Health Fraud - III. Machine Learning.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "15c5589c",
   "metadata": {},
   "source": [
    "# Machine Learning Processing\n",
    "\n",
    "### Contents\n",
    "\n",
    " - I. [Data Importing & Processing](#Data-Import-and-Processing)\n",
    " - II. [Modeling Plan](#Modeling-Plan)\n",
    "     - [Iteration 0 - Unprocessed](#Iteration-0:-Non-Scaled-Data-for-Tree-based-Boosting-Models)\n",
    "     - [Iteration 1 - Scaled](#Iteration-1:-Baby's-First-Pipe)\n",
    "     - [Iteration 2 - PCA](#Iteration-2:-Give-the-Computer-a-Chance)\n",
    "     - [Iteration 3 - PCA Boosted](#Iteration-3:-PCABoosted-Modeling)\n",
    "     - [Best Models](#Best-Models)\n",
    " - III. [Preliminary Analysis](#Preliminary-Results-&-Analysis)\n",
    "     - [Permutation Importance Insights](#Permutation-Importance)\n",
    " - IV. [Stacking](#Running-the-Best-Models-for-Full-Training-Data)\n",
    "     - [Confidence Visualization](#Stacking-Visualization)\n",
    " - V. [Conclusions & Discussion](#Conclusions-&-Discussion)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a22f8be7",
   "metadata": {},
   "source": [
    "## Data Import and Processing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "88e3f429",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Import packages etc.\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from scipy import stats\n",
    "import pickle\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "0345fb83",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Data shape: (4328, 71)\n",
      "Test Data Shape: (1082, 71)\n"
     ]
    }
   ],
   "source": [
    "Xdata = pickle.load(open('Xtrain.pickle', 'rb'))\n",
    "ydata = pickle.load(open('ytrain.pickle', 'rb'))\n",
    "\n",
    "# Split the data into train & test sets for supervised learning\n",
    "Xtrain, Xtest, ytrain, ytest = train_test_split(Xdata, ydata,\n",
    "                                               test_size=0.2,\n",
    "                                               random_state=430)\n",
    "\n",
    "# Quick Heads Up\n",
    "print('Training Data shape: ' + str(Xtrain.shape))\n",
    "print('Test Data Shape: ' + str(Xtest.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "revolutionary-medicaid",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "PotentialFraud    506\n",
      "dtype: int64\n",
      "Total for len(ydata)\n"
     ]
    }
   ],
   "source": [
    "print(np.sum(ydata))\n",
    "print('Total for len(ydata)')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "aging-crown",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "PotentialFraud    395\n",
       "dtype: int64"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.sum(ytrain)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c1e25a40",
   "metadata": {},
   "source": [
    "## Modeling Plan\n",
    "The general strategy of our approach incorporates an initial, broad spectrum testing of models, including the following.\n",
    " - Logistic Regression\n",
    " - KNN\n",
    " - Discriminant Analysis\n",
    " - Naive Bayes\n",
    " - SVM\n",
    " - CatBoost\n",
    " - Light GBM\n",
    "\n",
    "This grouping should represent a decent collection of well-worn, classical methods with exceptionally long history & documentation as well as newer, useful methodologies. Of particular interest to this last point will be **Catboost & Light GBM**, two Gradient Boosting models that have shown a great deal of promise in terms of efficiency & accuracy.\n",
    "\n",
    "### Iterative Strategy | What We Care About \n",
    " - Iteration 0: Non-Scaled models - Naive modeling with tree-based boosting models\n",
    " - Iteration 1: Scaled models\n",
    " - Iteration 2: PCA transformed models\n",
    " - Iteration 3: PCA Boosting models (PCA features reattached to original data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "electronic-stamp",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "from sklearn.naive_bayes import GaussianNB\n",
    "from sklearn.svm import SVC\n",
    "from catboost import CatBoostClassifier\n",
    "from lightgbm import LGBMClassifier\n",
    "from sklearn.model_selection import GridSearchCV"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "db6e6268",
   "metadata": {},
   "source": [
    "### Iteration 0: Non-Scaled Data for Tree-based Boosting Models\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "Both Catboost & Light GBM are capable of handling unscaled and even non-numeric data, and encourage testing their models with datasets that are not extensively transformed. As such, for these 2 models, we have run a separate iteration with non-scaled data. For further reference, please consult their respective documentation. Here we will process these for some initial yield results."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "facial-columbus",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'depth': 8, 'min_data_in_leaf': 1}\n"
     ]
    }
   ],
   "source": [
    "# IT0: CatBoost Classifier\n",
    "cb_0 = CatBoostClassifier(verbose = False)\n",
    "cb_0.fit(Xtrain, ytrain)\n",
    "cb_0_train, cb_0_test = cb_0.score(Xtrain, ytrain), cb_0.score(Xtest, ytest)\n",
    "\n",
    "cb_grid = {'depth': [6, 8, 10],\n",
    "           'min_data_in_leaf': [1, 2, 4]}\n",
    "cb_0_opt = GridSearchCV(cb_0, cb_grid, n_jobs=-2, cv=3)\n",
    "cb_0_opt.fit(Xtrain, ytrain)\n",
    "cb_0_opt_train, cb_0_opt_test, cb_0_opt_params = cb_0_opt.score(Xtrain, ytrain), cb_0_opt.score(Xtest, ytest), cb_0_opt.best_params_\n",
    "\n",
    "print('Best Params:', cb_0_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "nonprofit-tuner",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/theodore/anaconda3/lib/python3.8/site-packages/joblib/externals/loky/process_executor.py:688: UserWarning: A worker stopped while some jobs were given to the executor. This can be caused by a too short worker timeout or by a memory leak.\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Warning] min_data_in_leaf is set=17, min_child_samples=20 will be ignored. Current value: min_data_in_leaf=17\n",
      "Best Params: {'max_depth': 13, 'min_data_in_leaf': 17, 'num_leaves': 46}\n"
     ]
    }
   ],
   "source": [
    "# IT0: Light Gradient Boosting Machine Classifier\n",
    "lgbm_0 = LGBMClassifier()\n",
    "lgbm_0.fit(Xtrain, ytrain.values.ravel())\n",
    "lgbm_0_train, lgbm_0_test = lgbm_0.score(Xtrain, ytrain.values.ravel()), lgbm_0.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "lgbm_grid = {'num_leaves': [44, 46, 48],\n",
    "             'max_depth': [9, 11, 13],\n",
    "             'min_data_in_leaf': [15, 17, 19]}\n",
    "lgbm_0_opt = GridSearchCV(lgbm_0, lgbm_grid, n_jobs   = -2, cv=3)\n",
    "lgbm_0_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "lgbm_0_opt_train, lgbm_0_opt_test, lgbm_0_opt_params = lgbm_0_opt.score(Xtrain, ytrain.values.ravel()), lgbm_0_opt.score(Xtest, ytest.values.ravel()), lgbm_0_opt.best_params_\n",
    "\n",
    "print('Best Params:', lgbm_0_opt.best_params_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "equipped-israeli",
   "metadata": {},
   "source": [
    "Creating an iterative list of all new models' train and tests scores generated through hypertransformation tuning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "photographic-designer",
   "metadata": {},
   "outputs": [],
   "source": [
    "modellist = [ele for ele in ['cb', 'lgbm', 'lgr', 'knn', 'lda', 'gnb', 'svc'] for i in range(16)]\n",
    "iterlist = [ele for ele in ['_0', '_1', '_2', '_3'] for i in range(2)] * 14\n",
    "optlist = [ele for ele in ['', '_opt'] for i in range(8)] * 7\n",
    "traintestlist = ['_train', '_test'] * 56\n",
    "for idx in range(len(modellist)):\n",
    "    modellist[idx] = modellist[idx] + iterlist[idx] + optlist[idx] + traintestlist[idx]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "lyric-holocaust",
   "metadata": {},
   "source": [
    "Creating a function to tabulate all train and test scores generated by the hypertransformation tuning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "western-ethernet",
   "metadata": {},
   "outputs": [],
   "source": [
    "def scoring_table(score_list):\n",
    "    columns = [[ele for ele in ['Iteration0', 'Iteration1', 'Iteration2', 'Iteration3'] for i in range(2)],\n",
    "               [ele for ele in ['NonScaled', 'Scaled', 'PCA', 'PCABoost'] for i in range(2)],\n",
    "               ['Train', 'Test'] * 4]\n",
    "    c_tuples = list(zip(*columns))\n",
    "    multi_columns = pd.MultiIndex.from_tuples(c_tuples)\n",
    "\n",
    "    index = [[ele for ele in ['CatBoost', 'LightGBM', 'Logistic', 'KNN', 'LDA', 'GNB', 'SVC'] for i in range(2)], \n",
    "             ['default', 'optimized'] * 7]\n",
    "    i_tuples = list(zip(*index))\n",
    "    multi_index = pd.MultiIndex.from_tuples(i_tuples)\n",
    "\n",
    "    scores_df = pd.DataFrame(np.array(score_list).reshape((len(multi_index), len(multi_columns))), \n",
    "                 index=multi_index,\n",
    "                 columns=multi_columns)\n",
    "    return scores_df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "affiliated-partner",
   "metadata": {},
   "source": [
    "Initializing a new list of scores calculated after Iteration 0 is complete"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "julian-stretch",
   "metadata": {},
   "outputs": [
    {
     "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 tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration0</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration1</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration2</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration3</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">NonScaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Scaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCA</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCABoost</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">CatBoost</th>\n",
       "      <th>default</th>\n",
       "      <td>0.997689</td>\n",
       "      <td>0.931608</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.930684</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LightGBM</th>\n",
       "      <th>default</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.927911</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.926987</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">Logistic</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">KNN</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LDA</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">GNB</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">SVC</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   Iteration0           Iteration1      Iteration2       \\\n",
       "                    NonScaled               Scaled             PCA        \n",
       "                        Train      Test      Train Test      Train Test   \n",
       "CatBoost default     0.997689  0.931608        NaN  NaN        NaN  NaN   \n",
       "         optimized   1.000000  0.930684        NaN  NaN        NaN  NaN   \n",
       "LightGBM default     1.000000  0.927911        NaN  NaN        NaN  NaN   \n",
       "         optimized   1.000000  0.926987        NaN  NaN        NaN  NaN   \n",
       "Logistic default          NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "         optimized        NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "KNN      default          NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "         optimized        NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "LDA      default          NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "         optimized        NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "GNB      default          NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "         optimized        NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "SVC      default          NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "         optimized        NaN       NaN        NaN  NaN        NaN  NaN   \n",
       "\n",
       "                   Iteration3       \n",
       "                     PCABoost       \n",
       "                        Train Test  \n",
       "CatBoost default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "LightGBM default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "Logistic default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "KNN      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "LDA      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "GNB      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "SVC      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  "
