1.5. Modeling-Feature importance and  Evaluation_another model testing.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#  Modeling and feature importance visual\n",
    "\n",
    "## Feature Selection\n",
    "\n",
    "In this section, I will select specific features for model training. The original dataset contains 265 features, most of which are categorical attributes, dominantly the list of drug names that are approved as pain medications that need special AED licensed prescriber.Here,my main target is to predict the overdoses using the  the list of prescriber specialists with NPI information and the FDA approved drug lists. The data set has only total opiod cliams and total claim counts and i took the standard prescription for opiod drugs is 84 days which is 12 weeks. Therefore, an additional feature was created in the data cleaning, pre processing that shows avg_op_supply which i limit to the standard to be 84 and any ones who prescribes the category of medications under the controlled substance act(CSA). The CSA is used for organizing drugs based on the risk of abuse or harm to users in five and my capestone focuses on the approved by [FDA](https://www.fda.gov/drugs/drugsafety/informationbydrugclass/ucm251735.htm) and [DEA](https://www.deadiversion.usdoj.gov/schedules/) to be prescribed by licensed and certified health professional.  These drug lists that are considered controlled substances under the Controlled Substances Act (CSA) are divided into five schedules.  An updated and complete list of the schedules is published annually in [Title 21 Code of Federal Regulations (C.F.R.) §§ 1308.11 through 1308.15](https://www.deadiversion.usdoj.gov/21cfr/cfr/2108cfrt.htm).\n",
    "\n",
    "Therfore, for analysis and visualization  i only took those physicians who have greater opood cliams from the center for medical services dataset and created a feature average opiod supply  by dividing the total cliam by the opiod cliam and reviewed those specialists who prescribe more than the set limit of 12 weeks and that also become a new feature that measures the length of opiod prescription supply more than 84 days. As been visualized in notebook 1.3, significant amount of prescribers are family physicians, followed by internal medicine,nurse practitioners, Physicain Assistant and physical medicine and rehabilitation. When i group based on speciality and the vsuals can be accessed   [here top 20](https://plot.ly/~kiros/112) and the lists are as follows:-\n",
    "  > $'Cardiology', 'Nephrology', 'Endocrinology', 'Neurology'  'General Practice', 'Internal Medicine', 'Urology'$,   > $'Pulmonary Disease', 'Gastroenterology', 'Family Practice', 'Infectious Disease','Psychiatry'$ \n",
    "  >$'Nurse Practitioner', 'Ophthalmology','Allergy/Immunology', 'Otolaryngology', 'Psychiatry & Neurology'$ , $'Rheumatology', 'Dermatology', 'Obstetrics/Gynecology'$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# importing necessary libraries and packages\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.model_selection import train_test_split\n",
    "np.random.seed(2018)\n",
    "\n",
    "# Configure visual settings:\n",
    "from matplotlib import rcParams\n",
    "rcParams.update({'figure.autolayout':True})\n",
    "plt.style.use('fivethirtyeight')\n",
    "sns.set(font_scale=2)\n",
    "cmap = sns.diverging_palette(220, 10, as_cmap=True) # one of the many color mappings\n",
    "sns.set_style('whitegrid')\n",
    "plt.rcParams['figure.figsize'] = (10.0, 8.0) \n",
    "%config InlineBackend.figure_format = 'retina'\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/IPython/core/interactiveshell.py:2785: DtypeWarning: Columns (5,6) have mixed types. Specify dtype option on import or set low_memory=False.\n",
      "  interactivity=interactivity, compiler=compiler, result=result)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>NPPES Provider Last Name</th>\n",
       "      <th>NPPES Provider First Name</th>\n",
       "      <th>NPPES Provider ZIP Code</th>\n",
       "      <th>NPPES Provider State</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>EKLUND</td>\n",
       "      <td>D.</td>\n",
       "      <td>39208.0</td>\n",
       "      <td>MS</td>\n",
       "      <td>129</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>RIGGLE</td>\n",
       "      <td>ERIN</td>\n",
       "      <td>80120.0</td>\n",
       "      <td>CO</td>\n",
       "      <td>537</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 264 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease   \n",
       "\n",
       "  NPPES Provider Last Name NPPES Provider First Name  NPPES Provider ZIP Code  \\\n",
       "0                   EKLUND                        D.                  39208.0   \n",
       "1                   RIGGLE                      ERIN                  80120.0   \n",
       "\n",
       "  NPPES Provider State  Totalclaimcount      ...        VYTORIN  \\\n",
       "0                   MS              129      ...              0   \n",
       "1                   CO              537      ...              0   \n",
       "\n",
       "   WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  ZOLPIDEM.TARTRATE  \\\n",
       "0                0        0      0                0                  0   \n",
       "1                0       16      0                0                  0   \n",
       "\n",
       "   avg_op_supply  op_longer  Opioid_Prescriber  op_prescriber  \n",
       "0            4.5        1.0                  1            1.0  \n",
       "1            4.5        0.0                  0            0.0  \n",
       "\n",
       "[2 rows x 264 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## Read Data\n",
    "df=pd.read_csv('./inputs/cleaned_prescriber_info18.csv')\n",
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(25000, 264)\n"
     ]
    }
   ],
   "source": [
    "print(df.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "opioids=pd.read_csv('./inputs/list_opioids.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "name=opioids['Drug Name']\n",
    "import re\n",
    "new_name=name.apply(lambda x:re.sub(\"\\ |-\",\".\",str(x)))\n",
    "columns=df.columns\n",
    "Abandoned_variables = set(columns).intersection(set(new_name))\n",
    "Kept_variable=[]\n",
    "for each in columns:\n",
    "    if each in Abandoned_variables:\n",
    "        pass\n",
    "    else:\n",
    "        Kept_variable.append(each)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(25000, 253)\n"
     ]
    }
   ],
   "source": [
    "df=df[Kept_variable]\n",
    "print(df.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "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>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>NPPES Provider Last Name</th>\n",
       "      <th>NPPES Provider First Name</th>\n",
       "      <th>NPPES Provider ZIP Code</th>\n",
       "      <th>NPPES Provider State</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>EKLUND</td>\n",
       "      <td>D.</td>\n",
       "      <td>39208.0</td>\n",
       "      <td>MS</td>\n",
       "      <td>129</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>RIGGLE</td>\n",
       "      <td>ERIN</td>\n",
       "      <td>80120.0</td>\n",
       "      <td>CO</td>\n",
       "      <td>537</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 253 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease   \n",
       "\n",
       "  NPPES Provider Last Name NPPES Provider First Name  NPPES Provider ZIP Code  \\\n",
       "0                   EKLUND                        D.                  39208.0   \n",
       "1                   RIGGLE                      ERIN                  80120.0   \n",
       "\n",
       "  NPPES Provider State  Totalclaimcount      ...        VYTORIN  \\\n",
       "0                   MS              129      ...              0   \n",
       "1                   CO              537      ...              0   \n",
       "\n",
