DRAGON_noVPAS.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Initialisations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:03.298659Z",
     "start_time": "2020-01-23T12:09:00.029831Z"
    }
   },
   "outputs": [],
   "source": [
    "import pandas as pd ; pd.set_option('display.max_columns', 500) # dataframes\n",
    "import numpy as np # mathsy bits\n",
    "import ipywidgets as widgets # widgets\n",
    "\n",
    "from sklearn.model_selection import train_test_split #split data into train and test sets\n",
    "\n",
    "# feature selection + gridsearch\n",
    "from sklearn.feature_selection import SelectFromModel, SelectKBest\n",
    "from sklearn.feature_selection import f_regression, mutual_info_regression\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "\n",
    "# models\n",
    "from sklearn.linear_model import Lars, ElasticNet, Lasso\n",
    "from sklearn.tree import DecisionTreeRegressor\n",
    "from sklearn.ensemble import RandomForestRegressor, ExtraTreesRegressor\n",
    "from sklearn.ensemble import AdaBoostRegressor, GradientBoostingRegressor\n",
    "\n",
    "import joblib # saving models\n",
    "from datetime import datetime # get time for labelling saved models\n",
    "\n",
    "# visualization\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "import mpld3 # hover-over labels for plots\n",
    "mpld3.disable_notebook()\n",
    "\n",
    "train_scores = dict() ; test_scores  = dict() #to hold scores for viewing at the end"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:03.325255Z",
     "start_time": "2020-01-23T12:09:03.300687Z"
    }
   },
   "outputs": [],
   "source": [
    "def grid_search_return_hp_dict(model, X_train_gs, y_train_gs, param_grid):\n",
    "    \"\"\"Run GridSearchCV on given model with defined param_grid, return best params as dict.\n",
    "    Pre-assign sklearn model to variable. Pass param_grid as dict.\"\"\"\n",
    "    print(str(model).split(sep='(')[0], \"grid search.\")\n",
    "    GS = GridSearchCV(model, param_grid, n_jobs=2)\n",
    "    GS.fit(X_train_gs, y_train_gs)\n",
    "    print(GS.best_params_)\n",
    "    return(GS.best_params_)\n",
    "\n",
    "def train_model(model, feature_set, hyperparameters, \n",
    "                grid_search = False):\n",
    "    \"\"\"Wrapper for training models with optional grid search and feature selection.\n",
    "    Pass model with no brackets, ie Lars not Lars()\n",
    "    If grid_search = True, hyperparameters is the hp ranges dictionary. Outputs trained model, best_hp\n",
    "    If grid_search = False, hyperparameters is the hp dictionary. Outputs trained model\"\"\"\n",
    "    X_train, X_test = X_train_main[feature_set], X_test_main[feature_set]\n",
    "    \n",
    "    if grid_search == True:\n",
    "        best_hp = grid_search_return_hp_dict(model(),\n",
    "                                             X_train, y_train, hyperparameters)\n",
    "        print(\"Grid search complete!\\nHyperparameters:\\n\", best_hp)\n",
    "        training_model = model(**best_hp).fit(X_train, y_train)\n",
    "        return training_model, best_hp\n",
    "\n",
    "    if grid_search == False:\n",
    "        print(\"No grid search\\nHyperparameters:\\n\", hyperparameters)\n",
    "        training_model = model(**hyperparameters).fit(X_train, y_train)\n",
    "        return training_model\n",
    "\n",
    "def produce_exp_vs_pred_df(features_list, model, codename): # generalise!\n",
    "    \"\"\"Pass features_list as a string\"\"\"\n",
    "    pred_list = []\n",
    "    for i in range(0, len(dragon_features)):\n",
    "        prediction = model.predict([dragon_features[eval(features_list)].iloc[i]])\n",
    "        pred_list.append(prediction)\n",
    "    \n",
    "    exp_vs_calc = pd.DataFrame(constants_first)\n",
    "    exp_vs_calc['Predicted'] = pred_list\n",
    "    exp_vs_calc.rename({'Kh_first':'Experimental'}, inplace=True, axis=1)\n",
    "    exp_vs_calc.to_csv('models/DRAGON_noVPAS/%s'%(\"expvcalc_\" + str(model).split(sep='(')[0] + \"_\" + features_list[16:] + \".csv\"))\n",
    "    now = datetime.now()\n",
    "\n",
    "    dt_string = now.strftime(\"_%d_%m_%Y_%H_%M_%S\")\n",
    "    filename = \"models/DRAGON_noVPAS/\" + str(model).split(sep='(')[0] + \"_\" + features_list[16:] + dt_string + \".joblib\"\n",
    "    print(str(model).split(sep='(')[0], \"run at:\", now, \". Saving to\", filename)\n",
    "    \n",
    "    joblib.dump(model, filename)\n",
    "    return exp_vs_calc\n",
    "\n",
    "def prediction_plot_scores(model_func, pred_df):\n",
    "    \"\"\"Print train+test scores, then plot scatter of predicted vs actual HLCs. Uses {X/y}_{train/test},\n",
    "    redefining these variables will change the output.\"\"\"\n",
    "    fig = plt.figure(figsize=(10, 10))\n",
    "    scatter = plt.scatter(pred_df['Experimental'], pred_df['Predicted'])\n",
    "    plt.xlabel('Experimental')\n",
    "    plt.ylabel('Predicted')\n",
    "    plt.title('%s predictions of HLCs' %str(model_func).split(sep='(')[0])\n",
    "    labels = ['{}'.format(i) for i in species_names]\n",
    "    tooltip = mpld3.plugins.PointLabelTooltip(scatter, labels=labels)\n",
    "    mpld3.plugins.connect(fig, tooltip)\n",
    "    plt.plot([-30, 10], [-30, 10], c='red')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# DRAGON descriptors\n",
    "## Bringing in data and splitting it into parts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:03.898548Z",
     "start_time": "2020-01-23T12:09:03.329263Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input Shape (1404, 1479)\n",
      "Removed NaN, new shape (1355, 1479)\n"
     ]
    }
   ],
   "source": [
    "csv = pd.read_csv('organics_noVPAS_desc_kh.csv') # contains non-VP/AS HLCs\n",
    "\n",
    "print(\"Input Shape\", csv.shape) #input shape\n",
    "\n",
    "csv.dropna(axis=0, inplace=True)\n",
    "\n",
    "print(\"Removed NaN, new shape\", csv.shape) #removed NaN shape\n",