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores = []\n",
    "for idx in range(len(modellist)):\n",
    "    try:\n",
    "        scores.append(locals()[modellist[idx]])\n",
    "    except:\n",
    "        scores.append(np.nan)\n",
    "scoring_table(scores)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9cd9349d",
   "metadata": {},
   "source": [
    "### Iteration 1: Baby's First Pipe\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "Our goal here will be to test the results with some initial scaling built into a pipeline. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "lined-earthquake",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'catboost__depth': 8, 'catboost__min_data_in_leaf': 1}\n"
     ]
    }
   ],
   "source": [
    "# IT1: CatBoost Classifier\n",
    "cb_1 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                       ('catboost', CatBoostClassifier(random_state=430, verbose = False))])\n",
    "cb_1.fit(Xtrain, ytrain)\n",
    "cb_1_train, cb_1_test = cb_1.score(Xtrain, ytrain), cb_1.score(Xtest, ytest)\n",
    "\n",
    "cb_grid = {'catboost__depth': [6, 8, 10],\n",
    "           'catboost__min_data_in_leaf': [1, 2, 4]}\n",
    "cb_1_opt = GridSearchCV(cb_1, cb_grid, n_jobs=-2, cv=3)\n",
    "cb_1_opt.fit(Xtrain, ytrain)\n",
    "cb_1_opt_train, cb_1_opt_test, cb_1_opt_params = cb_1_opt.score(Xtrain, ytrain), cb_1_opt.score(Xtest, ytest), cb_1_opt.best_params_\n",
    "\n",
    "print('Best Params:', cb_1_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "maritime-civilization",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Warning] min_data_in_leaf is set=19, min_child_samples=20 will be ignored. Current value: min_data_in_leaf=19\n",
      "Best Params: {'lightgbm__max_depth': 11, 'lightgbm__min_data_in_leaf': 19, 'lightgbm__num_leaves': 46}\n"
     ]
    }
   ],
   "source": [
    "# IT1: Light Gradient Boosting Machine Classifier\n",
    "lgbm_1 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                         ('lightgbm', LGBMClassifier(n_jobs=-2, random_state=430))])\n",
    "lgbm_1.fit(Xtrain, ytrain.values.ravel())\n",
    "lgbm_1_train, lgbm_1_test = lgbm_1.score(Xtrain, ytrain.values.ravel()), lgbm_1.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "lgbm_grid = {'lightgbm__num_leaves': [44, 46, 48],\n",
    "             'lightgbm__max_depth': [9, 11, 13],\n",
    "             'lightgbm__min_data_in_leaf': [15, 17, 19]}\n",
    "lgbm_1_opt = GridSearchCV(lgbm_1, lgbm_grid, n_jobs=-2, cv=3)\n",
    "lgbm_1_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "lgbm_1_opt_train, lgbm_1_opt_test, lgbm_1_opt_params = lgbm_1_opt.score(Xtrain, ytrain.values.ravel()), lgbm_1_opt.score(Xtest, ytest.values.ravel()), lgbm_1_opt.best_params_\n",
    "\n",
    "print('Best Params:', lgbm_1_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "separate-mixture",
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/theodore/anaconda3/lib/python3.8/site-packages/sklearn/model_selection/_search.py:922: UserWarning: One or more of the test scores are non-finite: [       nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan 0.93438152        nan 0.93484336\n",
      "        nan 0.9343812         nan 0.93438088        nan 0.93414972\n",
      "        nan 0.93368756        nan 0.93345639        nan 0.93299439\n",
      "        nan 0.93299439        nan 0.93322555        nan 0.93322555\n",
      "        nan 0.93322555        nan 0.93322555        nan 0.93322555\n",
      "        nan 0.93322555        nan 0.93322555        nan 0.93322555\n",
      "        nan 0.93322555        nan 0.93322555        nan 0.93322555\n",
      "        nan 0.93322555        nan 0.93345655        nan 0.93345655\n",
      "        nan 0.93345655        nan 0.93345655]\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'logistic__C': 0.24489795918367285, 'logistic__penalty': 'l2'}\n"
     ]
    }
   ],
   "source": [
    "# IT1: Logistic Regression\n",
    "lgr_1 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('logistic', LogisticRegression(n_jobs=-2))])\n",
    "lgr_1.fit(Xtrain, ytrain.values.ravel())\n",
    "lgr_1_train, lgr_1_test = lgr_1.score(Xtrain, ytrain.values.ravel()), lgr_1.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "lgr_grid = {'logistic__C': np.linspace(-4, 4, 50),\n",
    "            'logistic__penalty': ['l1', 'l2']}\n",
    "lgr_1_opt = GridSearchCV(lgr_1, lgr_grid, n_jobs=-2, cv=3)\n",
    "lgr_1_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "lgr_1_opt_train, lgr_1_opt_test, lgr_1_opt_params = lgr_1_opt.score(Xtrain, ytrain.values.ravel()), lgr_1_opt.score(Xtest, ytest.values.ravel()), lgr_1_opt.best_params_\n",
    "\n",
    "print('Best Params:', lgr_1_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "front-tulsa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'knn__metric': 'euclidean', 'knn__n_neighbors': 61, 'knn__weights': 'distance'}\n"
     ]
    }
   ],
   "source": [
    "# IT1: K-Nearest Neighbors\n",
    "knn_1 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('knn', KNeighborsClassifier(n_jobs=-2))])\n",
    "knn_1.fit(Xtrain, ytrain.values.ravel())\n",
    "knn_1_train, knn_1_test = knn_1.score(Xtrain, ytrain.values.ravel()), knn_1.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "ksqrt = int(np.round(np.sqrt(Xtrain.shape[0])))       # Get sqrt of observations for k\n",
    "knn_grid = {'knn__n_neighbors': [ksqrt - 5, ksqrt - 3, ksqrt, ksqrt + 3, ksqrt + 5],\n",
    "            'knn__weights': ['uniform', 'distance'],\n",
    "            'knn__metric': ['euclidean', 'manhattan']}\n",
    "knn_1_opt = GridSearchCV(knn_1, knn_grid, n_jobs=-2, cv=3)\n",
    "knn_1_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "knn_1_opt_train, knn_1_opt_test, knn_1_opt_params = knn_1_opt.score(Xtrain, ytrain.values.ravel()), knn_1_opt.score(Xtest, ytest.values.ravel()), knn_1_opt.best_params_\n",
    "\n",
    "print('Best Params:', knn_1_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "decent-customs",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'lda__solver': 'lsqr'}\n"
     ]
    }
   ],
   "source": [
    "# IT1: Linear Discriminant Analysis\n",
    "lda_1 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('lda', LinearDiscriminantAnalysis())])\n",
    "lda_1.fit(Xtrain, ytrain.values.ravel())\n",
    "lda_1_train, lda_1_test = lda_1.score(Xtrain, ytrain.values.ravel()), lda_1.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "lda_grid = {'lda__solver': ['svd', 'lsqr', 'eigen']}\n",
    "lda_1_opt = GridSearchCV(lda_1, lda_grid, n_jobs=-2, cv=3)\n",
    "lda_1_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "lda_1_opt_train, lda_1_opt_test, lda_1_opt_params = lda_1_opt.score(Xtrain, ytrain.values.ravel()), lda_1_opt.score(Xtest, ytest.values.ravel()), lda_1_opt.best_params_\n",
    "\n",
    "print('Best Params:', lda_1_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "beautiful-guatemala",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'gnb__var_smoothing': 0.0005336699231206307}\n"
     ]
    }
   ],
   "source": [
    "# IT1: Gaussian Naive Bayes\n",
    "gnb_1 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('gnb', GaussianNB())])\n",
    "gnb_1.fit(Xtrain, ytrain.values.ravel())\n",
    "gnb_1_train, gnb_1_test = gnb_1.score(Xtrain, ytrain.values.ravel()), gnb_1.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "gnb_grid = {'gnb__var_smoothing': np.logspace(0, -9, num = 100)}\n",
    "gnb_1_opt = GridSearchCV(gnb_1, gnb_grid, n_jobs=-2, cv=3)\n",
    "gnb_1_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "gnb_1_opt_train, gnb_1_opt_test, gnb_1_opt_params = gnb_1_opt.score(Xtrain, ytrain.values.ravel()), gnb_1_opt.score(Xtest, ytest.values.ravel()), gnb_1_opt.best_params_\n",
    "\n",
    "print('Best Params:', gnb_1_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "organized-court",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'svc__C': 0.1, 'svc__gamma': 'scale', 'svc__kernel': 'linear'}\n"
     ]
    }
   ],
   "source": [
    "# IT1: Support Vector Machine Classifier\n",
    "svc_1 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('svc', SVC())])\n",
    "svc_1.fit(Xtrain, ytrain.values.ravel())\n",
    "svc_1_train, svc_1_test = svc_1.score(Xtrain, ytrain.values.ravel()), svc_1.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "svc_grid = {'svc__kernel': ['linear', 'poly', 'rbf', 'sigmoid'],\n",
    "            'svc__gamma': ['scale'],\n",
    "            'svc__C': [0.001, 0.01, 0.1, 1, 10, 100]}\n",
    "svc_1_opt = GridSearchCV(svc_1, svc_grid, n_jobs=-2, cv=3)\n",
    "svc_1_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "svc_1_opt_train, svc_1_opt_test, svc_1_opt_params = svc_1_opt.score(Xtrain, ytrain.values.ravel()), svc_1_opt.score(Xtest, ytest.values.ravel()), svc_1_opt.best_params_\n",
    "\n",
    "print('Best Params:', svc_1_opt.best_params_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "elegant-ultimate",
   "metadata": {},
   "source": [
    "Initializing a new list of scores calculated after Iteration 1 is complete"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "temporal-antibody",
   "metadata": {},
   "outputs": [
    {
     "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 tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration0</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration1</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration2</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration3</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">NonScaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Scaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCA</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCABoost</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">CatBoost</th>\n",
       "      <th>default</th>\n",
       "      <td>0.997689</td>\n",
       "      <td>0.931608</td>\n",
       "      <td>0.997689</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.930684</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LightGBM</th>\n",
       "      <th>default</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.927911</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.931608</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.926987</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">Logistic</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.936922</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">KNN</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.944547</td>\n",
       "      <td>0.916821</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LDA</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.935074</td>\n",
       "      <td>0.920518</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.935536</td>\n",
       "      <td>0.921442</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">GNB</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.698937</td>\n",
       "      <td>0.668207</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.698937</td>\n",
       "      <td>0.668207</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">SVC</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.939464</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.937847</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   Iteration0           Iteration1           Iteration2       \\\n",
       "                    NonScaled               Scaled                  PCA        \n",
       "                        Train      Test      Train      Test      Train Test   \n",
       "CatBoost default     0.997689  0.931608   0.997689  0.928835        NaN  NaN   \n",
       "         optimized   1.000000  0.930684   1.000000  0.928835        NaN  NaN   \n",
       "LightGBM default     1.000000  0.927911   1.000000  0.931608        NaN  NaN   \n",
       "         optimized   1.000000  0.926987   1.000000  0.928835        NaN  NaN   \n",
       "Logistic default          NaN       NaN   0.937616  0.922366        NaN  NaN   \n",
       "         optimized        NaN       NaN   0.936922  0.922366        NaN  NaN   \n",
       "KNN      default          NaN       NaN   0.944547  0.916821        NaN  NaN   \n",
       "         optimized        NaN       NaN   1.000000  0.922366        NaN  NaN   \n",
       "LDA      default          NaN       NaN   0.935074  0.920518        NaN  NaN   \n",