       "   WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  ZOLPIDEM.TARTRATE  \\\n",
       "0                0        0      0                0                  0   \n",
       "1                0       16      0                0                  0   \n",
       "\n",
       "   avg_op_supply  op_longer  Opioid_Prescriber  op_prescriber  \n",
       "0            4.5        1.0                  1            1.0  \n",
       "1            4.5        0.0                  0            0.0  \n",
       "\n",
       "[2 rows x 253 columns]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# df['NPPES Provider ZIP Code'] = df['NPPES Provider ZIP Code'].str.replace(',', '')\n",
    "# df['NPPES Provider ZIP Code'] = df['NPPES Provider ZIP Code'].str.replace(' ', '')\n",
    "df['NPPES Provider ZIP Code'] = df['NPPES Provider ZIP Code'].astype(float)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NPI                                  0\n",
       "Gender                               0\n",
       "State                                0\n",
       "Credentials                         82\n",
       "Specialty                            0\n",
       "NPPES Provider Last Name          2610\n",
       "NPPES Provider First Name         2610\n",
       "NPPES Provider ZIP Code           2610\n",
       "NPPES Provider State                 0\n",
       "Totalclaimcount                      0\n",
       "OpioidClaimCount                     0\n",
       "ACYCLOVIR                            0\n",
       "ADVAIR.DISKUS                        0\n",
       "AGGRENOX                             0\n",
       "ALENDRONATE.SODIUM                   0\n",
       "ALLOPURINOL                          0\n",
       "ALPRAZOLAM                           0\n",
       "AMIODARONE.HCL                       0\n",
       "AMITRIPTYLINE.HCL                    0\n",
       "AMLODIPINE.BESYLATE                  0\n",
       "AMLODIPINE.BESYLATE.BENAZEPRIL       0\n",
       "AMOXICILLIN                          0\n",
       "AMOX.TR.POTASSIUM.CLAVULANATE        0\n",
       "AMPHETAMINE.SALT.COMBO               0\n",
       "ATENOLOL                             0\n",
       "ATORVASTATIN.CALCIUM                 0\n",
       "AVODART                              0\n",
       "AZITHROMYCIN                         0\n",
       "BACLOFEN                             0\n",
       "BD.ULTRA.FINE.PEN.NEEDLE             0\n",
       "                                  ... \n",
       "TERAZOSIN.HCL                        0\n",
       "TIMOLOL.MALEATE                      0\n",
       "TIZANIDINE.HCL                       0\n",
       "TOLTERODINE.TARTRATE.ER              0\n",
       "TOPIRAMATE                           0\n",
       "TOPROL.XL                            0\n",
       "TORSEMIDE                            0\n",
       "TRAVATAN.Z                           0\n",
       "TRAZODONE.HCL                        0\n",
       "TRIAMCINOLONE.ACETONIDE              0\n",
       "TRIAMTERENE.HYDROCHLOROTHIAZID       0\n",
       "VALACYCLOVIR                         0\n",
       "VALSARTAN                            0\n",
       "VALSARTAN.HYDROCHLOROTHIAZIDE        0\n",
       "VENLAFAXINE.HCL                      0\n",
       "VENLAFAXINE.HCL.ER                   0\n",
       "VENTOLIN.HFA                         0\n",
       "VERAPAMIL.ER                         0\n",
       "VESICARE                             0\n",
       "VOLTAREN                             0\n",
       "VYTORIN                              0\n",
       "WARFARIN.SODIUM                      0\n",
       "XARELTO                              0\n",
       "ZETIA                                0\n",
       "ZIPRASIDONE.HCL                      0\n",
       "ZOLPIDEM.TARTRATE                    0\n",
       "avg_op_supply                     2610\n",
       "op_longer                         2610\n",
       "Opioid_Prescriber                    0\n",
       "op_prescriber                     2610\n",
       "Length: 253, dtype: int64"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.isnull().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.fillna(0,inplace = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['NPI', 'Gender', 'State', 'Credentials', 'Specialty',\n",
       "       'NPPES Provider Last Name', 'NPPES Provider First Name',\n",
       "       'NPPES Provider ZIP Code', 'NPPES Provider State', 'Totalclaimcount',\n",
       "       ...\n",
       "       'VYTORIN', 'WARFARIN.SODIUM', 'XARELTO', 'ZETIA', 'ZIPRASIDONE.HCL',\n",
       "       'ZOLPIDEM.TARTRATE', 'avg_op_supply', 'op_longer', 'Opioid_Prescriber',\n",
       "       'op_prescriber'],\n",
       "      dtype='object', length=253)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.drop(['NPPES Provider Last Name', 'NPPES Provider First Name','NPPES Provider ZIP Code',\n",
    "         'NPPES Provider State'], axis =1, inplace = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "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>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>OpioidClaimCount</th>\n",
       "      <th>ACYCLOVIR</th>\n",
       "      <th>ADVAIR.DISKUS</th>\n",
       "      <th>AGGRENOX</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>129</td>\n",
       "      <td>29</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>537</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>61</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 249 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  Totalclaimcount  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist              129   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease              537   \n",
       "\n",
       "   OpioidClaimCount  ACYCLOVIR  ADVAIR.DISKUS  AGGRENOX      ...        \\\n",
       "0                29          0              0         0      ...         \n",
       "1                 0          0             61         0      ...         \n",
       "\n",
       "   VYTORIN  WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  \\\n",
       "0        0                0        0      0                0   \n",
       "1        0                0       16      0                0   \n",
       "\n",
       "   ZOLPIDEM.TARTRATE  avg_op_supply  op_longer  Opioid_Prescriber  \\\n",
       "0                  0            4.5        1.0                  1   \n",
       "1                  0            4.5        0.0                  0   \n",
       "\n",
       "   op_prescriber  \n",
       "0            1.0  \n",
       "1            0.0  \n",
       "\n",
       "[2 rows x 249 columns]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(17500, 249)\n",
      "(7500, 249)\n"
     ]
    }
   ],
   "source": [
    "train,test = train_test_split(df,test_size=0.3,random_state=42)\n",
    "print(train.shape)\n",
    "print(test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/ipykernel/__main__.py:3: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  app.launch_new_instance()\n",
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/ipykernel/__main__.py:4: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n"
     ]
    }
   ],
   "source": [
    "Categorical_columns=['Gender','State','Credentials','Specialty']\n",
    "for col in Categorical_columns:\n",
    "    train[col]=pd.factorize(train[col], sort=True)[0]\n",
    "    test[col] =pd.factorize(test[col],sort=True)[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split, cross_val_score, GridSearchCV\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from sklearn.neighbors import KNeighborsRegressor\n",
    "from sklearn.tree import DecisionTreeRegressor\n",
    "from sklearn.ensemble import BaggingRegressor, RandomForestRegressor, AdaBoostRegressor\n",
    "from sklearn.svm import SVR\n",
    "\n",
    "y = train['Opioid_Prescriber']\n",
    "\n",
    "X = train.iloc[:,1:245]\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "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>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>OpioidClaimCount</th>\n",