    "\n",
    "#species_names = csv.pop('Unnamed: 0')\n",
    "smiles_strings = csv.pop('SMILES')\n",
    "constants_mean = csv.pop('Kh_mean')\n",
    "constants_first = csv.pop('Kh_first')\n",
    "\n",
    "varying_columns = csv[['Varying_1', 'Varying_2', 'Varying_3', 'Varying_4',\n",
    " 'Varying_5', 'Varying_6', 'Varying_7', 'Varying_8', 'Varying_9']] # popping one-hot encoding columns\n",
    "\n",
    "dragon_features = csv.drop(['Varying_1', 'Varying_2', 'Varying_3', 'Varying_4',\n",
    " 'Varying_5', 'Varying_6', 'Varying_7', 'Varying_8', 'Varying_9'], axis=1) #seperating features\n",
    "\n",
    "X_train_main, X_test_main, y_train, y_test = train_test_split(dragon_features, constants_first, test_size=0.3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:06.055662Z",
     "start_time": "2020-01-23T12:09:03.901558Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\pandas\\core\\frame.py:4117: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  errors=errors,\n",
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\ipykernel_launcher.py:3: FutureWarning: The signature of `Series.to_csv` was aligned to that of `DataFrame.to_csv`, and argument 'header' will change its default value from False to True: please pass an explicit value to suppress this warning.\n",
      "  This is separate from the ipykernel package so we can avoid doing imports until\n"
     ]
    }
   ],
   "source": [
    "for i in ['X_train_main', 'X_test_main', 'y_train', 'y_test']:\n",
    "    #joblib.dump(exec('i'), \"models/DRAGON_noVPAS/%s.joblib\"%i)\n",
    "    eval('%s'%i).to_csv('models/DRAGON_noVPAS/%s.csv'%i)\n",
    "    if 'X' in i:\n",
    "        eval('%s'%i).drop('Unnamed: 0', axis=1, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature Selection\n",
    "Made aware of features selection methods in scikitlearn via https://towardsdatascience.com/feature-selection-techniques-in-machine-learning-with-python-f24e7da3f36e\n",
    "\n",
    "See also [scikit-learn docs](https://scikit-learn.org/stable/modules/feature_selection.html)\n",
    "### Importances via 'dummy' GradientBoostingRegressor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:10.927122Z",
     "start_time": "2020-01-23T12:09:06.058671Z"
    }
   },
   "outputs": [],
   "source": [
    "fs_model = GradientBoostingRegressor()\n",
    "fs_model.fit(X_train_main, y_train)\n",
    "importances = pd.Series(fs_model.feature_importances_, index = X_train_main.columns)\n",
    "top_15_features_model = list(importances.nlargest(15).index.values)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Importances via 'dummy' LASSO model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:18.586561Z",
     "start_time": "2020-01-23T12:09:10.929128Z"
    }
   },
   "outputs": [],
   "source": [
    "dummy_lasso = Lasso(max_iter=10000, alpha=0.5)\n",
    "selector = SelectFromModel(dummy_lasso).fit(X_train_main, y_train)\n",
    "coefs = selector.estimator_.coef_\n",
    "importances_lasso = pd.Series(coefs, index = X_train_main.columns)\n",
    "top_15_features_lasso = list(importances_lasso.nlargest(15).index.values)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### SelectKBest using f_regression scoring function\n",
    "F-test estimates degree of linear dependency"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:18.639703Z",
     "start_time": "2020-01-23T12:09:18.588066Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\feature_selection\\univariate_selection.py:299: RuntimeWarning: invalid value encountered in true_divide\n",
      "  corr /= X_norms\n",
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\scipy\\stats\\_distn_infrastructure.py:901: RuntimeWarning: invalid value encountered in greater\n",
      "  return (a < x) & (x < b)\n",
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\scipy\\stats\\_distn_infrastructure.py:901: RuntimeWarning: invalid value encountered in less\n",
      "  return (a < x) & (x < b)\n",
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\scipy\\stats\\_distn_infrastructure.py:1892: RuntimeWarning: invalid value encountered in less_equal\n",
      "  cond2 = cond0 & (x <= _a)\n"
     ]
    }
   ],
   "source": [
    "fs_kbest = SelectKBest(score_func= f_regression, k=15)\n",
    "fit = fs_kbest.fit(X_train_main, y_train)\n",
    "scores_f = pd.Series(fit.scores_, index=X_train_main.columns)\n",
    "top_15_features_kbest = list(scores_f.nlargest(15).index.values)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### SelectKBest using mutual_info_regression\n",
    "Catch any statistical dependancy, but requires more samples than f_reg"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:42.219729Z",
     "start_time": "2020-01-23T12:09:18.644716Z"
    }
   },
   "outputs": [],
   "source": [
    "fs_mut_info = SelectKBest(score_func= mutual_info_regression, k=15)\n",
    "fit = fs_mut_info.fit(X_train_main, y_train)\n",
    "scores_mutual = pd.Series(fit.scores_, index=X_train_main.columns)\n",
    "top_15_features_mutual_info = list(scores_mutual.nlargest(15).index.values)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Feature Selection results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:43.033898Z",
     "start_time": "2020-01-23T12:09:42.223241Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, imp_plot = plt.subplots(nrows=1, ncols=4, figsize=(15, 10))\n",
    "imp_plot[0].barh(importances.nlargest(15).index.values, importances.nlargest(15))\n",
    "imp_plot[0].set_title('Dummy Model')\n",
    "imp_plot[1].barh(scores_mutual.nlargest(15).index.values, scores_mutual.nlargest(15))\n",
    "imp_plot[1].set_title('Mutual Information')\n",
    "imp_plot[2].barh(scores_f.nlargest(15).index.values, scores_f.nlargest(15))\n",
    "imp_plot[2].set_title('F Regression')\n",
    "imp_plot[3].barh(importances_lasso.nlargest(15).index.values, importances_lasso.nlargest(15))\n",