       "         optimized        NaN       NaN   0.935536  0.921442        NaN  NaN   \n",
       "GNB      default          NaN       NaN   0.698937  0.668207        NaN  NaN   \n",
       "         optimized        NaN       NaN   0.698937  0.668207        NaN  NaN   \n",
       "SVC      default          NaN       NaN   0.939464  0.922366        NaN  NaN   \n",
       "         optimized        NaN       NaN   0.937847  0.928835        NaN  NaN   \n",
       "\n",
       "                   Iteration3       \n",
       "                     PCABoost       \n",
       "                        Train Test  \n",
       "CatBoost default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "LightGBM default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "Logistic default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "KNN      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "LDA      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "GNB      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  \n",
       "SVC      default          NaN  NaN  \n",
       "         optimized        NaN  NaN  "
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores = []\n",
    "for idx in range(len(modellist)):\n",
    "    try:\n",
    "        scores.append(locals()[modellist[idx]])\n",
    "    except:\n",
    "        scores.append(np.nan)\n",
    "scoring_table(scores)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bb3eed25",
   "metadata": {},
   "source": [
    "## Iteration 2: Give the Computer a Chance\n",
    "\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "focal-tribute",
   "metadata": {},
   "outputs": [],
   "source": [
    "# IT2: PCA Only\n",
    "from sklearn.decomposition import PCA "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "controversial-compensation",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'catboost__depth': 8, 'catboost__min_data_in_leaf': 1}\n"
     ]
    }
   ],
   "source": [
    "# IT2: CatBoost Classifier\n",
    "cb_2 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                       ('PCA', PCA(n_components=0.95, random_state=430)),\n",
    "                       ('catboost', CatBoostClassifier(random_state=430, verbose = False))])\n",
    "cb_2.fit(Xtrain, ytrain)\n",
    "cb_2_train, cb_2_test = cb_2.score(Xtrain, ytrain), cb_2.score(Xtest, ytest)\n",
    "\n",
    "cb_2_opt = GridSearchCV(cb_2, cb_grid, n_jobs=-2, cv=3)\n",
    "cb_2_opt.fit(Xtrain, ytrain)\n",
    "cb_2_opt_train, cb_2_opt_test, cb_2_opt_params = cb_2_opt.score(Xtrain, ytrain), cb_2_opt.score(Xtest, ytest), cb_2_opt.best_params_\n",
    "\n",
    "print('Best Params:', cb_2_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "signed-watershed",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Warning] min_data_in_leaf is set=19, min_child_samples=20 will be ignored. Current value: min_data_in_leaf=19\n",
      "Best Params: {'lightgbm__max_depth': 11, 'lightgbm__min_data_in_leaf': 19, 'lightgbm__num_leaves': 48}\n"
     ]
    }
   ],
   "source": [
    "# IT2: Light Gradient Boosting Machine Classifier\n",
    "lgbm_2 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                         ('PCA', PCA(n_components=0.95, random_state=430)),\n",
    "                         ('lightgbm', LGBMClassifier(n_jobs=-2, random_state=430))])\n",
    "lgbm_2.fit(Xtrain, ytrain.values.ravel())\n",
    "lgbm_2_train, lgbm_2_test = lgbm_2.score(Xtrain, ytrain.values.ravel()), lgbm_2.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "lgbm_2_opt = GridSearchCV(lgbm_2, lgbm_grid, n_jobs=-2, cv=3)\n",
    "lgbm_2_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "lgbm_2_opt_train, lgbm_2_opt_test, lgbm_2_opt_params = lgbm_2_opt.score(Xtrain, ytrain.values.ravel()), lgbm_2_opt.score(Xtest, ytest.values.ravel()), lgbm_2_opt.best_params_\n",
    "\n",
    "print('Best Params:', lgbm_2_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "delayed-version",
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/theodore/anaconda3/lib/python3.8/site-packages/sklearn/model_selection/_search.py:922: UserWarning: One or more of the test scores are non-finite: [       nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan 0.92999235        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887        nan 0.92929887\n",
      "        nan 0.92929887        nan 0.92929887]\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'logistic__C': 0.08163265306122458, 'logistic__penalty': 'l2'}\n"
     ]
    }
   ],
   "source": [
    "# IT2: Logistic Regression\n",
    "lgr_2 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('PCA', PCA(n_components=0.95, random_state=430)),\n",
    "                        ('logistic', LogisticRegression(n_jobs=-2))])\n",
    "lgr_2.fit(Xtrain, ytrain.values.ravel())\n",
    "lgr_2_train, lgr_2_test = lgr_2.score(Xtrain, ytrain.values.ravel()), lgr_2.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "lgr_2_opt = GridSearchCV(lgr_2, lgr_grid, n_jobs=-2, cv=3)\n",
    "lgr_2_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "lgr_2_opt_train, lgr_2_opt_test, lgr_2_opt_params = lgr_2_opt.score(Xtrain, ytrain.values.ravel()), lgr_2_opt.score(Xtest, ytest.values.ravel()), lgr_2_opt.best_params_\n",
    "\n",
    "print('Best Params:', lgr_2_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "reserved-hands",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'knn__metric': 'euclidean', 'knn__n_neighbors': 69, 'knn__weights': 'distance'}\n"
     ]
    }
   ],
   "source": [
    "# IT2: K-Nearest Neighbors\n",
    "knn_2 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('PCA', PCA(n_components=0.95, random_state=430)),\n",
    "                        ('knn', KNeighborsClassifier(n_jobs=-2))])\n",
    "knn_2.fit(Xtrain, ytrain.values.ravel())\n",
    "knn_2_train, knn_2_test = knn_2.score(Xtrain, ytrain.values.ravel()), knn_2.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "knn_2_opt = GridSearchCV(knn_2, knn_grid, n_jobs=-2, cv=3)\n",
    "knn_2_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "knn_2_opt_train, knn_2_opt_test, knn_2_opt_params = knn_2_opt.score(Xtrain, ytrain.values.ravel()), knn_2_opt.score(Xtest, ytest.values.ravel()), knn_2_opt.best_params_\n",
    "\n",
    "print('Best Params:', knn_2_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "cleared-sending",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'lda__solver': 'svd'}\n"
     ]
    }
   ],
   "source": [
    "# IT2: Linear Discriminant Analysis\n",
    "lda_2 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('PCA', PCA(n_components=0.95, random_state=430)),\n",
    "                        ('lda', LinearDiscriminantAnalysis())])\n",
    "lda_2.fit(Xtrain, ytrain.values.ravel())\n",
    "lda_2_train, lda_2_test = lda_2.score(Xtrain, ytrain.values.ravel()), lda_2.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "lda_2_opt = GridSearchCV(lda_2, lda_grid, n_jobs=-2, cv=3)\n",
    "lda_2_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "lda_2_opt_train, lda_2_opt_test, lda_2_opt_params = lda_2_opt.score(Xtrain, ytrain.values.ravel()), lda_2_opt.score(Xtest, ytest.values.ravel()), lda_2_opt.best_params_\n",
    "\n",
    "print('Best Params:', lda_2_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "coordinate-phone",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'gnb__var_smoothing': 0.8111308307896871}\n"
     ]
    }
   ],
   "source": [
    "# IT2: Gaussian Naive Bayes\n",
    "gnb_2 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                        ('PCA', PCA(n_components=0.95, random_state=430)),\n",
    "                        ('gnb', GaussianNB())])\n",
    "gnb_2.fit(Xtrain, ytrain.values.ravel())\n",
    "gnb_2_train, gnb_2_test = gnb_2.score(Xtrain, ytrain.values.ravel()), gnb_2.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "gnb_2_opt = GridSearchCV(gnb_2, gnb_grid, n_jobs=-2, cv=3)\n",
    "gnb_2_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "gnb_2_opt_train, gnb_2_opt_test, gnb_2_opt_params = gnb_2_opt.score(Xtrain, ytrain.values.ravel()), gnb_2_opt.score(Xtest, ytest.values.ravel()), gnb_2_opt.best_params_\n",
    "print('Best Params:', gnb_2_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "approximate-halloween",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'svc__C': 1, 'svc__gamma': 'scale', 'svc__kernel': 'rbf'}\n"
     ]
    }
   ],
   "source": [
    "# IT2: Support Vector Machine Classifier\n",
    "svc_2 = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                       ('PCA', PCA(n_components=0.95, random_state=430)),\n",
    "                       ('svc', SVC())])\n",
    "svc_2.fit(Xtrain, ytrain.values.ravel())\n",
    "svc_2_train, svc_2_test = svc_2.score(Xtrain, ytrain.values.ravel()), svc_2.score(Xtest, ytest.values.ravel())\n",
    "\n",
    "svc_2_opt = GridSearchCV(svc_2, svc_grid, n_jobs=-2, cv=3)\n",
    "svc_2_opt.fit(Xtrain, ytrain.values.ravel())\n",
    "svc_2_opt_train, svc_2_opt_test, svc_2_opt_params = svc_2_opt.score(Xtrain, ytrain.values.ravel()), svc_2_opt.score(Xtest, ytest.values.ravel()), svc_2_opt.best_params_\n",
    "print('Best Params:', svc_2_opt.best_params_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "binary-still",
   "metadata": {},
   "source": [
    "Initializing a new list of scores calculated after Iteration 2 is complete"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "brutal-updating",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "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 tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration0</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration1</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration2</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration3</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">NonScaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Scaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCA</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCABoost</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">CatBoost</th>\n",
       "      <th>default</th>\n",
       "      <td>0.997689</td>\n",
       "      <td>0.931608</td>\n",
       "      <td>0.997689</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>0.998152</td>\n",
       "      <td>0.917745</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.930684</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LightGBM</th>\n",
       "      <th>default</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.927911</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.931608</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.919593</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.926987</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.916821</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">Logistic</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>0.933457</td>\n",
       "      <td>0.917745</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.936922</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>0.932763</td>\n",
       "      <td>0.916821</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">KNN</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.944547</td>\n",
       "      <td>0.916821</td>\n",
       "      <td>0.945702</td>\n",
       "      <td>0.903882</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.923290</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LDA</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.935074</td>\n",
       "      <td>0.920518</td>\n",
       "      <td>0.933688</td>\n",
       "      <td>0.914048</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.935536</td>\n",
       "      <td>0.921442</td>\n",
       "      <td>0.933688</td>\n",
       "      <td>0.914048</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">GNB</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.698937</td>\n",
       "      <td>0.668207</td>\n",
       "      <td>0.554529</td>\n",
       "      <td>0.538817</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.698937</td>\n",
       "      <td>0.668207</td>\n",
       "      <td>0.926987</td>\n",
       "      <td>0.913124</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">SVC</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.939464</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.920518</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.937847</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.920518</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   Iteration0           Iteration1           Iteration2  \\\n",