       "      <th>ACYCLOVIR</th>\n",
       "      <th>ADVAIR.DISKUS</th>\n",
       "      <th>AGGRENOX</th>\n",
       "      <th>ALENDRONATE.SODIUM</th>\n",
       "      <th>...</th>\n",
       "      <th>VENTOLIN.HFA</th>\n",
       "      <th>VERAPAMIL.ER</th>\n",
       "      <th>VESICARE</th>\n",
       "      <th>VOLTAREN</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>4913</th>\n",
       "      <td>0</td>\n",
       "      <td>18</td>\n",
       "      <td>517</td>\n",
       "      <td>48</td>\n",
       "      <td>153</td>\n",
       "      <td>36</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9338</th>\n",
       "      <td>0</td>\n",
       "      <td>6</td>\n",
       "      <td>284</td>\n",
       "      <td>25</td>\n",
       "      <td>229</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 244 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      Gender  State  Credentials  Specialty  Totalclaimcount  \\\n",
       "4913       0     18          517         48              153   \n",
       "9338       0      6          284         25              229   \n",
       "\n",
       "      OpioidClaimCount  ACYCLOVIR  ADVAIR.DISKUS  AGGRENOX  \\\n",
       "4913                36          0              0         0   \n",
       "9338                 0          0              0         0   \n",
       "\n",
       "      ALENDRONATE.SODIUM        ...          VENTOLIN.HFA  VERAPAMIL.ER  \\\n",
       "4913                   0        ...                     0             0   \n",
       "9338                   0        ...                     0             0   \n",
       "\n",
       "      VESICARE  VOLTAREN  VYTORIN  WARFARIN.SODIUM  XARELTO  ZETIA  \\\n",
       "4913         0         0        0                0        0      0   \n",
       "9338         0         0        0                0        0     14   \n",
       "\n",
       "      ZIPRASIDONE.HCL  ZOLPIDEM.TARTRATE  \n",
       "4913                0                  0  \n",
       "9338                0                  0  \n",
       "\n",
       "[2 rows x 244 columns]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.iloc[:,1:245][:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "X =train.iloc[:,1:245]\n",
    "y = train['Opioid_Prescriber']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "====================================Models,5Fold CV and Test Scores==========================================\n",
      "Linear Regrssion: \n",
      "Five cross-val score means: 0.558833316275933\n",
      "Test score: 0.5913652537140699\n",
      "-----------------------------------------------------------------------------------------------------------------\n",
      "k-nearest:\n",
      "Five cross-val score means: 0.5671962663791562\n",
      "Test score: 0.5269664604001213\n",
      "------------------------------------------------------------------------------------------------------------\n",
      "Decision Tree Regressor: \n",
      "Five cross-val score means: 0.02158379846227975\n",
      "Test score: 0.3185779003291218\n",
      "------------------------------------------------------------------------------------------------------------\n",
      "Bagging Regressor: \n",
      "Five cross-val score means: 0.5475622646936321\n",
      "Test score: 0.6222648709399\n",
      "-----------------------------------------------------------------------------------------------------------\n",
      "Random Forest Regressor: \n",
      "Five cross-val score means: 0.5617690592939093\n",
      "Test score: 0.5484221752906351\n",
      "------------------------------------------------------------------------------------------------------------\n",
      "AdaBoost Regressor:\n",
      "Five cross-val score means: -2.2015968271310635\n",
      "Test score: -1.308676658081274\n",
      "----------------------------------------------------------------------------------------------------------\n",
      "Support Vector Regressor:\n",
      "five cross-val score means: -0.006076829101235237\n",
      "Test score: -0.0060379236471603814\n",
      "----------------------------------------------------------------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "print(\"====================================Models,5Fold CV and Test Scores==========================================\")\n",
    "lr_model = LinearRegression()\n",
    "lr_model.fit(X_train, y_train)\n",
    "\n",
    "print(\"Linear Regrssion: \")\n",
    "print(\"Five cross-val score means: {}\".format(cross_val_score(lr_model, X_train, y_train, cv = 5).mean()))\n",
    "print(\"Test score:\", lr_model.score(X_test, y_test))\n",
    "print(\"-----------------------------------------------------------------------------------------------------------------\")\n",
    "\n",
    "knn_reg = KNeighborsRegressor()\n",
    "knn_reg_params = {\n",
    "    'n_neighbors':[1,2,3,4,5,6,7,8,9,10]\n",
    "}\n",
    "gs_knn_reg = GridSearchCV(knn_reg, knn_reg_params)\n",
    "gs_knn_reg.fit(X_train, y_train)\n",
    "\n",
    "print(\"k-nearest:\")\n",
    "print(\"Five cross-val score means: {}\".format(cross_val_score(gs_knn_reg, X_train, y_train, cv = 5).mean()))\n",
    "print(\"Test score:\", gs_knn_reg.score(X_test, y_test))\n",
    "print(\"------------------------------------------------------------------------------------------------------------\")\n",
    "\n",
    "tree_reg = DecisionTreeRegressor()\n",
    "tree_reg_params = {\n",
    "#     'n_neighbors':[1,2,3,4,5,6,7,8,9,10]\n",
    "}\n",
    "gs_tree_reg = GridSearchCV(tree_reg, tree_reg_params)\n",
    "gs_tree_reg.fit(X_train, y_train)\n",
    "\n",
    "print(\"Decision Tree Regressor: \")\n",
    "print(\"Five cross-val score means: {}\".format(cross_val_score(gs_tree_reg, X_train, y_train, cv = 5).mean()))\n",
    "print(\"Test score:\", gs_tree_reg.score(X_test, y_test))\n",
    "print(\"------------------------------------------------------------------------------------------------------------\")\n",
    "\n",
    "bag_reg = BaggingRegressor()\n",
    "bag_reg_params = {}\n",
    "\n",
    "gs_bag_reg = GridSearchCV(bag_reg, bag_reg_params)\n",
    "gs_bag_reg.fit(X_train, y_train)\n",
    "\n",
    "print(\"Bagging Regressor: \")\n",
    "print(\"Five cross-val score means: {}\".format(cross_val_score(gs_bag_reg, X_train, y_train, cv = 5).mean()))\n",
    "print(\"Test score:\", gs_bag_reg.score(X_test, y_test))\n",
    "print(\"-----------------------------------------------------------------------------------------------------------\")\n",
    "\n",
    "rf_reg = RandomForestRegressor()\n",
    "rf_reg_params = {}\n",
    "\n",
    "gs_rf_reg = GridSearchCV(rf_reg, rf_reg_params)\n",
    "gs_rf_reg.fit(X_train, y_train)\n",
    "\n",
    "print(\"Random Forest Regressor: \")\n",
    "print(\"Five cross-val score means: {}\".format(cross_val_score(gs_rf_reg, X_train, y_train, cv = 5).mean()))\n",
    "print(\"Test score:\", gs_rf_reg.score(X_test, y_test))\n",
    "print(\"------------------------------------------------------------------------------------------------------------\")\n",
    "\n",
    "ada_reg = AdaBoostRegressor()\n",
    "ada_reg_params = {}\n",
    "\n",
    "gs_ada_reg = GridSearchCV(ada_reg, ada_reg_params)\n",
    "gs_ada_reg.fit(X_train, y_train)\n",
    "\n",
    "print(\"AdaBoost Regressor:\")\n",
    "print(\"Five cross-val score means: {}\".format(cross_val_score(gs_ada_reg, X_train, y_train, cv = 5).mean()))\n",
    "print(\"Test score:\", gs_ada_reg.score(X_test, y_test))\n",
    "print(\"----------------------------------------------------------------------------------------------------------\")\n",
    "\n",
    "svr = SVR()\n",
    "svr_params = {\n",
    "    'C':[.1,1,.1]\n",
    "}\n",
    "\n",
    "gs_svr = GridSearchCV(svr, svr_params)\n",
    "gs_svr.fit(X_train, y_train)\n",
    "\n",
    "print(\"Support Vector Regressor:\")\n",
    "print(\"five cross-val score means: {}\".format(cross_val_score(gs_svr, X_train, y_train, cv = 5).mean()))\n",