    "imp_plot[3].set_title('LASSO')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Exporting feature importances"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:43.201942Z",
     "start_time": "2020-01-23T12:09:43.036373Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['models/DRAGON_noVPAS/mutual_importances.joblib']"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "joblib.dump(importances, \"models/DRAGON_noVPAS/model_importances.joblib\")\n",
    "joblib.dump(importances_lasso, \"models/DRAGON_noVPAS/lasso_importances.joblib\")\n",
    "joblib.dump(scores_f, \"models/DRAGON_noVPAS/freg_importances.joblib\")\n",
    "joblib.dump(scores_mutual, \"models/DRAGON_noVPAS/mutual_importances.joblib\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## GradientBoosting\n",
    "### Model-chosen features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T12:09:43.302557Z",
     "start_time": "2020-01-23T12:09:43.204952Z"
    }
   },
   "outputs": [],
   "source": [
    "gbr_param_grid = {\n",
    "    'n_estimators': [500, 1000, 2000],\n",
    "    'max_depth': [2, 4, 6],\n",
    "    'min_samples_leaf': [3, 5, 9, 17],\n",
    "    'learning_rate': [0.1, 0.05, 0.02],\n",
    "    'max_features': [1.0, 0.3, 0.1],\n",
    "    'loss': ['ls', 'lad', 'huber']\n",
    "} #hyperparameters for each gbr grid search"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:00:04.214057Z",
     "start_time": "2020-01-23T12:09:43.305567Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GradientBoostingRegressor grid search.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'learning_rate': 0.05, 'loss': 'lad', 'max_depth': 6, 'max_features': 0.1, 'min_samples_leaf': 3, 'n_estimators': 1000}\n",
      "Grid search complete!\n",
      "Hyperparameters:\n",
      " {'learning_rate': 0.05, 'loss': 'lad', 'max_depth': 6, 'max_features': 0.1, 'min_samples_leaf': 3, 'n_estimators': 1000}\n",
      "GradientBoostingRegressor run at: 2020-01-23 13:00:04.100789 . Saving to models/DRAGON_noVPAS/GradientBoostingRegressor_model_23_01_2020_13_00_04.joblib\n"
     ]
    }
   ],
   "source": [
    "gbr_model, gbr_hp = train_model(model = GradientBoostingRegressor,\n",
    "           feature_set= top_15_features_model,\n",
    "            grid_search=True,\n",
    "           hyperparameters=gbr_param_grid)\n",
    "\n",
    "exp_vs_calc_gbr_model = produce_exp_vs_pred_df('top_15_features_model', gbr_model,\n",
    "                                         \"GradBoost, model-chosen features\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:00:06.764874Z",
     "start_time": "2020-01-23T13:00:04.219070Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GradientBoostingRegressor run at: 2020-01-23 13:00:06.686664 . Saving to models/DRAGON_noVPAS/GradientBoostingRegressor_model_23_01_2020_13_00_06.joblib\n"
     ]
    }
   ],
   "source": [
    "exp_vs_calc_gbr_model = produce_exp_vs_pred_df('top_15_features_model', gbr_model,\n",
    "                                              \"Gradboost, model features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F-regression features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:00:12.781821Z",
     "start_time": "2020-01-23T13:00:06.767852Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'learning_rate': 0.05, 'loss': 'lad', 'max_depth': 6, 'max_features': 0.1, 'min_samples_leaf': 3, 'n_estimators': 1000}\n",
      "GradientBoostingRegressor run at: 2020-01-23 13:00:12.712633 . Saving to models/DRAGON_noVPAS/GradientBoostingRegressor_kbest_23_01_2020_13_00_12.joblib\n"
     ]
    }
   ],
   "source": [
    "gbr_freg = train_model(model = GradientBoostingRegressor,\n",
    "           feature_set= top_15_features_kbest,\n",
    "            grid_search=False,\n",
    "           hyperparameters=gbr_hp)\n",
    "\n",
    "exp_vs_calc_gbr_freg = produce_exp_vs_pred_df('top_15_features_kbest', gbr_freg,\n",
    "                                                    \"GradBoost, f_reg features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mutual info features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:00:18.670487Z",
     "start_time": "2020-01-23T13:00:12.784830Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'learning_rate': 0.05, 'loss': 'lad', 'max_depth': 6, 'max_features': 0.1, 'min_samples_leaf': 3, 'n_estimators': 1000}\n",
      "GradientBoostingRegressor run at: 2020-01-23 13:00:18.603308 . Saving to models/DRAGON_noVPAS/GradientBoostingRegressor_mutual_info_23_01_2020_13_00_18.joblib\n"
     ]
    }
   ],
   "source": [
    "gbr_mu = train_model(model = GradientBoostingRegressor,\n",
    "           feature_set= top_15_features_mutual_info,\n",
    "            grid_search=False,\n",
    "           hyperparameters=gbr_hp)\n",
    "\n",
    "exp_vs_calc_gbr_mu = produce_exp_vs_pred_df('top_15_features_mutual_info', gbr_mu,\n",
    "                                            \"GradBoost, Mutual Info\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LASSO features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:00:24.432951Z",
     "start_time": "2020-01-23T13:00:18.673494Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'learning_rate': 0.05, 'loss': 'lad', 'max_depth': 6, 'max_features': 0.1, 'min_samples_leaf': 3, 'n_estimators': 1000}\n",
      "GradientBoostingRegressor run at: 2020-01-23 13:00:24.358619 . Saving to models/DRAGON_noVPAS/GradientBoostingRegressor_lasso_23_01_2020_13_00_24.joblib\n"
     ]
    }
   ],
   "source": [
    "gbr_lasso = train_model(model = GradientBoostingRegressor,\n",
    "           feature_set= top_15_features_lasso,\n",
    "            grid_search=False,\n",
    "           hyperparameters=gbr_hp)\n",
    "exp_vs_calc_gbr_lasso = produce_exp_vs_pred_df('top_15_features_lasso', gbr_lasso,\n",
    "                                               \"LASSO-features, GBR model\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Decision Tree\n",
    "### Model-Chosen Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:00:24.443981Z",
     "start_time": "2020-01-23T13:00:24.434957Z"