       "                    NonScaled               Scaled                  PCA   \n",
       "                        Train      Test      Train      Test      Train   \n",
       "CatBoost default     0.997689  0.931608   0.997689  0.928835   0.998152   \n",
       "         optimized   1.000000  0.930684   1.000000  0.928835   1.000000   \n",
       "LightGBM default     1.000000  0.927911   1.000000  0.931608   1.000000   \n",
       "         optimized   1.000000  0.926987   1.000000  0.928835   1.000000   \n",
       "Logistic default          NaN       NaN   0.937616  0.922366   0.933457   \n",
       "         optimized        NaN       NaN   0.936922  0.922366   0.932763   \n",
       "KNN      default          NaN       NaN   0.944547  0.916821   0.945702   \n",
       "         optimized        NaN       NaN   1.000000  0.922366   1.000000   \n",
       "LDA      default          NaN       NaN   0.935074  0.920518   0.933688   \n",
       "         optimized        NaN       NaN   0.935536  0.921442   0.933688   \n",
       "GNB      default          NaN       NaN   0.698937  0.668207   0.554529   \n",
       "         optimized        NaN       NaN   0.698937  0.668207   0.926987   \n",
       "SVC      default          NaN       NaN   0.939464  0.922366   0.937616   \n",
       "         optimized        NaN       NaN   0.937847  0.928835   0.937616   \n",
       "\n",
       "                             Iteration3       \n",
       "                               PCABoost       \n",
       "                        Test      Train Test  \n",
       "CatBoost default    0.917745        NaN  NaN  \n",
       "         optimized  0.922366        NaN  NaN  \n",
       "LightGBM default    0.919593        NaN  NaN  \n",
       "         optimized  0.916821        NaN  NaN  \n",
       "Logistic default    0.917745        NaN  NaN  \n",
       "         optimized  0.916821        NaN  NaN  \n",
       "KNN      default    0.903882        NaN  NaN  \n",
       "         optimized  0.923290        NaN  NaN  \n",
       "LDA      default    0.914048        NaN  NaN  \n",
       "         optimized  0.914048        NaN  NaN  \n",
       "GNB      default    0.538817        NaN  NaN  \n",
       "         optimized  0.913124        NaN  NaN  \n",
       "SVC      default    0.920518        NaN  NaN  \n",
       "         optimized  0.920518        NaN  NaN  "
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores = []\n",
    "for idx in range(len(modellist)):\n",
    "    try:\n",
    "        scores.append(locals()[modellist[idx]])\n",
    "    except:\n",
    "        scores.append(np.nan)\n",
    "scoring_table(scores)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "rising-edward",
   "metadata": {},
   "source": [
    "## Iteration 3: PCABoosted Modeling\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "Considering that the bespoke function would not play well with Pipeline, we have pre-processed the data here, first with scaling and then with the PCA Boost. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "functioning-hepatitis",
   "metadata": {},
   "outputs": [],
   "source": [
    "def PCAmerge(train, test, n_comp=0.95, r_state = 42):\n",
    "    from sklearn.decomposition import PCA \n",
    "    pca = PCA(n_components = n_comp, random_state = r_state)\n",
    "    train_pca = pd.DataFrame(pca.fit_transform(train), index=train.index)\n",
    "    test_pca = pd.DataFrame(pca.transform(test), index=test.index)\n",
    "    PCA_col_list = []\n",
    "    for num in range(1, train_pca.shape[1]+1):\n",
    "        PCA_col_list.append(('PCA' + str(num)))\n",
    "    train_pca.columns = PCA_col_list\n",
    "    train = pd.concat([train, train_pca], axis=1)\n",
    "    test_pca.columns = PCA_col_list\n",
    "    test = pd.concat([test, test_pca], axis=1)\n",
    "    return train, test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "atmospheric-abraham",
   "metadata": {},
   "outputs": [],
   "source": [
    "# IT3: PCA Boosted Models\n",
    "scaled = StandardScaler()\n",
    "Xtrain_Scaled = pd.DataFrame(scaled.fit_transform(Xtrain), columns = Xtrain.columns, index = Xtrain.index)\n",
    "Xtest_Scaled = pd.DataFrame(scaled.fit_transform(Xtest), columns = Xtest.columns, index = Xtest.index)\n",
    "XtrainPCA, XtestPCA = PCAmerge(Xtrain_Scaled, Xtest_Scaled, n_comp=0.95)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "egyptian-surface",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'depth': 8, 'min_data_in_leaf': 1}\n"
     ]
    }
   ],
   "source": [
    "# IT3: CatBoost Classifier\n",
    "cb_3 = CatBoostClassifier(random_state=430, verbose = False)\n",
    "cb_3.fit(XtrainPCA, ytrain)\n",
    "cb_3_train, cb_3_test = cb_3.score(XtrainPCA, ytrain), cb_3.score(XtestPCA, ytest)\n",
    "\n",
    "cb_grid = {'depth': [6, 8, 10],\n",
    "          'min_data_in_leaf': [1, 2, 4]}\n",
    "cb_3_opt = GridSearchCV(cb_3, cb_grid, n_jobs=-2, cv=3)\n",
    "cb_3_opt.fit(XtrainPCA, ytrain)\n",
    "cb_3_opt_train, cb_3_opt_test, cb_3_opt_params = cb_3_opt.score(XtrainPCA, ytrain), cb_3_opt.score(XtestPCA, ytest), cb_3_opt.best_params_\n",
    "\n",
    "print('Best Params:', cb_3_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "authorized-quarter",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/theodore/anaconda3/lib/python3.8/site-packages/joblib/externals/loky/process_executor.py:688: UserWarning: A worker stopped while some jobs were given to the executor. This can be caused by a too short worker timeout or by a memory leak.\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[LightGBM] [Warning] min_data_in_leaf is set=15, min_child_samples=20 will be ignored. Current value: min_data_in_leaf=15\n",
      "Best Params: {'max_depth': 13, 'min_data_in_leaf': 15, 'num_leaves': 48}\n"
     ]
    }
   ],
   "source": [
    "# IT3: Light Gradient Boosting Machine Classifier\n",
    "lgbm_3 = LGBMClassifier(n_jobs=-2, random_state=430)\n",
    "lgbm_3.fit(XtrainPCA, ytrain.values.ravel())\n",
    "lgbm_3_train, lgbm_3_test = lgbm_3.score(XtrainPCA, ytrain.values.ravel()), lgbm_3.score(XtestPCA, ytest.values.ravel())\n",
    "\n",
    "lgbm_grid = {'num_leaves': [44, 46, 48],\n",
    "             'max_depth': [9, 11, 13],\n",
    "             'min_data_in_leaf': [15, 17, 19]}\n",
    "lgbm_3_opt = GridSearchCV(lgbm_3, lgbm_grid, n_jobs=-2, cv=3)\n",
    "lgbm_3_opt.fit(XtrainPCA, ytrain.values.ravel())\n",
    "lgbm_3_opt_train, lgbm_3_opt_test, lgbm_3_opt_params = lgbm_3_opt.score(XtrainPCA, ytrain.values.ravel()), lgbm_3_opt.score(XtestPCA, ytest.values.ravel()), lgbm_3_opt.best_params_\n",
    "\n",
    "print('Best Params:', lgbm_3_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "solved-decline",
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/theodore/anaconda3/lib/python3.8/site-packages/sklearn/model_selection/_search.py:922: UserWarning: One or more of the test scores are non-finite: [       nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan 0.93438168        nan 0.93461236\n",
      "        nan 0.93391888        nan 0.93438088        nan 0.93391856\n",
      "        nan 0.93368756        nan 0.93345639        nan 0.93322539\n",
      "        nan 0.93322555        nan 0.93322555        nan 0.93322555\n",
      "        nan 0.93345655        nan 0.93322555        nan 0.93322555\n",
      "        nan 0.93322555        nan 0.93368756        nan 0.93322555\n",
      "        nan 0.93322555        nan 0.93345655        nan 0.93345655\n",
      "        nan 0.93345655        nan 0.93368756        nan 0.93345655\n",
      "        nan 0.93345655        nan 0.93322539]\n",
      "  warnings.warn(\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'C': 0.24489795918367285, 'penalty': 'l2'}\n"
     ]
    }
   ],
   "source": [
    "# IT3: Logistic Regression\n",
    "lgr_3 = LogisticRegression(n_jobs=-2)\n",
    "lgr_3.fit(XtrainPCA, ytrain.values.ravel())\n",
    "lgr_3_train, lgr_3_test = lgr_3.score(XtrainPCA, ytrain.values.ravel()), lgr_3.score(XtestPCA, ytest.values.ravel())\n",
    "\n",
    "lgr_grid = {'C': np.linspace(-4, 4, 50),\n",
    "            'penalty': ['l1', 'l2']}\n",
    "lgr_3_opt = GridSearchCV(lgr_3, lgr_grid, n_jobs=-2, cv=3)\n",
    "lgr_3_opt.fit(XtrainPCA, ytrain.values.ravel())\n",
    "lgr_3_opt_train, lgr_3_opt_test, lgr_3_opt_params = lgr_3_opt.score(XtrainPCA, ytrain.values.ravel()), lgr_3_opt.score(XtestPCA, ytest.values.ravel()), lgr_3_opt.best_params_\n",
    "\n",
    "print('Best Params:', lgr_3_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "heard-grocery",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'metric': 'euclidean', 'n_neighbors': 63, 'weights': 'distance'}\n"
     ]
    }
   ],
   "source": [
    "# IT3: K-Nearest Neighbors\n",
    "knn_3 = KNeighborsClassifier(n_jobs=-2)\n",
    "knn_3.fit(XtrainPCA, ytrain.values.ravel())\n",
    "knn_3_train, knn_3_test = knn_3.score(XtrainPCA, ytrain.values.ravel()), knn_3.score(XtestPCA, ytest.values.ravel())\n",
    "\n",
    "knn_grid = {'n_neighbors': [61, 63, 66, 69, 71],\n",
    "            'weights': ['uniform', 'distance'],\n",
    "            'metric': ['euclidean', 'manhattan']}\n",
    "knn_3_opt = GridSearchCV(knn_3, knn_grid, n_jobs=-2, cv=3)\n",
    "knn_3_opt.fit(XtrainPCA, ytrain.values.ravel())\n",
    "knn_3_opt_train, knn_3_opt_test, knn_3_opt_params = knn_3_opt.score(XtrainPCA, ytrain.values.ravel()), knn_3_opt.score(XtestPCA, ytest.values.ravel()), knn_3_opt.best_params_\n",
    "\n",
    "print('Best Params:', knn_3_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "sorted-hours",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'solver': 'lsqr'}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/theodore/anaconda3/lib/python3.8/site-packages/sklearn/model_selection/_search.py:922: UserWarning: One or more of the test scores are non-finite: [0.93438168 0.93461268        nan]\n",
      "  warnings.warn(\n"
     ]
    }
   ],
   "source": [
    "# IT3: Linear Discriminant Analysis\n",
    "lda_3 = LinearDiscriminantAnalysis()\n",
    "lda_3.fit(XtrainPCA, ytrain.values.ravel())\n",
    "lda_3_train, lda_3_test = lda_3.score(XtrainPCA, ytrain.values.ravel()), lda_3.score(XtestPCA, ytest.values.ravel())\n",
    "\n",
    "lda_grid = {'solver': ['svd', 'lsqr', 'eigen']}\n",
    "lda_3_opt = GridSearchCV(lda_3, lda_grid, n_jobs=-2, cv=3)\n",
    "lda_3_opt.fit(XtrainPCA, ytrain.values.ravel())\n",
    "lda_3_opt_train, lda_3_opt_test, lda_3_opt_params = lda_3_opt.score(XtrainPCA, ytrain.values.ravel()), lda_3_opt.score(XtestPCA, ytest.values.ravel()), lda_3_opt.best_params_\n",
    "\n",
    "print('Best Params:', lda_3_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "unauthorized-ratio",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'var_smoothing': 1.0}\n"
     ]
    }
   ],
   "source": [
    "# IT3: Gaussian Naive Bayes\n",
    "gnb_3 = GaussianNB()\n",
    "gnb_3.fit(XtrainPCA, ytrain.values.ravel())\n",
    "gnb_3_train, gnb_3_test = gnb_3.score(XtrainPCA, ytrain.values.ravel()), gnb_3.score(XtestPCA, ytest.values.ravel())\n",
    "\n",
    "gnb_grid = {'var_smoothing': np.logspace(0, -9, num = 100)}\n",
    "gnb_3_opt = GridSearchCV(gnb_3, gnb_grid, n_jobs=-2, cv=3)\n",
    "gnb_3_opt.fit(XtrainPCA, ytrain.values.ravel())\n",
    "gnb_3_opt_train, gnb_3_opt_test, gnb_3_opt_params = gnb_3_opt.score(XtrainPCA, ytrain.values.ravel()), gnb_3_opt.score(XtestPCA, ytest.values.ravel()), gnb_3_opt.best_params_\n",
    "print('Best Params:', gnb_3_opt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "distributed-newark",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Params: {'C': 0.1, 'gamma': 'scale', 'kernel': 'linear'}\n"
     ]
    }
   ],
   "source": [
    "# IT3: Support Vector Machine Classifier\n",
    "svc_3 = SVC()\n",
    "svc_3.fit(XtrainPCA, ytrain.values.ravel())\n",
    "svc_3_train, svc_3_test = svc_3.score(XtrainPCA, ytrain.values.ravel()), svc_3.score(XtestPCA, ytest.values.ravel())\n",
    "\n",
    "svc_grid = {'kernel': ['linear', 'poly', 'rbf', 'sigmoid'],\n",
    "            'gamma': ['scale'],\n",
    "            'C': [0.001, 0.01, 0.1, 1, 10, 100]}\n",
    "svc_3_opt = GridSearchCV(svc_3, svc_grid, n_jobs=-2, cv=3)\n",
    "svc_3_opt.fit(XtrainPCA, ytrain.values.ravel())\n",
    "svc_3_opt_train, svc_3_opt_test, svc_3_opt_params = svc_3_opt.score(XtrainPCA, ytrain.values.ravel()), svc_3_opt.score(XtestPCA, ytest.values.ravel()), svc_3_opt.best_params_\n",
    "print('Best Params:', svc_3_opt.best_params_)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "finished-edward",