    "print(\"Test score:\", gs_svr.score(X_test, y_test))\n",
    "print(\"----------------------------------------------------------------------------------------------------------------\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.pipeline import Pipeline\n",
    "\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "\n",
    "from sklearn.svm import SVR\n",
    "\n",
    "from sklearn.neighbors import KNeighborsRegressor\n",
    "\n",
    "from sklearn.model_selection import cross_val_score\n",
    "\n",
    "from sklearn.ensemble import RandomForestRegressor\n",
    "\n",
    "from sklearn.ensemble import AdaBoostRegressor\n",
    "from sklearn.model_selection import cross_val_score\n",
    "from sklearn import svm\n",
    "from sklearn.datasets import samples_generator\n",
    "from sklearn.feature_selection import SelectKBest\n",
    "from sklearn.feature_selection import f_regression"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "=============================================<Root-Mean-Square Error>============================================\n",
      "RMSE scores\n",
      "\n",
      "\n",
      "LR:\n",
      "-----------------------------------------------------------------------------------------------------------\n",
      "Train score: 5.196921296239649\n",
      "Test score: 7.115028225059413\n",
      "\n",
      "\n",
      "KNN:\n",
      "--------------------------------------------------------------------------------------------------------\n",
      "Train score: 6.440482734791147\n",
      "Test score: 7.65517192118225\n",
      "\n",
      "\n",
      "Tree:\n",
      "------------------------------------------------------------------------------------------------------\n",
      "Train score: 0.0\n",
      "Test score: 9.187917531814735\n",
      "\n",
      "\n",
      "Bag:\n",
      "----------------------------------------------------------------------------------------------------------\n",
      "Train score: 3.055023279468626\n",
      "Test score: 6.840733461092772\n",
      "\n",
      "\n",
      "Random Forest:\n",
      "------------------------------------------------------------------------------------------\n",
      "Train score: 3.107677497516663\n",
      "Test score: 7.479546815912617\n",
      "\n",
      "\n",
      "AdaBoost:\n",
      "---------------------------------------------------------------------------------------------\n",
      "Train score: 15.967394170681626\n",
      "Test score: 16.911827385705465\n",
      "\n",
      "\n",
      "SVR:\n",
      "--------------------------------------------------------------------------------------\n",
      "Train score: 11.014892519291795\n",
      "Test score: 11.163905647360451\n"
     ]
    }
   ],
   "source": [
    "print(\"=============================================<Root-Mean-Square Error>============================================\")\n",
    "from sklearn.metrics import mean_squared_error\n",
    "import numpy as np\n",
    "print(\"RMSE scores\")\n",
    "print(\"\\n\")\n",
    "print(\"LR:\\n-------------------------------------------------------------------------------------------------------\")\n",
    "print(\"Train score:\", np.sqrt(mean_squared_error(lr_model.predict(X_train), y_train)))\n",
    "print(\"Test score:\", np.sqrt(mean_squared_error(lr_model.predict(X_test), y_test)))\n",
    "\n",
    "print(\"\\n\")\n",
    "print(\"KNN:\\n-------------------------------------------------------------------------------------------------------\")\n",
    "print(\"Train score:\", np.sqrt(mean_squared_error(gs_knn_reg.predict(X_train), y_train)))\n",
    "print(\"Test score:\", np.sqrt(mean_squared_error(gs_knn_reg.predict(X_test), y_test)))\n",
    "\n",
    "print(\"\\n\")\n",
    "print(\"Tree:\\n-----------------------------------------------------------------------------------------------------\")\n",
    "print(\"Train score:\", np.sqrt(mean_squared_error(gs_tree_reg.predict(X_train), y_train)))\n",
    "print(\"Test score:\", np.sqrt(mean_squared_error(gs_tree_reg.predict(X_test), y_test)))\n",
    "\n",
    "print(\"\\n\")\n",
    "print(\"Bag:\\n------------------------------------------------------------------------------------------------------\")\n",
    "print(\"Train score:\", np.sqrt(mean_squared_error(gs_bag_reg.predict(X_train), y_train)))\n",
    "print(\"Test score:\", np.sqrt(mean_squared_error(gs_bag_reg.predict(X_test), y_test)))\n",
    "\n",
    "print(\"\\n\")\n",
    "print(\"Random Forest:\\n--------------------------------------------------------------------------------------------\")\n",
    "print(\"Train score:\", np.sqrt(mean_squared_error(gs_rf_reg.predict(X_train), y_train)))\n",
    "print(\"Test score:\", np.sqrt(mean_squared_error(gs_rf_reg.predict(X_test), y_test)))\n",
    "\n",
    "print(\"\\n\")\n",
    "print(\"AdaBoost:\\n------------------------------------------------------------------------------------------------\")\n",
    "print(\"Train score:\", np.sqrt(mean_squared_error(gs_ada_reg.predict(X_train), y_train)))\n",
    "print(\"Test score:\", np.sqrt(mean_squared_error(gs_ada_reg.predict(X_test), y_test)))\n",
    "\n",
    "print(\"\\n\")\n",
    "print(\"SVR:\\n---------------------------------------------------------------------------------------------------\")\n",
    "print(\"Train score:\", np.sqrt(mean_squared_error(gs_svr.predict(X_train), y_train)))\n",
    "print(\"Test score:\", np.sqrt(mean_squared_error(gs_svr.predict(X_test), y_test)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "According to the models the data was split in to training,validation and test in the ratio of 70:30:1. The models were developed and trained  using the training dataset and then calculated the root mean square error(RMSE) for each of the models pipelined above. Then i tried to compare the models using the RMSE of the validation performance. The model with the lowest RMSE is technically supposed to be chosen as highest performing model. bagging tree was with the lowest RMSE(3.01 on traing and 6.8 on test). The thing here is that to generalize the model its important to  evaluate it using the methods you choose."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['NPI', 'Gender', 'State', 'Credentials', 'Specialty', 'Totalclaimcount',\n",
       "       'OpioidClaimCount', 'ACYCLOVIR', 'ADVAIR.DISKUS', 'AGGRENOX',\n",
       "       ...\n",
       "       'VYTORIN', 'WARFARIN.SODIUM', 'XARELTO', 'ZETIA', 'ZIPRASIDONE.HCL',\n",
       "       'ZOLPIDEM.TARTRATE', 'avg_op_supply', 'op_longer', 'Opioid_Prescriber',\n",
       "       'op_prescriber'],\n",
       "      dtype='object', length=249)"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.columns"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 3. Model Building step 2_Rework with tunning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Ineligible and not Opioid_Prescriber:\t\t 11123 \n",
      "Ineligible and has a Opioid_Prescriber:\t 0 \n",
      "Eligible and noOpioid_Prescriber:\t\t 4 \n",
      "Eligible and has prescribed opiods:\t 13331\n"
     ]
    }
   ],
   "source": [
    "ineligible_noOpioidsPrescriber = (df['Opioid_Prescriber'] == 0) & (df['op_longer'] == 0)\n",
    "ineligible_yesOpioid_Prescriber = (df['Opioid_Prescriber'] == 0) & (df['op_longer'] == 1)\n",
    "eligible_noOpioid_Prescriber = (df['Opioid_Prescriber'] == 1) & (df['op_longer'] == 0)\n",
    "eligible_yesOpioid_Prescriber = (df['Opioid_Prescriber'] == 1) & (df['op_longer'] == 1)\n",
    "print(\n",
    "\"Ineligible and not Opioid_Prescriber:\\t\\t\", df[ineligible_noOpioidsPrescriber].shape[0],\n",
    "\"\\nIneligible and has a Opioid_Prescriber:\\t\", df[ineligible_yesOpioid_Prescriber].shape[0],\n",
    "\"\\nEligible and noOpioid_Prescriber:\\t\\t\", df[eligible_noOpioid_Prescriber].shape[0],\n",
    "\"\\nEligible and has prescribed opiods:\\t\", df[eligible_yesOpioid_Prescriber].shape[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/cross_validation.py:41: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.\n",