    }
   },
   "outputs": [],
   "source": [
    "dtr_param_grid = {\n",
    "    'criterion': ['mse', 'friedman_mse', 'mae'], #function measuring quality of a split\n",
    "    'max_depth': [5, 10, 20] , # max depth of tree\n",
    "    'max_features': [5, 10, 15], # N features to be considered when looking for split\n",
    "    'max_leaf_nodes': [None, 10, 15, 20, 30], #grows tree with N nodes in best-first fashion\n",
    "    'min_impurity_decrease': [0.0, 0.1],\n",
    "    'min_samples_leaf': [1, 2, 3], \n",
    "    'min_samples_split': [1.0, 2, 3],\n",
    "    'min_weight_fraction_leaf': [0.0, 0.1, 0.5] , \n",
    "    'splitter': ['best', 'random']\n",
    "} #hyperparameters for dtr gridsearch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:03:02.356562Z",
     "start_time": "2020-01-23T13:00:24.446988Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DecisionTreeRegressor grid search.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'criterion': 'friedman_mse', 'max_depth': 20, 'max_features': 15, 'max_leaf_nodes': None, 'min_impurity_decrease': 0.0, 'min_samples_leaf': 3, 'min_samples_split': 3, 'min_weight_fraction_leaf': 0.0, 'splitter': 'random'}\n",
      "Grid search complete!\n",
      "Hyperparameters:\n",
      " {'criterion': 'friedman_mse', 'max_depth': 20, 'max_features': 15, 'max_leaf_nodes': None, 'min_impurity_decrease': 0.0, 'min_samples_leaf': 3, 'min_samples_split': 3, 'min_weight_fraction_leaf': 0.0, 'splitter': 'random'}\n",
      "DecisionTreeRegressor run at: 2020-01-23 13:03:02.346536 . Saving to models/DRAGON_noVPAS/DecisionTreeRegressor_model_23_01_2020_13_03_02.joblib\n"
     ]
    }
   ],
   "source": [
    "dtr_model, dtr_hp = train_model(model = DecisionTreeRegressor,\n",
    "           feature_set= top_15_features_model,\n",
    "            grid_search=True,\n",
    "           hyperparameters=dtr_param_grid)\n",
    "\n",
    "\n",
    "exp_vs_calc_dtr_model = produce_exp_vs_pred_df('top_15_features_model', dtr_model,\n",
    "                                         \"Decision Tree, model-chosen features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F-regression features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:03:04.628606Z",
     "start_time": "2020-01-23T13:03:02.358567Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'criterion': 'friedman_mse', 'max_depth': 20, 'max_features': 15, 'max_leaf_nodes': None, 'min_impurity_decrease': 0.0, 'min_samples_leaf': 3, 'min_samples_split': 3, 'min_weight_fraction_leaf': 0.0, 'splitter': 'random'}\n",
      "DecisionTreeRegressor run at: 2020-01-23 13:03:04.621589 . Saving to models/DRAGON_noVPAS/DecisionTreeRegressor_kbest_23_01_2020_13_03_04.joblib\n"
     ]
    }
   ],
   "source": [
    "dtr_freg = train_model(model = DecisionTreeRegressor,\n",
    "           feature_set= top_15_features_kbest,\n",
    "            grid_search=False,\n",
    "           hyperparameters=dtr_hp)\n",
    "exp_vs_calc_dtr_freg = produce_exp_vs_pred_df('top_15_features_kbest', dtr_freg,\n",
    "                                         \"Decision Tree, freg features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mutual info features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:03:06.361544Z",
     "start_time": "2020-01-23T13:03:04.630612Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'criterion': 'friedman_mse', 'max_depth': 20, 'max_features': 15, 'max_leaf_nodes': None, 'min_impurity_decrease': 0.0, 'min_samples_leaf': 3, 'min_samples_split': 3, 'min_weight_fraction_leaf': 0.0, 'splitter': 'random'}\n",
      "DecisionTreeRegressor run at: 2020-01-23 13:03:06.354523 . Saving to models/DRAGON_noVPAS/DecisionTreeRegressor_mutual_info_23_01_2020_13_03_06.joblib\n"
     ]
    }
   ],
   "source": [
    "dtr_mu = train_model(model = DecisionTreeRegressor,\n",
    "           feature_set= top_15_features_mutual_info,\n",
    "            grid_search=False,\n",
    "           hyperparameters=dtr_hp)\n",
    "exp_vs_calc_dtr_mu = produce_exp_vs_pred_df('top_15_features_mutual_info', dtr_mu,\n",
    "                                         \"Decision Tree, mutual info features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LASSO features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:03:07.889611Z",
     "start_time": "2020-01-23T13:03:06.364553Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'criterion': 'friedman_mse', 'max_depth': 20, 'max_features': 15, 'max_leaf_nodes': None, 'min_impurity_decrease': 0.0, 'min_samples_leaf': 3, 'min_samples_split': 3, 'min_weight_fraction_leaf': 0.0, 'splitter': 'random'}\n",
      "DecisionTreeRegressor run at: 2020-01-23 13:03:07.882591 . Saving to models/DRAGON_noVPAS/DecisionTreeRegressor_lasso_23_01_2020_13_03_07.joblib\n"
     ]
    }
   ],
   "source": [
    "dtr_lasso = train_model(model = DecisionTreeRegressor,\n",
    "           feature_set= top_15_features_lasso,\n",
    "            grid_search=False,\n",
    "           hyperparameters=dtr_hp)\n",
    "\n",
    "exp_vs_calc_dtr_lasso = produce_exp_vs_pred_df('top_15_features_lasso', dtr_lasso,\n",
    "                                         \"Decision Tree, lasso features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## AdaBoost Regressor\n",
    "### Model Chosen Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:03:07.898633Z",
     "start_time": "2020-01-23T13:03:07.892617Z"
    }
   },
   "outputs": [],
   "source": [
    "base_models = [ExtraTreesRegressor(n_estimators= 5,\n",
    "                                   criterion= 'mse',\n",
    "                                   max_features = 'log2'),\n",
    "               RandomForestRegressor(n_estimators= 5,\n",
    "                                     criterion= 'mse',\n",
    "                                     max_features = 'sqrt',\n",
    "                                     min_samples_split = 3),\n",
    "              GradientBoostingRegressor(),\n",