   "metadata": {},
   "source": [
    "Initializing a final list of scores calculated after Iteration 3 is complete"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "marked-warner",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "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 tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration0</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration1</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration2</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Iteration3</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th colspan=\"2\" halign=\"left\">NonScaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">Scaled</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCA</th>\n",
       "      <th colspan=\"2\" halign=\"left\">PCABoost</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "      <th>Train</th>\n",
       "      <th>Test</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">CatBoost</th>\n",
       "      <th>default</th>\n",
       "      <td>0.997689</td>\n",
       "      <td>0.931608</td>\n",
       "      <td>0.997689</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>0.998152</td>\n",
       "      <td>0.917745</td>\n",
       "      <td>0.998614</td>\n",
       "      <td>0.920518</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.930684</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.926987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LightGBM</th>\n",
       "      <th>default</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.927911</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.931608</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.919593</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.922366</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.926987</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.916821</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.922366</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">Logistic</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>0.933457</td>\n",
       "      <td>0.917745</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.924214</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.936922</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>0.932763</td>\n",
       "      <td>0.916821</td>\n",
       "      <td>0.937153</td>\n",
       "      <td>0.926987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">KNN</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.944547</td>\n",
       "      <td>0.916821</td>\n",
       "      <td>0.945702</td>\n",
       "      <td>0.903882</td>\n",
       "      <td>0.945009</td>\n",
       "      <td>0.909427</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.923290</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.919593</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">LDA</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.935074</td>\n",
       "      <td>0.920518</td>\n",
       "      <td>0.933688</td>\n",
       "      <td>0.914048</td>\n",
       "      <td>0.935074</td>\n",
       "      <td>0.926987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.935536</td>\n",
       "      <td>0.921442</td>\n",
       "      <td>0.933688</td>\n",
       "      <td>0.914048</td>\n",
       "      <td>0.935536</td>\n",
       "      <td>0.354898</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">GNB</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.698937</td>\n",
       "      <td>0.668207</td>\n",
       "      <td>0.554529</td>\n",
       "      <td>0.538817</td>\n",
       "      <td>0.631238</td>\n",
       "      <td>0.637708</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.698937</td>\n",
       "      <td>0.668207</td>\n",
       "      <td>0.926987</td>\n",
       "      <td>0.913124</td>\n",
       "      <td>0.880776</td>\n",
       "      <td>0.875231</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">SVC</th>\n",
       "      <th>default</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.939464</td>\n",
       "      <td>0.922366</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.920518</td>\n",
       "      <td>0.938771</td>\n",
       "      <td>0.923290</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>optimized</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.937847</td>\n",
       "      <td>0.928835</td>\n",
       "      <td>0.937616</td>\n",
       "      <td>0.920518</td>\n",
       "      <td>0.937847</td>\n",
       "      <td>0.926063</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   Iteration0           Iteration1           Iteration2  \\\n",
       "                    NonScaled               Scaled                  PCA   \n",
       "                        Train      Test      Train      Test      Train   \n",
       "CatBoost default     0.997689  0.931608   0.997689  0.928835   0.998152   \n",
       "         optimized   1.000000  0.930684   1.000000  0.928835   1.000000   \n",
       "LightGBM default     1.000000  0.927911   1.000000  0.931608   1.000000   \n",
       "         optimized   1.000000  0.926987   1.000000  0.928835   1.000000   \n",
       "Logistic default          NaN       NaN   0.937616  0.922366   0.933457   \n",
       "         optimized        NaN       NaN   0.936922  0.922366   0.932763   \n",
       "KNN      default          NaN       NaN   0.944547  0.916821   0.945702   \n",
       "         optimized        NaN       NaN   1.000000  0.922366   1.000000   \n",
       "LDA      default          NaN       NaN   0.935074  0.920518   0.933688   \n",
       "         optimized        NaN       NaN   0.935536  0.921442   0.933688   \n",
       "GNB      default          NaN       NaN   0.698937  0.668207   0.554529   \n",
       "         optimized        NaN       NaN   0.698937  0.668207   0.926987   \n",
       "SVC      default          NaN       NaN   0.939464  0.922366   0.937616   \n",
       "         optimized        NaN       NaN   0.937847  0.928835   0.937616   \n",
       "\n",
       "                             Iteration3            \n",
       "                               PCABoost            \n",
       "                        Test      Train      Test  \n",
       "CatBoost default    0.917745   0.998614  0.920518  \n",
       "         optimized  0.922366   1.000000  0.926987  \n",
       "LightGBM default    0.919593   1.000000  0.922366  \n",
       "         optimized  0.916821   1.000000  0.922366  \n",
       "Logistic default    0.917745   0.937616  0.924214  \n",
       "         optimized  0.916821   0.937153  0.926987  \n",
       "KNN      default    0.903882   0.945009  0.909427  \n",
       "         optimized  0.923290   1.000000  0.919593  \n",
       "LDA      default    0.914048   0.935074  0.926987  \n",
       "         optimized  0.914048   0.935536  0.354898  \n",
       "GNB      default    0.538817   0.631238  0.637708  \n",
       "         optimized  0.913124   0.880776  0.875231  \n",
       "SVC      default    0.920518   0.938771  0.923290  \n",
       "         optimized  0.920518   0.937847  0.926063  "
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores = []\n",
    "for idx in range(len(modellist)):\n",
    "    try:\n",
    "        scores.append(locals()[modellist[idx]])\n",
    "    except:\n",
    "        scores.append(np.nan)\n",
    "scoring_table(scores)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "selective-switch",
   "metadata": {},
   "source": [
    "## Best Models\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "\n",
    "Creating a table of the best hypertransformations for each ML model, comparing test scores among all iterations.    \n",
    "These will guide our treatment of the full Xtrain data in the next section."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "mature-excitement",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "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>Iteration</th>\n",
       "      <th>Transformation</th>\n",
       "      <th>Optimized</th>\n",
       "      <th>Test Score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>CatBoost</th>\n",
       "      <td>Iteration0</td>\n",
       "      <td>NonScaled</td>\n",
       "      <td>default</td>\n",
       "      <td>0.931608</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LightGBM</th>\n",
       "      <td>Iteration1</td>\n",
       "      <td>Scaled</td>\n",
       "      <td>default</td>\n",
       "      <td>0.931608</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Logistic</th>\n",
       "      <td>Iteration3</td>\n",
       "      <td>PCABoost</td>\n",
       "      <td>optimized</td>\n",
       "      <td>0.926987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>KNN</th>\n",
       "      <td>Iteration2</td>\n",
       "      <td>PCA</td>\n",
       "      <td>optimized</td>\n",
       "      <td>0.923290</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LDA</th>\n",
       "      <td>Iteration3</td>\n",
       "      <td>PCABoost</td>\n",
       "      <td>default</td>\n",
       "      <td>0.926987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GNB</th>\n",
       "      <td>Iteration2</td>\n",
       "      <td>PCA</td>\n",
       "      <td>optimized</td>\n",
       "      <td>0.913124</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SVC</th>\n",
       "      <td>Iteration1</td>\n",
       "      <td>Scaled</td>\n",
       "      <td>optimized</td>\n",
       "      <td>0.928835</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           Iteration Transformation  Optimized  Test Score\n",
       "CatBoost  Iteration0      NonScaled    default    0.931608\n",
       "LightGBM  Iteration1         Scaled    default    0.931608\n",
       "Logistic  Iteration3       PCABoost  optimized    0.926987\n",
       "KNN       Iteration2            PCA  optimized    0.923290\n",
       "LDA       Iteration3       PCABoost    default    0.926987\n",
       "GNB       Iteration2            PCA  optimized    0.913124\n",
       "SVC       Iteration1         Scaled  optimized    0.928835"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "models = ['CatBoost', 'LightGBM', 'Logistic', 'KNN', 'LDA', 'GNB', 'SVC']\n",
    "model_scores = scoring_table(scores)\n",
    "test_scores = model_scores.loc[:, model_scores.columns.get_level_values(2) == 'Test'].unstack(1).T\n",
    "best_models = pd.DataFrame([], columns = ['Iteration', 'Transformation', 'Test_Train', 'Optimized', 'Test Score'])\n",
    "for model in models:\n",
    "    best_models.loc[model] = test_scores[model].sort_values(ascending=False).reset_index().iloc[0].values\n",
    "best_models = best_models.drop('Test_Train', axis=1)\n",
    "best_models"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "separated-absolute",
   "metadata": {},
   "source": [
    "Getting rid of '_0' entries by generating a reduced list of model names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "aquatic-devil",
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['cb_0_train',\n",
       " 'cb_0_test',\n",
       " 'cb_1_train',\n",
       " 'cb_1_test',\n",
       " 'cb_2_train',\n",
       " 'cb_2_test',\n",
       " 'cb_3_train',\n",
       " 'cb_3_test',\n",
       " 'cb_0_opt_train',\n",
       " 'cb_0_opt_test',\n",
       " 'cb_1_opt_train',\n",
       " 'cb_1_opt_test',\n",
       " 'cb_2_opt_train',\n",
       " 'cb_2_opt_test',\n",
       " 'cb_3_opt_train',\n",
       " 'cb_3_opt_test',\n",
       " 'lgbm_0_train',\n",
       " 'lgbm_0_test',\n",
       " 'lgbm_1_train',\n",
       " 'lgbm_1_test',\n",
       " 'lgbm_2_train',\n",
       " 'lgbm_2_test',\n",
       " 'lgbm_3_train',\n",
       " 'lgbm_3_test',\n",
       " 'lgbm_0_opt_train',\n",
       " 'lgbm_0_opt_test',\n",
       " 'lgbm_1_opt_train',\n",
       " 'lgbm_1_opt_test',\n",
       " 'lgbm_2_opt_train',\n",
       " 'lgbm_2_opt_test',\n",
       " 'lgbm_3_opt_train',\n",
       " 'lgbm_3_opt_test',\n",
       " 'lgr_1_train',\n",
       " 'lgr_1_test',\n",
       " 'lgr_2_train',\n",
       " 'lgr_2_test',\n",
       " 'lgr_3_train',\n",
       " 'lgr_3_test',\n",
       " 'lgr_1_opt_train',\n",
       " 'lgr_1_opt_test',\n",
       " 'lgr_2_opt_train',\n",
       " 'lgr_2_opt_test',\n",
       " 'lgr_3_opt_train',\n",
       " 'lgr_3_opt_test',\n",
       " 'knn_1_train',\n",
       " 'knn_1_test',\n",
       " 'knn_2_train',\n",
       " 'knn_2_test',\n",
       " 'knn_3_train',\n",
       " 'knn_3_test',\n",
       " 'knn_1_opt_train',\n",