      "  \"This module will be removed in 0.20.\", DeprecationWarning)\n"
     ]
    }
   ],
   "source": [
    "import re\n",
    "from sklearn import svm\n",
    "from sklearn.pipeline import Pipeline\n",
    "from imblearn.over_sampling import SMOTE\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from sklearn.neighbors import KNeighborsRegressor\n",
    "from sklearn.tree import DecisionTreeRegressor\n",
    "from sklearn.metrics import confusion_matrix as cm\n",
    "from sklearn.ensemble import BaggingRegressor, RandomForestRegressor, AdaBoostRegressor\n",
    "from sklearn.svm import SVR\n",
    "from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier\n",
    "from sklearn.model_selection import train_test_split,GridSearchCV\n",
    "from sklearn.cross_validation import cross_val_score\n",
    "from sklearn import cross_validation,metrics\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from matplotlib import rcParams"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>OpioidClaimCount</th>\n",
       "      <th>ACYCLOVIR</th>\n",
       "      <th>ADVAIR.DISKUS</th>\n",
       "      <th>AGGRENOX</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>129</td>\n",
       "      <td>29</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>537</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>61</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 249 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  Totalclaimcount  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist              129   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease              537   \n",
       "\n",
       "   OpioidClaimCount  ACYCLOVIR  ADVAIR.DISKUS  AGGRENOX      ...        \\\n",
       "0                29          0              0         0      ...         \n",
       "1                 0          0             61         0      ...         \n",
       "\n",
       "   VYTORIN  WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  \\\n",
       "0        0                0        0      0                0   \n",
       "1        0                0       16      0                0   \n",
       "\n",
       "   ZOLPIDEM.TARTRATE  avg_op_supply  op_longer  Opioid_Prescriber  \\\n",
       "0                  0            4.5        1.0                  1   \n",
       "1                  0            4.5        0.0                  0   \n",
       "\n",
       "   op_prescriber  \n",
       "0            1.0  \n",
       "1            0.0  \n",
       "\n",
       "[2 rows x 249 columns]"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# df.drop(columns=[''], axis=1, inplace=True)\n",
    "\n",
    "df = pd.get_dummies(df, columns=['Gender'], drop_first=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['NPI', 'State', 'Credentials', 'Specialty', 'Totalclaimcount',\n",
       "       'OpioidClaimCount', 'ACYCLOVIR', 'ADVAIR.DISKUS', 'AGGRENOX',\n",
       "       'ALENDRONATE.SODIUM',\n",
       "       ...\n",
       "       'WARFARIN.SODIUM', 'XARELTO', 'ZETIA', 'ZIPRASIDONE.HCL',\n",
       "       'ZOLPIDEM.TARTRATE', 'avg_op_supply', 'op_longer', 'Opioid_Prescriber',\n",
       "       'op_prescriber', 'Gender_M'],\n",
       "      dtype='object', length=249)"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(25000, 249)"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Ustandardization and applying all models at once\n",
    "ss  = StandardScaler()\n",
    "\n",
    "logreg = LogisticRegression()\n",
    "rfc = RandomForestClassifier(n_estimators=14, random_state=42)\n",
    "adb = AdaBoostClassifier(n_estimators=100, learning_rate=0.1, random_state=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# making pipeline for choosing best model\n",
    "pipe_logreg = Pipeline([\n",
    "    ('ss', ss),\n",
    "    ('logreg', logreg)\n",
    "])\n",
    "pipe_rfc = Pipeline([\n",
    "    ('ss', ss),\n",
    "    ('rfc', rfc),\n",
    "])\n",
    "pipe_adb = Pipeline([\n",
    "    ('ss', ss),\n",
    "    ('adb', adb)\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Score Train/Test: Logistic Regression\n",
      "0.8729904761904762\n",
      "0.8464\n",
      "\n",
      "Score Train/Test: Random Forest\n",
      "0.9954285714285714\n",
      "0.8731428571428571\n",
      "\n",
      "Score Train/Test: Adaboost\n",
      "0.44739047619047617\n",
      "0.4342857142857143\n",
      "CPU times: user 2min 47s, sys: 3.5 s, total: 2min 51s\n",
      "Wall time: 2min 45s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "pipes    = [ \n",
    "    pipe_logreg, \n",
    "    pipe_rfc, \n",
    "    pipe_adb,\n",
    "]\n",
    "pipe_idx = {0: 'Logistic Regression', \n",
    "            1: 'Random Forest', \n",
    "            2: 'Adaboost'\n",
    "           }\n",
    "\n",
    "for idx, pipe in enumerate(pipes):\n",
    "    pipe.fit(X_train, y_train)\n",
    "    print('\\nScore Train/Test: %s' % pipe_idx[idx])\n",
    "    print(pipe.score(X_train, y_train))\n",
    "    print(pipe.score(X_test, y_test))\n",
    "# print('Best params: %s' % pipe.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/IPython/core/interactiveshell.py:2785: DtypeWarning: Columns (5,6) have mixed types. Specify dtype option on import or set low_memory=False.\n",
      "  interactivity=interactivity, compiler=compiler, result=result)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>NPPES Provider Last Name</th>\n",
       "      <th>NPPES Provider First Name</th>\n",
       "      <th>NPPES Provider ZIP Code</th>\n",
       "      <th>NPPES Provider State</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>EKLUND</td>\n",
       "      <td>D.</td>\n",
       "      <td>39208.0</td>\n",
       "      <td>MS</td>\n",
       "      <td>129</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>RIGGLE</td>\n",
       "      <td>ERIN</td>\n",
       "      <td>80120.0</td>\n",
       "      <td>CO</td>\n",
       "      <td>537</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 264 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease   \n",
       "\n",
       "  NPPES Provider Last Name NPPES Provider First Name  NPPES Provider ZIP Code  \\\n",
       "0                   EKLUND                        D.                  39208.0   \n",
       "1                   RIGGLE                      ERIN                  80120.0   \n",
       "\n",
       "  NPPES Provider State  Totalclaimcount      ...        VYTORIN  \\\n",
       "0                   MS              129      ...              0   \n",
       "1                   CO              537      ...              0   \n",
       "\n",
       "   WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  ZOLPIDEM.TARTRATE  \\\n",
       "0                0        0      0                0                  0   \n",
       "1                0       16      0                0                  0   \n",
       "\n",
       "   avg_op_supply  op_longer  Opioid_Prescriber  op_prescriber  \n",
       "0            4.5        1.0                  1            1.0  \n",
       "1            4.5        0.0                  0            0.0  \n",
       "\n",
       "[2 rows x 264 columns]"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## Read Data\n",
    "df=pd.read_csv('./inputs/cleaned_prescriber_info18.csv')\n",
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(25000, 264)\n"
     ]
    }
   ],
   "source": [
    "print(df.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "opioids=pd.read_csv('./inputs/list_opioids.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "name=opioids['Drug Name']\n",
    "import re\n",
    "new_name=name.apply(lambda x:re.sub(\"\\ |-\",\".\",str(x)))\n",
    "columns=df.columns\n",
    "Abandoned_variables = set(columns).intersection(set(new_name))\n",
    "Kept_variable=[]\n",
    "for each in columns:\n",
    "    if each in Abandoned_variables:\n",
    "        pass\n",
    "    else:\n",