    "              DecisionTreeRegressor(),\n",
    "              Lars(),\n",
    "              ElasticNet()]\n",
    "\n",
    "ada_param_grid = {\n",
    "    'base_estimator' : base_models, \n",
    "    'learning_rate' : [0.3, 0.5, 0.8, 1], \n",
    "    'loss' : ['linear', 'square', 'exponential'], \n",
    "    'n_estimators' : [50, 100]\n",
    "} #hyperparameters for adaboost gridsearch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:09:15.956598Z",
     "start_time": "2020-01-23T13:03:07.908661Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AdaBoostRegressor grid search.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'base_estimator': ExtraTreesRegressor(bootstrap=False, criterion='mse', max_depth=None,\n",
      "                    max_features='log2', 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=5, n_jobs=None,\n",
      "                    oob_score=False, random_state=None, verbose=0,\n",
      "                    warm_start=False), 'learning_rate': 0.5, 'loss': 'exponential', 'n_estimators': 100}\n",
      "Grid search complete!\n",
      "Hyperparameters:\n",
      " {'base_estimator': ExtraTreesRegressor(bootstrap=False, criterion='mse', max_depth=None,\n",
      "                    max_features='log2', 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=5, n_jobs=None,\n",
      "                    oob_score=False, random_state=None, verbose=0,\n",
      "                    warm_start=False), 'learning_rate': 0.5, 'loss': 'exponential', 'n_estimators': 100}\n",
      "AdaBoostRegressor run at: 2020-01-23 13:09:15.692894 . Saving to models/DRAGON_noVPAS/AdaBoostRegressor_model_23_01_2020_13_09_15.joblib\n"
     ]
    }
   ],
   "source": [
    "ada_model, ada_hp = train_model(model = AdaBoostRegressor,\n",
    "           feature_set= top_15_features_model,\n",
    "            grid_search=True,\n",
    "           hyperparameters=ada_param_grid)\n",
    "\n",
    "exp_vs_calc_ada_model = produce_exp_vs_pred_df('top_15_features_model', ada_model,\n",
    "                                               \"AdaBoost, model features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F-regression features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:10:42.666775Z",
     "start_time": "2020-01-23T13:09:15.961614Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'base_estimator': ExtraTreesRegressor(bootstrap=False, criterion='mse', max_depth=None,\n",
      "                    max_features='log2', 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=5, n_jobs=None,\n",
      "                    oob_score=False, random_state=None, verbose=0,\n",
      "                    warm_start=False), 'learning_rate': 0.5, 'loss': 'exponential', 'n_estimators': 100}\n",
      "AdaBoostRegressor run at: 2020-01-23 13:10:42.381717 . Saving to models/DRAGON_noVPAS/AdaBoostRegressor_kbest_23_01_2020_13_10_42.joblib\n"
     ]
    }
   ],
   "source": [
    "ada_freg = train_model(model = AdaBoostRegressor,\n",
    "           feature_set= top_15_features_kbest,\n",
    "            grid_search=False,\n",
    "           hyperparameters=ada_hp)\n",
    "\n",
    "exp_vs_calc_ada_freg = produce_exp_vs_pred_df('top_15_features_kbest', ada_freg,\n",
    "                                               \"AdaBoost, freg features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mutual info features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:11:58.854433Z",
     "start_time": "2020-01-23T13:10:42.668748Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'base_estimator': ExtraTreesRegressor(bootstrap=False, criterion='mse', max_depth=None,\n",
      "                    max_features='log2', 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=5, n_jobs=None,\n",
      "                    oob_score=False, random_state=None, verbose=0,\n",
      "                    warm_start=False), 'learning_rate': 0.5, 'loss': 'exponential', 'n_estimators': 100}\n",
      "AdaBoostRegressor run at: 2020-01-23 13:11:58.615799 . Saving to models/DRAGON_noVPAS/AdaBoostRegressor_mutual_info_23_01_2020_13_11_58.joblib\n"
     ]
    }
   ],
   "source": [
    "ada_mu = train_model(model = AdaBoostRegressor,\n",
    "           feature_set= top_15_features_mutual_info,\n",
    "            grid_search=False,\n",
    "           hyperparameters=ada_hp)\n",
    "\n",
    "exp_vs_calc_ada_mu = produce_exp_vs_pred_df('top_15_features_mutual_info', ada_mu,\n",
    "                                               \"AdaBoost, mutual info features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LASSO features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:13:36.523268Z",
     "start_time": "2020-01-23T13:11:58.856438Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'base_estimator': ExtraTreesRegressor(bootstrap=False, criterion='mse', max_depth=None,\n",
      "                    max_features='log2', 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=5, n_jobs=None,\n",
      "                    oob_score=False, random_state=None, verbose=0,\n",
      "                    warm_start=False), 'learning_rate': 0.5, 'loss': 'exponential', 'n_estimators': 100}\n",
      "AdaBoostRegressor run at: 2020-01-23 13:13:36.118190 . Saving to models/DRAGON_noVPAS/AdaBoostRegressor_lasso_23_01_2020_13_13_36.joblib\n"
     ]
    }
   ],
   "source": [
    "ada_lasso = train_model(model = AdaBoostRegressor,\n",
    "           feature_set= top_15_features_lasso,\n",
    "            grid_search=False,\n",
    "           hyperparameters=ada_hp)\n",
    "\n",
    "exp_vs_calc_ada_lasso = produce_exp_vs_pred_df('top_15_features_lasso', ada_lasso,\n",
    "                                               \"AdaBoost, model features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## LASSO\n",
    "### Model Chosen Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:13:36.530288Z",
     "start_time": "2020-01-23T13:13:36.525274Z"
    }
   },
   "outputs": [],
   "source": [
    "lasso_param_grid = {\n",
