120482   4.774432\n",
       "1  6.210463  25.366572  11.287395\n",
       "2  3.474954   4.506942   5.321563\n",
       "3  3.741624   5.733119   5.849354\n",
       "4  3.293494   4.199712   4.456315\n",
       "5  5.060748   7.294209   8.452276\n",
       "6  5.981122  51.895294  13.494034\n",
       "7  3.870546   4.708183   7.264682\n",
       "8  4.425983  12.333225   8.390218\n",
       "9  4.404517  29.966831  10.519402"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Create a regression estimator with best parameters for cross validation\n",
    "regEstimator = regGridSearch.best_estimator_\n",
    "\n",
    "#Evaluate the regression estimator above using our pre-defined cross validation and scoring metrics.\n",
    "EvaluateRegressionEstimator(regEstimator, X_highSchools_Scl, Y, cv)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature Importance for the Top Performing Model\n",
    "**Here we look at the features driving the top performing model's decisions.**\n",
    "* The top performing model uses Support Vector Regression and performs at +-1.88% Mean Absolute Error during Cross Validation.\n",
    "* All training data is scaled to remove differences in coefficient scales since many features have different dynamic ranges.\n",
    "* A final model is created using all the scaled training data.  \n",
    "* All model coefficients are sotred by absolute value in descending order.\n",
    "* Sorted coefficient values are plotted showing the top 75 features with the strongest positive and negative correlations to the graduation rate.  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "**Next, we create one final model using all of the scaled training data for evaluating feature importance.**  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ExtraTreesRegressor(bootstrap=False, criterion='mae', max_depth=None,\n",
       "          max_features='auto', max_leaf_nodes=None,\n",
       "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "          min_samples_leaf=10, min_samples_split=4,\n",
       "          min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
       "          oob_score=False, random_state=0, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble import ExtraTreesRegressor\n",
    "\n",
    "#Create a regression estimator with best parameters for cross validation\n",
    "regEstimator = ExtraTreesRegressor(bootstrap=False, criterion='mae', max_depth=None,\n",
    "          max_features='auto', max_leaf_nodes=None,\n",
    "          min_impurity_decrease=0.0, min_impurity_split=None,\n",
    "          min_samples_leaf=10, min_samples_split=4,\n",
    "          min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=8,\n",
    "          oob_score=False, random_state=0, verbose=0, warm_start=False)\n",
    "\n",
    "#Fit the model using all of the scaled training data\n",
    "regEstimator.fit(X_highSchools_Scl, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Load the model's coefficient weights and feature names into a dataframe sorted by weights\n",
    "weights = regEstimator.feature_importances_.ravel()\n",
    "feature_names = X_highSchools.columns.values\n",
    "linreg_ft_imp_df = pd.DataFrame({'feature_names':feature_names, 'weights':weights, 'absolute_weights': np.abs(weights)})\n",
    "linreg_ft_imp_df.sort_values(by='absolute_weights', inplace=True, ascending=False )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x576 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Examine categorical variables of interest  \n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "#Plot the model's feature importances\n",
    "# REFERENCE:  Eric Larson, https://github.com/eclarson/DataMiningNotebooks\n",
    "plt.style.use('ggplot')\n",
    "\n",
    "wt_plt_df = linreg_ft_imp_df.head(75)\n",
    "\n",
    "weights = pd.Series(wt_plt_df['weights'].values,index=wt_plt_df['feature_names'])\n",
    "ax = weights.plot(kind='bar', figsize=(20,8))\n",
    "\n",
    "ax.set_title(\"Top Feature Correlations\")\n",
    "ax.set_ylabel(\"Coefficient Magnitude\\n(z-score)\")\n",
    "ax.set_xlabel(\"Feature Names\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "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": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}