    "        Kept_variable.append(each)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(25000, 253)\n"
     ]
    }
   ],
   "source": [
    "df=df[Kept_variable]\n",
    "print(df.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "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>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>NPPES Provider Last Name</th>\n",
       "      <th>NPPES Provider First Name</th>\n",
       "      <th>NPPES Provider ZIP Code</th>\n",
       "      <th>NPPES Provider State</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>EKLUND</td>\n",
       "      <td>D.</td>\n",
       "      <td>39208.0</td>\n",
       "      <td>MS</td>\n",
       "      <td>129</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>RIGGLE</td>\n",
       "      <td>ERIN</td>\n",
       "      <td>80120.0</td>\n",
       "      <td>CO</td>\n",
       "      <td>537</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 253 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease   \n",
       "\n",
       "  NPPES Provider Last Name NPPES Provider First Name  NPPES Provider ZIP Code  \\\n",
       "0                   EKLUND                        D.                  39208.0   \n",
       "1                   RIGGLE                      ERIN                  80120.0   \n",
       "\n",
       "  NPPES Provider State  Totalclaimcount      ...        VYTORIN  \\\n",
       "0                   MS              129      ...              0   \n",
       "1                   CO              537      ...              0   \n",
       "\n",
       "   WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  ZOLPIDEM.TARTRATE  \\\n",
       "0                0        0      0                0                  0   \n",
       "1                0       16      0                0                  0   \n",
       "\n",
       "   avg_op_supply  op_longer  Opioid_Prescriber  op_prescriber  \n",
       "0            4.5        1.0                  1            1.0  \n",
       "1            4.5        0.0                  0            0.0  \n",
       "\n",
       "[2 rows x 253 columns]"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "# df['NPPES Provider ZIP Code'] = df['NPPES Provider ZIP Code'].str.replace(',', '')\n",
    "# df['NPPES Provider ZIP Code'] = df['NPPES Provider ZIP Code'].str.replace(' ', '')\n",
    "df['NPPES Provider ZIP Code'] = df['NPPES Provider ZIP Code'].astype(float)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NPI                                  0\n",
       "Gender                               0\n",
       "State                                0\n",
       "Credentials                         82\n",
       "Specialty                            0\n",
       "NPPES Provider Last Name          2610\n",
       "NPPES Provider First Name         2610\n",
       "NPPES Provider ZIP Code           2610\n",
       "NPPES Provider State                 0\n",
       "Totalclaimcount                      0\n",
       "OpioidClaimCount                     0\n",
       "ACYCLOVIR                            0\n",
       "ADVAIR.DISKUS                        0\n",
       "AGGRENOX                             0\n",
       "ALENDRONATE.SODIUM                   0\n",
       "ALLOPURINOL                          0\n",
       "ALPRAZOLAM                           0\n",
       "AMIODARONE.HCL                       0\n",
       "AMITRIPTYLINE.HCL                    0\n",
       "AMLODIPINE.BESYLATE                  0\n",
       "AMLODIPINE.BESYLATE.BENAZEPRIL       0\n",
       "AMOXICILLIN                          0\n",
       "AMOX.TR.POTASSIUM.CLAVULANATE        0\n",
       "AMPHETAMINE.SALT.COMBO               0\n",
       "ATENOLOL                             0\n",
       "ATORVASTATIN.CALCIUM                 0\n",
       "AVODART                              0\n",
       "AZITHROMYCIN                         0\n",
       "BACLOFEN                             0\n",
       "BD.ULTRA.FINE.PEN.NEEDLE             0\n",
       "                                  ... \n",
       "TERAZOSIN.HCL                        0\n",
       "TIMOLOL.MALEATE                      0\n",
       "TIZANIDINE.HCL                       0\n",
       "TOLTERODINE.TARTRATE.ER              0\n",
       "TOPIRAMATE                           0\n",
       "TOPROL.XL                            0\n",
       "TORSEMIDE                            0\n",
       "TRAVATAN.Z                           0\n",
       "TRAZODONE.HCL                        0\n",
       "TRIAMCINOLONE.ACETONIDE              0\n",
       "TRIAMTERENE.HYDROCHLOROTHIAZID       0\n",
       "VALACYCLOVIR                         0\n",
       "VALSARTAN                            0\n",
       "VALSARTAN.HYDROCHLOROTHIAZIDE        0\n",
       "VENLAFAXINE.HCL                      0\n",
       "VENLAFAXINE.HCL.ER                   0\n",
       "VENTOLIN.HFA                         0\n",
       "VERAPAMIL.ER                         0\n",
       "VESICARE                             0\n",
       "VOLTAREN                             0\n",
       "VYTORIN                              0\n",
       "WARFARIN.SODIUM                      0\n",
       "XARELTO                              0\n",
       "ZETIA                                0\n",
       "ZIPRASIDONE.HCL                      0\n",
       "ZOLPIDEM.TARTRATE                    0\n",
       "avg_op_supply                     2610\n",
       "op_longer                         2610\n",
       "Opioid_Prescriber                    0\n",
       "op_prescriber                     2610\n",
       "Length: 253, dtype: int64"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.isnull().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.fillna(0,inplace = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['NPI', 'Gender', 'State', 'Credentials', 'Specialty',\n",
       "       'NPPES Provider Last Name', 'NPPES Provider First Name',\n",
       "       'NPPES Provider ZIP Code', 'NPPES Provider State', 'Totalclaimcount',\n",
       "       ...\n",
       "       'VYTORIN', 'WARFARIN.SODIUM', 'XARELTO', 'ZETIA', 'ZIPRASIDONE.HCL',\n",
       "       'ZOLPIDEM.TARTRATE', 'avg_op_supply', 'op_longer', 'Opioid_Prescriber',\n",
       "       'op_prescriber'],\n",
       "      dtype='object', length=253)"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "df.drop(['NPPES Provider Last Name', 'NPPES Provider First Name','NPPES Provider ZIP Code',\n",
    "         'NPPES Provider State'], axis =1, inplace = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "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>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>OpioidClaimCount</th>\n",
       "      <th>ACYCLOVIR</th>\n",
       "      <th>ADVAIR.DISKUS</th>\n",
       "      <th>AGGRENOX</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>129</td>\n",
       "      <td>29</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>537</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>61</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 249 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  Totalclaimcount  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist              129   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease              537   \n",
       "\n",
       "   OpioidClaimCount  ACYCLOVIR  ADVAIR.DISKUS  AGGRENOX      ...        \\\n",
       "0                29          0              0         0      ...         \n",
       "1                 0          0             61         0      ...         \n",
       "\n",
       "   VYTORIN  WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  \\\n",
       "0        0                0        0      0                0   \n",
       "1        0                0       16      0                0   \n",
       "\n",
       "   ZOLPIDEM.TARTRATE  avg_op_supply  op_longer  Opioid_Prescriber  \\\n",
       "0                  0            4.5        1.0                  1   \n",
       "1                  0            4.5        0.0                  0   \n",