    "    'alpha':[0.2, 0.4, 0.6, 0.8],\n",
    "    'max_iter':[1000, 5000, 10000, 50000],\n",
    "    'selection':['cyclic', 'random']\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:13:43.802634Z",
     "start_time": "2020-01-23T13:13:36.533296Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Lasso grid search.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'alpha': 0.2, 'max_iter': 10000, 'selection': 'random'}\n",
      "Grid search complete!\n",
      "Hyperparameters:\n",
      " {'alpha': 0.2, 'max_iter': 10000, 'selection': 'random'}\n",
      "Lasso run at: 2020-01-23 13:13:43.795617 . Saving to models/DRAGON_noVPAS/Lasso_model_23_01_2020_13_13_43.joblib\n"
     ]
    }
   ],
   "source": [
    "lasso_model, lasso_hp = train_model(model = Lasso,\n",
    "           feature_set= top_15_features_model,\n",
    "            grid_search=True,\n",
    "           hyperparameters=lasso_param_grid)\n",
    "\n",
    "\n",
    "exp_vs_calc_lasso_model = produce_exp_vs_pred_df('top_15_features_model', lasso_model,\n",
    "                                                 \"LASSO, model features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F-regression features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:13:46.080695Z",
     "start_time": "2020-01-23T13:13:43.805643Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'alpha': 0.2, 'max_iter': 10000, 'selection': 'random'}\n",
      "Lasso run at: 2020-01-23 13:13:46.058636 . Saving to models/DRAGON_noVPAS/Lasso_kbest_23_01_2020_13_13_46.joblib\n"
     ]
    }
   ],
   "source": [
    "lasso_freg = train_model(model = Lasso,\n",
    "           feature_set= top_15_features_kbest,\n",
    "            grid_search=False,\n",
    "           hyperparameters=lasso_hp)\n",
    "\n",
    "exp_vs_calc_lasso_freg = produce_exp_vs_pred_df('top_15_features_kbest', lasso_freg,\n",
    "                                                 \"LASSO, f-reg features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mutual info features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:13:48.388836Z",
     "start_time": "2020-01-23T13:13:46.082701Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'alpha': 0.2, 'max_iter': 10000, 'selection': 'random'}\n",
      "Lasso run at: 2020-01-23 13:13:48.382821 . Saving to models/DRAGON_noVPAS/Lasso_mutual_info_23_01_2020_13_13_48.joblib\n"
     ]
    }
   ],
   "source": [
    "lasso_mu = train_model(model = Lasso,\n",
    "           feature_set= top_15_features_mutual_info,\n",
    "            grid_search=False,\n",
    "           hyperparameters=lasso_hp)\n",
    "\n",
    "exp_vs_calc_lasso_mu = produce_exp_vs_pred_df('top_15_features_mutual_info', lasso_mu,\n",
    "                                                 \"LASSO, mutual info features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LASSO Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:13:50.994770Z",
     "start_time": "2020-01-23T13:13:48.390841Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'alpha': 0.2, 'max_iter': 10000, 'selection': 'random'}\n",
      "Lasso run at: 2020-01-23 13:13:50.963685 . Saving to models/DRAGON_noVPAS/Lasso_lasso_23_01_2020_13_13_50.joblib\n"
     ]
    }
   ],
   "source": [
    "lasso_lasso = train_model(model = Lasso,\n",
    "           feature_set= top_15_features_lasso,\n",
    "            grid_search=False,\n",
    "           hyperparameters=lasso_hp)\n",
    "\n",
    "exp_vs_calc_lasso_lasso = produce_exp_vs_pred_df('top_15_features_lasso', lasso_lasso,\n",
    "                                                 \"LASSO, lasso features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest Regressor\n",
    "### Model Chosen Features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T13:13:51.004795Z",
     "start_time": "2020-01-23T13:13:50.997777Z"
    }
   },
   "outputs": [],
   "source": [
    "random_forest_param_grid = {\n",
    "    'max_depth': [10, 20, 30, 40, 50, 60, 70, 80, 90, None],\n",
    "    'max_features': ['auto', 'sqrt'],\n",
    "    'min_samples_leaf': [1, 2, 4],\n",
    "    'min_samples_split': [2, 5, 10],\n",
    "    'n_estimators': [200, 400, 600, 800, 1000, 1200, 1400, 1600]\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:51:57.425086Z",
     "start_time": "2020-01-23T13:13:51.008809Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "RandomForestRegressor grid search.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'max_depth': None, 'max_features': 'auto', 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}\n",
      "Grid search complete!\n",
      "Hyperparameters:\n",
      " {'max_depth': None, 'max_features': 'auto', 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}\n",
      "RandomForestRegressor run at: 2020-01-23 14:51:57.317264 . Saving to models/DRAGON_noVPAS/RandomForestRegressor_model_23_01_2020_14_51_57.joblib\n"
     ]
    }
   ],
   "source": [
    "random_forest_model, rfr_hp = train_model(model = RandomForestRegressor,\n",
    "           feature_set= top_15_features_model,\n",
    "            grid_search=True,\n",
    "           hyperparameters=random_forest_param_grid)\n",
    "exp_vs_calc_random_forest_model = produce_exp_vs_pred_df('top_15_features_model', random_forest_model,\n",
    "                                                         \"RandomForest, model features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F-regression features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:52:17.438348Z",
     "start_time": "2020-01-23T14:51:57.429563Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'max_depth': None, 'max_features': 'auto', 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}\n",
      "RandomForestRegressor run at: 2020-01-23 14:52:17.353619 . Saving to models/DRAGON_noVPAS/RandomForestRegressor_kbest_23_01_2020_14_52_17.joblib\n"
     ]
    }
   ],
   "source": [
    "random_forest_freg = train_model(model = RandomForestRegressor,\n",