       " 'knn_1_opt_test',\n",
       " 'knn_2_opt_train',\n",
       " 'knn_2_opt_test',\n",
       " 'knn_3_opt_train',\n",
       " 'knn_3_opt_test',\n",
       " 'lda_1_train',\n",
       " 'lda_1_test',\n",
       " 'lda_2_train',\n",
       " 'lda_2_test',\n",
       " 'lda_3_train',\n",
       " 'lda_3_test',\n",
       " 'lda_1_opt_train',\n",
       " 'lda_1_opt_test',\n",
       " 'lda_2_opt_train',\n",
       " 'lda_2_opt_test',\n",
       " 'lda_3_opt_train',\n",
       " 'lda_3_opt_test',\n",
       " 'gnb_1_train',\n",
       " 'gnb_1_test',\n",
       " 'gnb_2_train',\n",
       " 'gnb_2_test',\n",
       " 'gnb_3_train',\n",
       " 'gnb_3_test',\n",
       " 'gnb_1_opt_train',\n",
       " 'gnb_1_opt_test',\n",
       " 'gnb_2_opt_train',\n",
       " 'gnb_2_opt_test',\n",
       " 'gnb_3_opt_train',\n",
       " 'gnb_3_opt_test',\n",
       " 'svc_1_train',\n",
       " 'svc_1_test',\n",
       " 'svc_2_train',\n",
       " 'svc_2_test',\n",
       " 'svc_3_train',\n",
       " 'svc_3_test',\n",
       " 'svc_1_opt_train',\n",
       " 'svc_1_opt_test',\n",
       " 'svc_2_opt_train',\n",
       " 'svc_2_opt_test',\n",
       " 'svc_3_opt_train',\n",
       " 'svc_3_opt_test']"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "modellistreduced = modellist.copy()\n",
    "\n",
    "num_list = [105, 104]\n",
    "i = 0\n",
    "while i < 9:\n",
    "    max_idx = len(num_list) - 1\n",
    "    num_list.append(num_list[max_idx - 1] - 8)\n",
    "    num_list.append(num_list[max_idx] - 8)\n",
    "    i += 1\n",
    "\n",
    "for num in num_list:\n",
    "    modellistreduced.pop(num)\n",
    "modellistreduced"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "labeled-yorkshire",
   "metadata": {},
   "source": [
    "Getting a list of model names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "inside-fighter",
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['cb_0',\n",
       " 'cb_1',\n",
       " 'cb_2',\n",
       " 'cb_3',\n",
       " 'cb_0_opt',\n",
       " 'cb_1_opt',\n",
       " 'cb_2_opt',\n",
       " 'cb_3_opt',\n",
       " 'lgbm_0',\n",
       " 'lgbm_1',\n",
       " 'lgbm_2',\n",
       " 'lgbm_3',\n",
       " 'lgbm_0_opt',\n",
       " 'lgbm_1_opt',\n",
       " 'lgbm_2_opt',\n",
       " 'lgbm_3_opt',\n",
       " 'lgr_1',\n",
       " 'lgr_2',\n",
       " 'lgr_3',\n",
       " 'lgr_1_opt',\n",
       " 'lgr_2_opt',\n",
       " 'lgr_3_opt',\n",
       " 'knn_1',\n",
       " 'knn_2',\n",
       " 'knn_3',\n",
       " 'knn_1_opt',\n",
       " 'knn_2_opt',\n",
       " 'knn_3_opt',\n",
       " 'lda_1',\n",
       " 'lda_2',\n",
       " 'lda_3',\n",
       " 'lda_1_opt',\n",
       " 'lda_2_opt',\n",
       " 'lda_3_opt',\n",
       " 'gnb_1',\n",
       " 'gnb_2',\n",
       " 'gnb_3',\n",
       " 'gnb_1_opt',\n",
       " 'gnb_2_opt',\n",
       " 'gnb_3_opt',\n",
       " 'svc_1',\n",
       " 'svc_2',\n",
       " 'svc_3',\n",
       " 'svc_1_opt',\n",
       " 'svc_2_opt',\n",
       " 'svc_3_opt']"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "modelinputlist = modellistreduced.copy()\n",
    "for idx in range(len(modelinputlist)):\n",
    "    if idx % 2 == 0:\n",
    "        modelinputlist[idx] = modelinputlist[idx][:-6]\n",
    "    else:\n",
    "        modelinputlist[idx] = modelinputlist[idx][:-5]\n",
    "modelinputlist = modelinputlist[::2]\n",
    "modelinputlist"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "double-weekly",
   "metadata": {},
   "source": [
    "Getting a list of saved optimized parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "directed-quality",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "dictlist = [ele for ele in ['cb', 'lgbm', 'lgr', 'knn', 'lda', 'gnb', 'svc'] for i in range(4)]\n",
    "iterdictlist = ['_0', '_1', '_2', '_3'] * 7\n",
    "for idx in range(len(dictlist)):\n",
    "    dictlist[idx] = dictlist[idx] + iterdictlist[idx] + '_opt_params'\n",
    "for num in [24, 20, 16, 12, 8]:\n",
    "    dictlist.pop(num)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "saved-strengthening",
   "metadata": {},
   "source": [
    "Combining these three lists to load all saved model elements into a single dictionary for pickling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "impossible-aerospace",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_dict = {}\n",
    "all_list = modelinputlist + modellistreduced + dictlist\n",
    "for ele in all_list:\n",
    "    model_dict[ele] = locals()[ele]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "answering-opera",
   "metadata": {},
   "outputs": [],
   "source": [
    "pickle.dump(model_scores, open('model_scores.pickle', 'wb'))\n",
    "pickle.dump(best_models, open('best_models.pickle', 'wb'))\n",
    "pickle.dump(model_dict, open('model_dict.pickle', 'wb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4f7d155d",
   "metadata": {},
   "source": [
    "# Preliminary Results & Analysis\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "To recap, thus far we have combined the techniques of hyper transformation tuning and hyper parameter tuning to find the very best versions of each of our machine learning models. Each of these performs well in the test set, but how do we apply what we have learned?\n",
    "\n",
    "The classical approach to this point would be to investigate the relative confusion matrices of the models and to compare them against our priorities. To contextualize briefly, our best application of these data for an insurance company would be in the savings of money; our goal is to ascertain actual instances of fraud, **True Positives**, and the least number of faulty detections, or **False Positives**. In this way we can best direct the use of funds for the client. This approach maximizes the *precision* of our detection."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "d26454ad",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CatBoost \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0       0.95      0.97      0.96       971\n",
      "           1       0.71      0.56      0.63       111\n",
      "\n",
      "    accuracy                           0.93      1082\n",
      "   macro avg       0.83      0.77      0.79      1082\n",
      "weighted avg       0.93      0.93      0.93      1082\n",
      "\n",
      "------------------------------\n",
      "LightGBM \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0       0.95      0.98      0.96       971\n",
      "           1       0.72      0.55      0.62       111\n",
      "\n",
      "    accuracy                           0.93      1082\n",
      "   macro avg       0.83      0.76      0.79      1082\n",
      "weighted avg       0.93      0.93      0.93      1082\n",
      "\n",
      "------------------------------\n",
      "Logistic Reg. \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0       0.94      0.98      0.96       971\n",
      "           1       0.74      0.45      0.56       111\n",
      "\n",
      "    accuracy                           0.93      1082\n",
      "   macro avg       0.84      0.72      0.76      1082\n",
      "weighted avg       0.92      0.93      0.92      1082\n",
      "\n",
      "------------------------------\n",
      "KNN \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0       0.93      0.98      0.96       971\n",
      "           1       0.74      0.39      0.51       111\n",
      "\n",
      "    accuracy                           0.92      1082\n",
      "   macro avg       0.84      0.69      0.73      1082\n",
      "weighted avg       0.91      0.92      0.91      1082\n",
      "\n",
      "------------------------------\n",
      "LDA \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0       0.95      0.97      0.96       971\n",
      "           1       0.68      0.54      0.60       111\n",
      "\n",
      "    accuracy                           0.93      1082\n",
      "   macro avg       0.82      0.76      0.78      1082\n",
      "weighted avg       0.92      0.93      0.92      1082\n",
      "\n",
      "------------------------------\n",
      "GNB \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0       0.92      0.99      0.95       971\n",
      "           1       0.81      0.20      0.32       111\n",
      "\n",
      "    accuracy                           0.91      1082\n",
      "   macro avg       0.87      0.60      0.64      1082\n",
      "weighted avg       0.91      0.91      0.89      1082\n",
      "\n",
      "------------------------------\n",
      "SVC \n",
      "               precision    recall  f1-score   support\n",
      "\n",
      "           0       0.94      0.99      0.96       971\n",
      "           1       0.78      0.42      0.55       111\n",
      "\n",
      "    accuracy                           0.93      1082\n",
      "   macro avg       0.86      0.71      0.76      1082\n",
      "weighted avg       0.92      0.93      0.92      1082\n",
      "\n",
      "------------------------------\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import plot_confusion_matrix, classification_report\n",
    "\n",
    "# Prep Graphing Procedure\n",
    "model_dict = pickle.load(open('model_dict.pickle', 'rb'))\n",
    "model_names = ['CatBoost', 'LightGBM', 'Logistic Reg.', \n",
    "               'KNN', 'LDA', 'GNB', 'SVC']\n",
    "top_dict = {'CatBoost': model_dict['cb_0'],\n",
    "           'LightGBM': model_dict['lgbm_1'],\n",
    "           'Logistic Reg.': model_dict['lgr_3_opt'],\n",
    "           'KNN': model_dict['knn_2_opt'],\n",
    "           'LDA': model_dict['lda_3'],\n",
    "           'GNB': model_dict['gnb_2_opt'],\n",
    "           'SVC': model_dict['svc_1_opt']}\n",
    "\n",
    "# Graph Matrices\n",
    "for model_name in model_names:\n",
    "    try:\n",
    "        plot_confusion_matrix(top_dict[model_name], Xtest, ytest)\n",
    "        plt.title(model_name)\n",
    "        print(model_name, '\\n',classification_report(ytest, top_dict[model_name].predict(Xtest)))\n",
    "        print(30*'-')\n",
    "    except:  # For models requiring XtestPCA\n",
    "        plot_confusion_matrix(top_dict[model_name], XtestPCA, ytest)\n",
    "        plt.title(model_name)\n",
    "        print(model_name, '\\n', classification_report(ytest, top_dict[model_name].predict(XtestPCA)))\n",
    "        print(30*'-')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b03e07dd",
   "metadata": {},
   "source": [
    "## Surface Level Discussion\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "In terms of our project goals, **CatBoost** features the optimal precision, with the fewest predicted False Positives and the most True Positives.\n",
    "\n",
    "-something something savings\n",
    "\n",
    "## Permutation Importance\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "From here, we can potentially mine further information by applying Permutation Importance to any of our given models. For the sake of example, we will demonstrate the procedure with CatBoost, considering it made it to the top as the most apparently useful model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "dd598211",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 1.0, 'CatBoost Permutation Importance')"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.inspection import permutation_importance\n",
    "fig, axs = plt.subplots(figsize=(12, 8))\n",
    "\n",
    "perms = permutation_importance(model_dict['cb_0'], Xtrain, ytrain,\n",
    "                              n_jobs=-2, random_state=430)\n",
    "perm_imp = pd.DataFrame({'Mean_Importance': list(perms['importances_mean']),\n",
    "                        'Features': Xtrain.columns.to_list()})\n",
    "perm_imp = perm_imp.sort_values(by='Mean_Importance', ascending=False).head(15)\n",
    "\n",
    "sns.barplot(x=perm_imp['Mean_Importance'], y=perm_imp['Features'])\n",
    "plt.ylabel('')\n",
    "plt.title('CatBoost Permutation Importance')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "overall-fluid",
   "metadata": {},
   "source": [
    "## Running the Best Models for Full Training Data\n",
    "[Back to Top](#Machine-Learning-Processing)\n",
    "\n",
    "Ordinarily this would make for a fine place to finish an ML analysis, however, we are interested in deriving as much precision as possible from our results and can take our findings one step further: we have all of the very best models of each of our selected classifiers and now we can collate their collective, predictive power through stacking techniques.\n",
    "\n",