       "\n",
       "   op_prescriber  \n",
       "0            1.0  \n",
       "1            0.0  \n",
       "\n",
       "[2 rows x 249 columns]"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "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>NPI</th>\n",
       "      <th>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>OpioidClaimCount</th>\n",
       "      <th>ACYCLOVIR</th>\n",
       "      <th>ADVAIR.DISKUS</th>\n",
       "      <th>AGGRENOX</th>\n",
       "      <th>...</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "      <th>avg_op_supply</th>\n",
       "      <th>op_longer</th>\n",
       "      <th>Opioid_Prescriber</th>\n",
       "      <th>op_prescriber</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1003002320</td>\n",
       "      <td>M</td>\n",
       "      <td>MS</td>\n",
       "      <td>D.M.D. P.A.</td>\n",
       "      <td>Dentist</td>\n",
       "      <td>129</td>\n",
       "      <td>29</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1003004771</td>\n",
       "      <td>F</td>\n",
       "      <td>CO</td>\n",
       "      <td>M.D.</td>\n",
       "      <td>Pulmonary Disease</td>\n",
       "      <td>537</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>61</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>16</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 249 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          NPI Gender State  Credentials          Specialty  Totalclaimcount  \\\n",
       "0  1003002320      M    MS  D.M.D. P.A.            Dentist              129   \n",
       "1  1003004771      F    CO         M.D.  Pulmonary Disease              537   \n",
       "\n",
       "   OpioidClaimCount  ACYCLOVIR  ADVAIR.DISKUS  AGGRENOX      ...        \\\n",
       "0                29          0              0         0      ...         \n",
       "1                 0          0             61         0      ...         \n",
       "\n",
       "   VYTORIN  WARFARIN.SODIUM  XARELTO  ZETIA  ZIPRASIDONE.HCL  \\\n",
       "0        0                0        0      0                0   \n",
       "1        0                0       16      0                0   \n",
       "\n",
       "   ZOLPIDEM.TARTRATE  avg_op_supply  op_longer  Opioid_Prescriber  \\\n",
       "0                  0            4.5        1.0                  1   \n",
       "1                  0            4.5        0.0                  0   \n",
       "\n",
       "   op_prescriber  \n",
       "0            1.0  \n",
       "1            0.0  \n",
       "\n",
       "[2 rows x 249 columns]"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(17500, 249)\n",
      "(7500, 249)\n"
     ]
    }
   ],
   "source": [
    "train,test = train_test_split(df,test_size=0.3,random_state=42)\n",
    "print(train.shape)\n",
    "print(test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/ipykernel/__main__.py:3: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  app.launch_new_instance()\n",
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/ipykernel/__main__.py:4: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n"
     ]
    }
   ],
   "source": [
    "Categorical_columns=['Gender','State','Credentials','Specialty']\n",
    "for col in Categorical_columns:\n",
    "    train[col]=pd.factorize(train[col], sort=True)[0]\n",
    "    test[col] =pd.factorize(test[col],sort=True)[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = train['Opioid_Prescriber']\n",
    "\n",
    "X = train.iloc[:,1:245]\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "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>Gender</th>\n",
       "      <th>State</th>\n",
       "      <th>Credentials</th>\n",
       "      <th>Specialty</th>\n",
       "      <th>Totalclaimcount</th>\n",
       "      <th>OpioidClaimCount</th>\n",
       "      <th>ACYCLOVIR</th>\n",
       "      <th>ADVAIR.DISKUS</th>\n",
       "      <th>AGGRENOX</th>\n",
       "      <th>ALENDRONATE.SODIUM</th>\n",
       "      <th>...</th>\n",
       "      <th>VENTOLIN.HFA</th>\n",
       "      <th>VERAPAMIL.ER</th>\n",
       "      <th>VESICARE</th>\n",
       "      <th>VOLTAREN</th>\n",
       "      <th>VYTORIN</th>\n",
       "      <th>WARFARIN.SODIUM</th>\n",
       "      <th>XARELTO</th>\n",
       "      <th>ZETIA</th>\n",
       "      <th>ZIPRASIDONE.HCL</th>\n",
       "      <th>ZOLPIDEM.TARTRATE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>4913</th>\n",
       "      <td>0</td>\n",
       "      <td>18</td>\n",
       "      <td>517</td>\n",
       "      <td>48</td>\n",
       "      <td>153</td>\n",
       "      <td>36</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9338</th>\n",
       "      <td>0</td>\n",
       "      <td>6</td>\n",
       "      <td>284</td>\n",
       "      <td>25</td>\n",
       "      <td>229</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>14</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 244 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      Gender  State  Credentials  Specialty  Totalclaimcount  \\\n",
       "4913       0     18          517         48              153   \n",
       "9338       0      6          284         25              229   \n",
       "\n",
       "      OpioidClaimCount  ACYCLOVIR  ADVAIR.DISKUS  AGGRENOX  \\\n",
       "4913                36          0              0         0   \n",
       "9338                 0          0              0         0   \n",
       "\n",
       "      ALENDRONATE.SODIUM        ...          VENTOLIN.HFA  VERAPAMIL.ER  \\\n",
       "4913                   0        ...                     0             0   \n",
       "9338                   0        ...                     0             0   \n",
       "\n",
       "      VESICARE  VOLTAREN  VYTORIN  WARFARIN.SODIUM  XARELTO  ZETIA  \\\n",
       "4913         0         0        0                0        0      0   \n",
       "9338         0         0        0                0        0     14   \n",
       "\n",
       "      ZIPRASIDONE.HCL  ZOLPIDEM.TARTRATE  \n",
       "4913                0                  0  \n",
       "9338                0                  0  \n",
       "\n",
       "[2 rows x 244 columns]"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.iloc[:,1:245][:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "X =train.iloc[:,1:245]\n",
    "y = train['Opioid_Prescriber']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "rfc=RandomForestClassifier()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "hyperparameters={\n",
    "                'n_estimators': [100],\n",
    "                'max_depth': [30],\n",
    "                'min_samples_split': [30],\n",
    "                'class_weight': ['balanced']\n",
    "}\n",
    "                \n",
    "rfgrid = GridSearchCV(rfc, hyperparameters, cv=3, scoring='roc_auc')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# df[columns] = df.columns.apply(lambda x: float(x.split()[0].replace(',', '')))\n",
    "# df[columns] = df.columns.apply(lambda x: float(x.split()[0].replace(',', '')))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "ss = StandardScaler()\n",
    "X_train = ss.fit_transform(X_train)\n",
    "X_test = ss.transform(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "from sklearn.preprocessing import scale\n",
    "X_test = scale(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "df1 = df[df['Opioid_Prescriber'] == 0]\n",
    "df2 = df[df['Opioid_Prescriber'] == 1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "total 25000\n",
      "Non opiod 11123\n",
      "opiod 13335\n"
     ]
    }
   ],
   "source": [
    "print ('total', len(df))\n",
    "print ('Non opiod', len(df1))\n",
    "print ('opiod', len(df2))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Repeating the above process after checking the existance of Data gap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "y = df.Opioid_Prescriber\n",