    "           feature_set= top_15_features_kbest,\n",
    "            grid_search=False,\n",
    "           hyperparameters=rfr_hp)\n",
    "exp_vs_calc_random_forest_freg = produce_exp_vs_pred_df('top_15_features_kbest', random_forest_freg,\n",
    "                                                         \"RandomForest, freg features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mutual info features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:52:38.440080Z",
     "start_time": "2020-01-23T14:52:17.441389Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'max_depth': None, 'max_features': 'auto', 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}\n",
      "RandomForestRegressor run at: 2020-01-23 14:52:38.345342 . Saving to models/DRAGON_noVPAS/RandomForestRegressor_mutual_info_23_01_2020_14_52_38.joblib\n"
     ]
    }
   ],
   "source": [
    "random_forest_mu = train_model(model = RandomForestRegressor,\n",
    "           feature_set= top_15_features_mutual_info,\n",
    "            grid_search=False,\n",
    "           hyperparameters=rfr_hp)\n",
    "exp_vs_calc_random_forest_mu = produce_exp_vs_pred_df('top_15_features_mutual_info', random_forest_mu,\n",
    "                                                         \"RandomForest, mutual info features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LASSO features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:52:58.501705Z",
     "start_time": "2020-01-23T14:52:38.442586Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'max_depth': None, 'max_features': 'auto', 'min_samples_leaf': 1, 'min_samples_split': 5, 'n_estimators': 200}\n",
      "RandomForestRegressor run at: 2020-01-23 14:52:58.403439 . Saving to models/DRAGON_noVPAS/RandomForestRegressor_lasso_23_01_2020_14_52_58.joblib\n"
     ]
    }
   ],
   "source": [
    "random_forest_lasso = train_model(model = RandomForestRegressor,\n",
    "           feature_set= top_15_features_lasso,\n",
    "            grid_search=False,\n",
    "           hyperparameters=rfr_hp)\n",
    "exp_vs_calc_random_forest_lasso = produce_exp_vs_pred_df('top_15_features_lasso', random_forest_lasso,\n",
    "                                                         \"RandomForest, lasso features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## LARS - Least angle regression\n",
    "### Model chosen features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:52:58.509223Z",
     "start_time": "2020-01-23T14:52:58.504711Z"
    }
   },
   "outputs": [],
   "source": [
    "lars_param_grid = {'n_nonzero_coefs': [0, 5, 10, 15]}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:12.844262Z",
     "start_time": "2020-01-23T14:52:58.512733Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Lars grid search.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n",
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\linear_model\\least_angle.py:578: ConvergenceWarning: Regressors in active set degenerate. Dropping a regressor, after 4 iterations, i.e. alpha=8.143e-02, with an active set of 4 regressors, and the smallest cholesky pivot element being 1.490e-08. Reduce max_iter or increase eps parameters.\n",
      "  ConvergenceWarning)\n",
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\linear_model\\least_angle.py:578: ConvergenceWarning: Regressors in active set degenerate. Dropping a regressor, after 4 iterations, i.e. alpha=8.143e-02, with an active set of 4 regressors, and the smallest cholesky pivot element being 1.490e-08. Reduce max_iter or increase eps parameters.\n",
      "  ConvergenceWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'n_nonzero_coefs': 10}\n",
      "Grid search complete!\n",
      "Hyperparameters:\n",
      " {'n_nonzero_coefs': 10}\n",
      "Lars run at: 2020-01-23 14:53:12.838246 . Saving to models/DRAGON_noVPAS/Lars_model_23_01_2020_14_53_12.joblib\n"
     ]
    }
   ],
   "source": [
    "lars_model, lars_hp = train_model(model = Lars,\n",
    "           feature_set= top_15_features_model,\n",
    "            grid_search=True,\n",
    "           hyperparameters= lars_param_grid)\n",
    "exp_vs_calc_lars_model = produce_exp_vs_pred_df('top_15_features_model', lars_model, \"LARS model features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F-regression features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:14.462418Z",
     "start_time": "2020-01-23T14:53:12.847271Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'n_nonzero_coefs': 10}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\linear_model\\least_angle.py:578: ConvergenceWarning: Regressors in active set degenerate. Dropping a regressor, after 7 iterations, i.e. alpha=1.905e-02, with an active set of 7 regressors, and the smallest cholesky pivot element being 1.490e-08. Reduce max_iter or increase eps parameters.\n",
      "  ConvergenceWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Lars run at: 2020-01-23 14:53:14.455897 . Saving to models/DRAGON_noVPAS/Lars_kbest_23_01_2020_14_53_14.joblib\n"
     ]
    }
   ],
   "source": [
    "lars_freg = train_model(model = Lars,\n",
    "           feature_set= top_15_features_kbest,\n",
    "            grid_search=False,\n",
    "           hyperparameters= lars_hp)\n",
    "exp_vs_calc_lars_freg = produce_exp_vs_pred_df('top_15_features_kbest', lars_freg, \"LARS freg features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mutual info features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:16.193523Z",
     "start_time": "2020-01-23T14:53:14.465925Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'n_nonzero_coefs': 10}\n",
      "Lars run at: 2020-01-23 14:53:16.186513 . Saving to models/DRAGON_noVPAS/Lars_mutual_info_23_01_2020_14_53_16.joblib\n"
     ]
    }
   ],
   "source": [
    "lars_mu = train_model(model = Lars,\n",
    "           feature_set= top_15_features_mutual_info,\n",
    "            grid_search=False,\n",
    "           hyperparameters= lars_hp)\n",
    "\n",