    "With this approach, our goal is to gather these models together so that we can generate a clear confidence interval of prediction so that we can better inform our client's decisions at the same time as generating our final predictions of fradulent providers."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "adjusted-racing",
   "metadata": {},
   "outputs": [],
   "source": [
    "# ### If starting from here, uncomment these #####\n",
    "# import numpy as np\n",
    "# import pandas as pd\n",
    "# import pickle\n",
    "\n",
    "# from sklearn.pipeline import Pipeline\n",
    "# from catboost import CatBoostClassifier\n",
    "# from lightgbm import LGBMClassifier\n",
    "# from sklearn.linear_model import LogisticRegression\n",
    "# from sklearn.preprocessing import StandardScaler\n",
    "# from sklearn.neighbors import KNeighborsClassifier\n",
    "# from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "# from sklearn.naive_bayes import GaussianNB\n",
    "# from sklearn.svm import SVC\n",
    "# from sklearn.decomposition import PCA \n",
    "\n",
    "\n",
    "def PCAmerge(train, test, n_comp=0.95, r_state = 42):\n",
    "    from sklearn.decomposition import PCA \n",
    "    pca = PCA(n_components = n_comp, random_state = r_state)\n",
    "    train_pca = pd.DataFrame(pca.fit_transform(train), index=train.index)\n",
    "    test_pca = pd.DataFrame(pca.transform(test), index=test.index)\n",
    "    PCA_col_list = []\n",
    "    for num in range(1, train_pca.shape[1]+1):\n",
    "        PCA_col_list.append(('PCA' + str(num)))\n",
    "    train_pca.columns = PCA_col_list\n",
    "    train = pd.concat([train, train_pca], axis=1)\n",
    "    test_pca.columns = PCA_col_list\n",
    "    test = pd.concat([test, test_pca], axis=1)\n",
    "    return train, test\n",
    "\n",
    "model_scores = pickle.load(open('model_scores.pickle', 'rb'))\n",
    "best_models = pickle.load(open('best_models.pickle', 'rb'))\n",
    "model_dict = pickle.load(open('model_dict.pickle', 'rb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "facial-heaven",
   "metadata": {},
   "outputs": [],
   "source": [
    "Xtrain_all = pickle.load(open('Xtrain.pickle', 'rb'))\n",
    "ytrain_all = pickle.load(open('ytrain.pickle', 'rb'))\n",
    "Xtest_all = pickle.load(open('Xtest.pickle', 'rb'))\n",
    "ytest_all = pickle.load(open('ytest.pickle', 'rb'))\n",
    "\n",
    "scaled = StandardScaler()\n",
    "Xtrain_all_Scaled = pd.DataFrame(scaled.fit_transform(Xtrain_all), columns=Xtrain_all.columns, index=Xtrain_all.index)\n",
    "Xtest_all_Scaled = pd.DataFrame(scaled.fit_transform(Xtest_all), columns=Xtest_all.columns, index=Xtest_all.index)\n",
    "Xtrain_all_PCA, Xtest_all_PCA = PCAmerge(Xtrain_all_Scaled, Xtest_all_Scaled, n_comp=0.95)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "lightweight-disclaimer",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "84"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_cb = model_dict['cb_0']\n",
    "best_cb.fit(Xtrain_all, ytrain_all)\n",
    "\n",
    "ytest_cb = ytest_all.copy()\n",
    "ytest_cb['PotentialFraud'] = best_cb.predict(Xtest_all)\n",
    "ytest_cb['ProbabilityFraud'] = best_cb.predict_proba(Xtest_all)[:,1] \n",
    "sum(ytest_cb['PotentialFraud'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "manual-supervisor",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "86"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_lgbm = model_dict['lgbm_1']\n",
    "best_lgbm.fit(Xtrain_all, ytrain_all.values.ravel())\n",
    "\n",
    "ytest_lgbm = ytest_all.copy()\n",
    "ytest_lgbm['PotentialFraud'] = best_lgbm.predict(Xtest_all)\n",
    "ytest_lgbm['ProbabilityFraud'] = best_lgbm.predict_proba(Xtest_all)[:,1] \n",
    "sum(ytest_lgbm['PotentialFraud'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "attached-launch",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\gottl\\Anaconda3\\lib\\site-packages\\sklearn\\model_selection\\_search.py:921: UserWarning: One or more of the test scores are non-finite: [       nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan        nan        nan        nan\n",
      "        nan        nan        nan 0.93160727        nan 0.9314225\n",
      "        nan 0.93123762        nan 0.93105285        nan 0.93123773\n",
      "        nan 0.93049842        nan 0.9306832         nan 0.929944\n",
      "        nan 0.929944          nan 0.929944          nan 0.93031375\n",
      "        nan 0.93031375        nan 0.9306835         nan 0.9306835\n",
      "        nan 0.93031375        nan 0.9306835         nan 0.93031375\n",
      "        nan 0.93049852        nan 0.93049863        nan 0.93049863\n",
      "        nan 0.93086838        nan 0.93086828        nan 0.9306835\n",
      "        nan 0.93105326        nan 0.9306835 ]\n",
      "  category=UserWarning\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "62"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_lgr = model_dict['lgr_3_opt']\n",
    "best_lgr.fit(Xtrain_all_PCA, ytrain_all.values.ravel())\n",
    "\n",
    "ytest_lgr = ytest_all.copy()\n",
    "ytest_lgr['PotentialFraud'] = best_lgr.predict(Xtest_all_PCA)\n",
    "ytest_lgr['ProbabilityFraud'] = best_lgr.predict_proba(Xtest_all_PCA)[:,1] \n",
    "sum(ytest_lgr['PotentialFraud'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "atlantic-acceptance",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "62"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_knn = model_dict['knn_2_opt']\n",
    "best_knn.fit(Xtrain_all, ytrain_all.values.ravel())\n",
    "\n",
    "ytest_knn = ytest_all.copy()\n",
    "ytest_knn['PotentialFraud'] = best_knn.predict(Xtest_all)\n",
    "ytest_knn['ProbabilityFraud'] = best_knn.predict_proba(Xtest_all)[:,1] \n",
    "sum(ytest_knn['PotentialFraud'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "fifth-lyric",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "71"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_lda = model_dict['lda_3']\n",
    "best_lda.fit(Xtrain_all_PCA, ytrain_all.values.ravel())\n",
    "\n",
    "ytest_lda = ytest_all.copy()\n",
    "ytest_lda['PotentialFraud'] = best_lda.predict(Xtest_all_PCA)\n",
    "ytest_lda['ProbabilityFraud'] = best_lda.predict_proba(Xtest_all_PCA)[:,1] \n",
    "sum(ytest_lda['PotentialFraud'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "appointed-vintage",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "27"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_gnb = model_dict['gnb_2_opt']\n",
    "best_gnb.fit(Xtrain_all, ytrain_all.values.ravel())\n",
    "\n",
    "ytest_gnb = ytest_all.copy()\n",
    "ytest_gnb['PotentialFraud'] = best_gnb.predict(Xtest_all)\n",
    "ytest_gnb['ProbabilityFraud'] = best_gnb.predict_proba(Xtest_all)[:,1] \n",
    "sum(ytest_gnb['PotentialFraud'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "given-documentation",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "41"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best_svc = Pipeline(steps=[('scaler', StandardScaler()),\n",
    "                           ('svc', SVC(probability=True))])\n",
    "best_svc.set_params(**model_dict['svc_1_opt_params'])\n",
    "best_svc.fit(Xtrain_all, ytrain_all.values.ravel())\n",
    "\n",
    "ytest_svc = ytest_all.copy()\n",
    "ytest_svc['PotentialFraud'] = best_svc.predict(Xtest_all)\n",
    "ytest_svc['ProbabilityFraud'] = best_svc.predict_proba(Xtest_all)[:,1] \n",
    "sum(ytest_svc['PotentialFraud'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "rocky-section",
   "metadata": {},
   "outputs": [
    {
     "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>PotentialFraud</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Provider</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>PRV51002</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV51006</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV51009</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV51010</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV51018</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV57713</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV57726</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV57745</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV57749</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV57750</th>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1353 rows × 1 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          PotentialFraud\n",
       "Provider                \n",
       "PRV51002             0.0\n",
       "PRV51006             0.0\n",
       "PRV51009             0.0\n",
       "PRV51010             0.0\n",
       "PRV51018             0.0\n",
       "...                  ...\n",
       "PRV57713             0.0\n",
       "PRV57726             0.0\n",
       "PRV57745             0.0\n",
       "PRV57749             0.0\n",
       "PRV57750             0.0\n",
       "\n",
       "[1353 rows x 1 columns]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ytest_stack_confidence = ytest_all.copy()\n",
    "ytest_stack_confidence['PotentialFraud'] = (ytest_cb['PotentialFraud'] + ytest_lgbm['PotentialFraud'] + \n",
    "                                            ytest_lgr['PotentialFraud'] + ytest_knn['PotentialFraud'] + \n",
    "                                            ytest_lda['PotentialFraud'] + ytest_gnb['PotentialFraud'] + \n",
    "                                            ytest_svc['PotentialFraud'])  / 7\n",
    "ytest_stack_confidence\n",
    "\n",
    "ytest_stacked = np.round(ytest_stack_confidence)\n",
    "ytest_stacked"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "continent-poland",
   "metadata": {},
   "source": [
    "Generate Model Fraud Probability Data\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 205,
   "id": "accompanied-loading",
   "metadata": {},
   "outputs": [
    {
     "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>Provider</th>\n",
       "      <th>Percentage</th>\n",
       "      <th>NumModels</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>30</td>\n",
       "      <td>0.022173</td>\n",
       "      <td>1/7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>30</td>\n",
       "      <td>0.022173</td>\n",
       "      <td>2/7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>10</td>\n",
       "      <td>0.007391</td>\n",
       "      <td>3/7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>14</td>\n",
       "      <td>0.010347</td>\n",
       "      <td>4/7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>9</td>\n",
       "      <td>0.006652</td>\n",
       "      <td>5/7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>12</td>\n",
       "      <td>0.008869</td>\n",
       "      <td>6/7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>20</td>\n",
       "      <td>0.014782</td>\n",
       "      <td>7/7</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Provider  Percentage NumModels\n",
       "0        30    0.022173       1/7\n",
       "1        30    0.022173       2/7\n",
       "2        10    0.007391       3/7\n",
       "3        14    0.010347       4/7\n",
       "4         9    0.006652       5/7\n",
       "5        12    0.008869       6/7\n",
       "6        20    0.014782       7/7"
      ]
     },
     "execution_count": 205,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ytest_stack_confidence_group = ytest_stack_confidence.drop('ProbabilityFraud', axis=1).reset_index().groupby('PotentialFraud').agg('count')\n",
    "ytest_stack_confidence_group.index = range(8)\n",
    "ytest_stack_confidence_group['Percentage'] = ytest_stack_confidence_group.Provider / np.sum(ytest_stack_confidence_group.Provider)\n",
    "ytest_stack_graph = ytest_stack_confidence_group.loc[1:].copy()\n",
    "ytest_stack_graph['NumModels'] = list(ytest_stack_graph.index)\n",
    "ytest_stack_graph['NumModels'] = ytest_stack_graph['NumModels'].apply(str) + '/7'\n",
    "ytest_stack_graph.index = ytest_stack_graph.index - 1\n",
    "ytest_stack_graph"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 182,
   "id": "consistent-ending",
   "metadata": {},
   "outputs": [
    {
     "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>PotentialFraud</th>\n",