    "exp_vs_calc_lars_mu = produce_exp_vs_pred_df('top_15_features_mutual_info', lars_mu, \"LARS mutual info features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LASSO features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:17.953204Z",
     "start_time": "2020-01-23T14:53:16.196531Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'n_nonzero_coefs': 10}\n",
      "Lars run at: 2020-01-23 14:53:17.945212 . Saving to models/DRAGON_noVPAS/Lars_lasso_23_01_2020_14_53_17.joblib\n"
     ]
    }
   ],
   "source": [
    "lars_lasso = train_model(model = Lars,\n",
    "           feature_set= top_15_features_lasso,\n",
    "            grid_search=False,\n",
    "           hyperparameters= lars_hp)\n",
    "exp_vs_calc_lars_lasso = produce_exp_vs_pred_df('top_15_features_lasso', lars_lasso, \"LARS lasso features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Elastic Net (linear regression l1 l2 norm regularization)\n",
    "### Model chosen features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:17.962229Z",
     "start_time": "2020-01-23T14:53:17.956214Z"
    }
   },
   "outputs": [],
   "source": [
    "net_param_grid = {\n",
    "    'alpha':[0.1, 0.3, 0.5, 0.8, 1.0],\n",
    "    'l1_ratio':[0.0, 0.2, 0.4, 0.6, 0.8, 1.0],\n",
    "    'fit_intercept':[True, False],\n",
    "    'max_iter': [1000, 3000, 5000],\n",
    "    'tol':[0.0001, 0.001, 0.01],\n",
    "    'selection':['cyclic', 'random']\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:41.415936Z",
     "start_time": "2020-01-23T14:53:17.965237Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ElasticNet grid search.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_split.py:1978: FutureWarning: The default value of cv will change from 3 to 5 in version 0.22. Specify it explicitly to silence this warning.\n",
      "  warnings.warn(CV_WARNING, FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'alpha': 0.1, 'fit_intercept': True, 'l1_ratio': 1.0, 'max_iter': 1000, 'selection': 'random', 'tol': 0.01}\n",
      "Grid search complete!\n",
      "Hyperparameters:\n",
      " {'alpha': 0.1, 'fit_intercept': True, 'l1_ratio': 1.0, 'max_iter': 1000, 'selection': 'random', 'tol': 0.01}\n",
      "ElasticNet run at: 2020-01-23 14:53:41.408913 . Saving to models/DRAGON_noVPAS/ElasticNet_model_23_01_2020_14_53_41.joblib\n"
     ]
    }
   ],
   "source": [
    "net_model, net_hp = train_model(model = ElasticNet,\n",
    "           feature_set= top_15_features_model,\n",
    "            grid_search=True,\n",
    "           hyperparameters= net_param_grid)\n",
    "exp_vs_calc_net_model = produce_exp_vs_pred_df('top_15_features_model', net_model, \"ElasticNet model features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### F regression features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:43.221738Z",
     "start_time": "2020-01-23T14:53:41.418943Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'alpha': 0.1, 'fit_intercept': True, 'l1_ratio': 1.0, 'max_iter': 1000, 'selection': 'random', 'tol': 0.01}\n",
      "ElasticNet run at: 2020-01-23 14:53:43.214220 . Saving to models/DRAGON_noVPAS/ElasticNet_kbest_23_01_2020_14_53_43.joblib\n"
     ]
    }
   ],
   "source": [
    "net_freg = train_model(model = ElasticNet,\n",
    "           feature_set= top_15_features_kbest,\n",
    "            grid_search=False,\n",
    "           hyperparameters= net_hp)\n",
    "exp_vs_calc_net_freg = produce_exp_vs_pred_df('top_15_features_kbest', net_freg, \"ElasticNet freg features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Mutual info features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:44.946024Z",
     "start_time": "2020-01-23T14:53:43.225247Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'alpha': 0.1, 'fit_intercept': True, 'l1_ratio': 1.0, 'max_iter': 1000, 'selection': 'random', 'tol': 0.01}\n",
      "ElasticNet run at: 2020-01-23 14:53:44.939507 . Saving to models/DRAGON_noVPAS/ElasticNet_mutual_info_23_01_2020_14_53_44.joblib\n"
     ]
    }
   ],
   "source": [
    "net_mu = train_model(model = ElasticNet,\n",
    "           feature_set= top_15_features_mutual_info,\n",
    "            grid_search=False,\n",
    "           hyperparameters= net_hp)\n",
    "exp_vs_calc_net_mu = produce_exp_vs_pred_df('top_15_features_mutual_info', net_mu, \"ElasticNet mutual info features\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### LASSO features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2020-01-23T14:53:46.646047Z",
     "start_time": "2020-01-23T14:53:44.949032Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No grid search\n",
      "Hyperparameters:\n",
      " {'alpha': 0.1, 'fit_intercept': True, 'l1_ratio': 1.0, 'max_iter': 1000, 'selection': 'random', 'tol': 0.01}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\jtd1g16\\AppData\\Local\\Continuum\\anaconda3\\lib\\site-packages\\sklearn\\linear_model\\coordinate_descent.py:475: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations. Duality gap: 1694.4525810404884, tolerance: 401.20752102064785\n",
      "  positive)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ElasticNet run at: 2020-01-23 14:53:46.639029 . Saving to models/DRAGON_noVPAS/ElasticNet_lasso_23_01_2020_14_53_46.joblib\n"
     ]
    }
   ],
   "source": [
    "net_lasso = train_model(model = ElasticNet,\n",
    "           feature_set= top_15_features_lasso,\n",
    "            grid_search=False,\n",
    "           hyperparameters= net_hp)\n",
    "exp_vs_calc_net_lasso = produce_exp_vs_pred_df('top_15_features_lasso', net_lasso, \"ElasticNet lasso features\")"
   ]
  }
 ],
 "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": "3.7.3"
  },
  "notify_time": "30",
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {
    "height": "calc(100% - 180px)",
    "left": "10px",
    "top": "150px",
    "width": "303.516px"
   },
   "toc_section_display": true,
   "toc_window_display": false
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}