       "      <th>ProbabilityFraud</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Provider</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>PRV53105</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.987689</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV56573</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.985942</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV53866</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.982490</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV52618</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.960715</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV52080</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.954112</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV54729</th>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.228291</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV54984</th>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.227770</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV56204</th>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.220669</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV51188</th>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.201094</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PRV54920</th>\n",
       "      <td>0.142857</td>\n",
       "      <td>0.196143</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>125 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          PotentialFraud  ProbabilityFraud\n",
       "Provider                                  \n",
       "PRV53105        1.000000          0.987689\n",
       "PRV56573        1.000000          0.985942\n",
       "PRV53866        1.000000          0.982490\n",
       "PRV52618        1.000000          0.960715\n",
       "PRV52080        1.000000          0.954112\n",
       "...                  ...               ...\n",
       "PRV54729        0.142857          0.228291\n",
       "PRV54984        0.142857          0.227770\n",
       "PRV56204        0.142857          0.220669\n",
       "PRV51188        0.142857          0.201094\n",
       "PRV54920        0.142857          0.196143\n",
       "\n",
       "[125 rows x 2 columns]"
      ]
     },
     "execution_count": 182,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ytest_stack_confidence['ProbabilityFraud'] = (ytest_cb['ProbabilityFraud'] + ytest_lgbm['ProbabilityFraud'] + \n",
    "                                              ytest_lgr['ProbabilityFraud'] + ytest_knn['ProbabilityFraud'] + \n",
    "                                              ytest_lda['ProbabilityFraud'] + ytest_gnb['ProbabilityFraud'] + \n",
    "                                              ytest_svc['ProbabilityFraud'])  / 7\n",
    "\n",
    "# Collate Potentially Fraudulent Providers\n",
    "PotentialFraud = ytest_stack_confidence[ytest_stack_confidence.PotentialFraud != 0]\n",
    "PotentialFraud = PotentialFraud.sort_values(['PotentialFraud', 'ProbabilityFraud', 'Provider'], ascending=[False, False, True])\n",
    "PotentialFraud['NumModels'] = (PotentialFraud['PotentialFraud'] * 7).apply(int).apply(str) + '/7'\n",
    "PotentialFraud"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "medical-bearing",
   "metadata": {},
   "source": [
    "Statistics for Probability of Fraud, Grouped by How Many of 7 Models Marked Provider as Fraudulent"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 185,
   "id": "framed-france",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "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 tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe thead tr:last-of-type th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th colspan=\"6\" halign=\"left\">ProbabilityFraud</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th>min</th>\n",
       "      <th>mean</th>\n",
       "      <th>median</th>\n",
       "      <th>max</th>\n",
       "      <th>std</th>\n",
       "      <th>range_</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PotentialFraud</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1.000000</th>\n",
       "      <td>0.822189</td>\n",
       "      <td>0.921706</td>\n",
       "      <td>0.916879</td>\n",
       "      <td>0.987689</td>\n",
       "      <td>0.044557</td>\n",
       "      <td>0.165501</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.857143</th>\n",
       "      <td>0.705410</td>\n",
       "      <td>0.804699</td>\n",
       "      <td>0.798145</td>\n",
       "      <td>0.885285</td>\n",
       "      <td>0.055267</td>\n",
       "      <td>0.179875</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.714286</th>\n",
       "      <td>0.551900</td>\n",
       "      <td>0.655504</td>\n",
       "      <td>0.648673</td>\n",
       "      <td>0.833580</td>\n",
       "      <td>0.075395</td>\n",
       "      <td>0.281679</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.571429</th>\n",
       "      <td>0.494100</td>\n",
       "      <td>0.575918</td>\n",
       "      <td>0.584629</td>\n",
       "      <td>0.674020</td>\n",
       "      <td>0.054462</td>\n",
       "      <td>0.179919</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.428571</th>\n",
       "      <td>0.461860</td>\n",
       "      <td>0.512040</td>\n",
       "      <td>0.497225</td>\n",
       "      <td>0.634824</td>\n",
       "      <td>0.049194</td>\n",
       "      <td>0.172964</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.285714</th>\n",
       "      <td>0.254147</td>\n",
       "      <td>0.380308</td>\n",
       "      <td>0.380284</td>\n",
       "      <td>0.526816</td>\n",
       "      <td>0.081925</td>\n",
       "      <td>0.272669</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.142857</th>\n",
       "      <td>0.196143</td>\n",
       "      <td>0.305068</td>\n",
       "      <td>0.304162</td>\n",
       "      <td>0.441164</td>\n",
       "      <td>0.063558</td>\n",
       "      <td>0.245021</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               ProbabilityFraud                                          \\\n",
       "                            min      mean    median       max       std   \n",
       "PotentialFraud                                                            \n",
       "1.000000               0.822189  0.921706  0.916879  0.987689  0.044557   \n",
       "0.857143               0.705410  0.804699  0.798145  0.885285  0.055267   \n",
       "0.714286               0.551900  0.655504  0.648673  0.833580  0.075395   \n",
       "0.571429               0.494100  0.575918  0.584629  0.674020  0.054462   \n",
       "0.428571               0.461860  0.512040  0.497225  0.634824  0.049194   \n",
       "0.285714               0.254147  0.380308  0.380284  0.526816  0.081925   \n",
       "0.142857               0.196143  0.305068  0.304162  0.441164  0.063558   \n",
       "\n",
       "                          \n",
       "                  range_  \n",
       "PotentialFraud            \n",
       "1.000000        0.165501  \n",
       "0.857143        0.179875  \n",
       "0.714286        0.281679  \n",
       "0.571429        0.179919  \n",
       "0.428571        0.172964  \n",
       "0.285714        0.272669  \n",
       "0.142857        0.245021  "
      ]
     },
     "execution_count": 185,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def range_(df):\n",
    "    return max(df) - min(df)\n",
    "PotentialFraud.groupby('PotentialFraud').agg(['min','mean','median','max', 'std', range_]).sort_values('PotentialFraud', ascending=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "caring-custom",
   "metadata": {},
   "source": [
    "Export Results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "employed-algorithm",
   "metadata": {},
   "outputs": [],
   "source": [
    "ytest_cb.to_csv('ytest_cb.csv', header=True)\n",
    "ytest_lgbm.to_csv('ytest_lgbm.csv', header=True)\n",
    "ytest_lgr.to_csv('ytest_lgr.csv', header=True)\n",
    "ytest_knn.to_csv('ytest_knn.csv', header=True)\n",
    "ytest_lda.to_csv('ytest_lda.csv', header=True)\n",
    "ytest_gnb.to_csv('ytest_gnb.csv', header=True)\n",
    "ytest_svc.to_csv('ytest_svc.csv', header=True)\n",
    "ytest_stack_confidence.to_csv('ytest_stack_confidence.csv', header=True)\n",
    "ytest_stacked.to_csv('ytest_stacked.csv', header=True)\n",
    "ytest_stack_graph.to_csv('ytest_stack_graph.csv', header=True)\n",
    "PotentialFraud.to_csv('PotentialFraud.csv', header=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a49d3bee",
   "metadata": {},
   "source": [
    "### Stacking Visualization\n",
    "[Back to Top](#Machine-Learning-Processing)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dense-davis",
   "metadata": {},
   "source": [
    "#### Grouping Potentially Fraudulent Providers by Model Agreement\n",
    " - How many providers are identified as fraudulent by all 7 models?   \n",
    " - How many are only identified as fraudulent by 1 model?    \n",
    " <b>Answering these questions can help eliminate the biases of a single model choice by displaying the divergences and consensus of a group of models </b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 233,
   "id": "vocal-format",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.lines import Line2D\n",
    "\n",
    "plt.figure(figsize=(10,6))\n",
    "ax = sns.barplot(x=ytest_stack_graph['NumModels'], y=ytest_stack_graph['Provider'], palette=\"coolwarm\")\n",
    "plt.title(\"Confidence of Suspected Provider Fraud\", size=14)\n",
    "plt.ylabel(\"Number of Providers\", size=12)\n",
    "plt.xlabel(\"How Many of the 7 ML Models Identified a Provider as Fraudulent\", size=12)\n",
    "plt.ylim(0,32)\n",
    "\n",
    "patches = ax.patches\n",
    "for i in range(len(patches)):\n",
    "    x = patches[i].get_x() + patches[i].get_width() / 2\n",
    "    y = patches[i].get_height() + .35\n",
    "    ax.annotate('{:.0f}'.format(ytest_stack_graph['Provider'][i]), (x, y), ha='center')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "restricted-damages",
   "metadata": {},
   "source": [
    "#### Is there variation in the probability of Fraud among the models identified as Fraudulent?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 234,
   "id": "adolescent-worker",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<function matplotlib.pyplot.show(close=None, block=None)>"
      ]
     },
     "execution_count": 234,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(10,8))\n",
    "PotentialFraudorder = PotentialFraud.sort_values('PotentialFraud', ascending=False)\n",
    "ax = sns.boxplot(y=\"NumModels\", x=\"ProbabilityFraud\", data=PotentialFraudorder, palette=\"coolwarm\",\n",
    "                 flierprops=red_circle, showmeans=True, meanprops=yellow_diamond)\n",
    "\n",
    "plt.title(\"Machine Learning Confidence in Provider Fraud\", size=16)\n",
    "ax.set_xlabel('Mean Model Probability of Fraud by Provider', fontsize=13)\n",
    "ax.set_ylabel('How Many ML Models Identified a Provider as Fraudulent', fontsize=13)\n",
    "ax.tick_params(labelright = False)\n",
    "plt.rcParams['ytick.right'] = plt.rcParams['ytick.labelright'] = False\n",
    "legend_elements = [Line2D([0], [0], marker='o', color='red', markeredgecolor='r', markersize=2.5, label='Outlier',\n",
    "                         linestyle = 'None'),\n",
    "                   Line2D([1], [1], color='black', lw=2, label='Median', marker='|',\n",
    "                         linestyle='None',\n",
    "                          markersize=10, markeredgewidth=1.5,),\n",
    "                   Line2D([0], [0], marker='D', color='yellow', label='Mean',\n",
    "                          markerfacecolor='yellow', markeredgecolor='black', markersize=5, linestyle = 'None')]\n",
    "\n",
    "plt.legend(handles=legend_elements[::-1], loc='lower right')\n",
    "plt.show      "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8fe71459",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9a8572a1",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "21a7cf81",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e8907269",
   "metadata": {},
   "outputs": [],
   "source": [
    "### For Importing Results\n",
    "# svc_preds = pd.read_csv('ytest_svc.csv', index_col='Provider')\n",
    "# lgr_preds = pd.read_csv('ytest_lgr.csv', index_col='Provider')\n",
    "# lda_preds = pd.read_csv('ytest_lda.csv', index_col='Provider')\n",
    "# knn_preds = pd.read_csv('ytest_knn.csv', index_col='Provider')\n",
    "# gnb_preds = pd.read_csv('ytest_gnb.csv', index_col='Provider')\n",
    "# lgbm_preds = pd.read_csv('ytest_lgbm.csv', index_col='Provider')\n",
    "# cb_preds = pd.read_csv('ytest_cb.csv', index_col='Provider')\n",
    "# stacked_preds = pd.read_csv('ytest_stacked.csv', index_col='Provider')"
   ]
  }
 ],
 "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}