    "X = df.drop(['Opioid_Prescriber','Speciality' \"op_longer\",'op_prescriber','avg_op_supply'], axis =1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(13125, 244)"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Verifying shapes are the same \n",
    "\n",
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(13125,)"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Verifying shapes are the same \n",
    "\n",
    "\n",
    "y_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(4375, 244)"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Verifying shapes are the same \n",
    "\n",
    "X_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(4375,)"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Verifying shapes are the same \n",
    "\n",
    "y_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip install tpot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "# TPOT used to optimize and model pipelines# TPOT u \n",
    "\n",
    "from tpot import TPOTClassifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "pipeline_optimizer = TPOTClassifier(generations=2, population_size=20,cv=5, verbosity=2,\n",
    "                      config_dict='TPOT light')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Optimization Progress:  68%|██████▊   | 41/60 [37:08<25:40, 81.08s/pipeline]   "
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generation 1 - Current best internal CV score: 0.8687291181299026\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "                                                                             \r"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generation 2 - Current best internal CV score: 0.8687291181299026\n",
      "\n",
      "Best pipeline: DecisionTreeClassifier(input_matrix, criterion=entropy, max_depth=6, min_samples_leaf=6, min_samples_split=8)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "TPOTClassifier(config_dict={'sklearn.naive_bayes.GaussianNB': {}, 'sklearn.naive_bayes.BernoulliNB': {'alpha': [0.001, 0.01, 0.1, 1.0, 10.0, 100.0], 'fit_prior': [True, False]}, 'sklearn.naive_bayes.MultinomialNB': {'alpha': [0.001, 0.01, 0.1, 1.0, 10.0, 100.0], 'fit_prior': [True, False]}, 'sklearn.tree.DecisionT...e_selection.VarianceThreshold': {'threshold': [0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.2]}},\n",
       "        crossover_rate=0.1, cv=5, disable_update_check=False,\n",
       "        early_stop=None, generations=2, max_eval_time_mins=5,\n",
       "        max_time_mins=None, memory=None, mutation_rate=0.9, n_jobs=1,\n",
       "        offspring_size=20, periodic_checkpoint_folder=None,\n",
       "        population_size=20, random_state=None, scoring=None, subsample=1.0,\n",
       "        verbosity=2, warm_start=False)"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pipeline_optimizer.fit(X_train,y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8637714285714285\n"
     ]
    }
   ],
   "source": [
    "print(pipeline_optimizer.score(X_test, y_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Machine learning library/functions\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.svm import SVC, LinearSVC\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.naive_bayes import GaussianNB\n",
    "np.random.seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [],
   "source": [
    "dt = DecisionTreeClassifier(criterion='gini', max_depth=9, min_samples_leaf=13, min_samples_split=12)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=9,\n",
       "            max_features=None, max_leaf_nodes=None,\n",
       "            min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "            min_samples_leaf=13, min_samples_split=12,\n",
       "            min_weight_fraction_leaf=0.0, presort=False, random_state=None,\n",
       "            splitter='best')"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dt.fit(X_train,y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.881447619047619"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dt.score(X_train,y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8562285714285714"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dt.score(X_test,y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/cross_validation.py:553: Warning: The least populated class in y has only 1 members, which is too few. The minimum number of labels for any class cannot be less than n_folds=5.\n",
      "  % (min_labels, self.n_folds)), Warning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.8532112382876257"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cross_val_score(dt, X_test,y_test,scoring='accuracy',cv=5).mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [],
   "source": [
    "rf = RandomForestClassifier(criterion='gini', max_depth=9, min_samples_leaf=13, min_samples_split=12)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "RandomForestClassifier(bootstrap=True, class_weight=None, criterion='gini',\n",
       "            max_depth=9, max_features='auto', max_leaf_nodes=None,\n",
       "            min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "            min_samples_leaf=13, min_samples_split=12,\n",
       "            min_weight_fraction_leaf=0.0, n_estimators=10, n_jobs=1,\n",
       "            oob_score=False, random_state=None, verbose=0,\n",
       "            warm_start=False)"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rf.fit(X_train,y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8758857142857143"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rf.score(X_train,y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8669714285714286"
      ]
     },
     "execution_count": 78,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rf.score(X_test,y_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# pd.DataFrame(important_features).plot(kind = 'barh', grid=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# pd.DataFrame(important_features).plot(kind = 'barh', grid=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Importing more packages for model evaluation\n",
    "from sklearn.metrics import roc_auc_score\n",
    "from sklearn.metrics import roc_curve, auc\n",
    "from sklearn.metrics import classification_report\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn import metrics\n",
    "from sklearn.multiclass import OneVsRestClassifier\n",
    "from sklearn.preprocessing import label_binarize, LabelBinarizer\n",
    "from sklearn.svm import LinearSVC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = rf.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to allow us to obtain aucroc score for a multi-class y variable\n",
    "\n",
    "def multiclass_roc_auc_score(y_test, y_pred, average=\"macro\"):\n",
    "\n",
    "    lb = LabelBinarizer()\n",
    "    lb.fit(y_test)\n",
    "\n",
    "    y_test = lb.transform(y_test)\n",
    "    y_pred = lb.transform(y_pred)\n",
    "\n",
    "    return roc_auc_score(y_test, y_pred, average=average)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.5122934678929896"
      ]
     },
     "execution_count": 89,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "multiclass_roc_auc_score(y_test,y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8669714285714286"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics.precision_score(y_test, y_pred,average='micro')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/metrics/classification.py:1137: UndefinedMetricWarning: Recall is ill-defined and being set to 0.0 in labels with no true samples.\n",
      "  'recall', 'true', average, warn_for)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.028210924936126643"
      ]
     },
     "execution_count": 91,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "metrics.recall_score(y_test, y_pred, average='macro')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to produce confusion matrix, precision, recall and f1-score# Functio \n",
    "\n",
    "def classifications(y_test, y_pred):\n",
    "    print('Mean accuracy score %s ' % metrics.accuracy_score(y_test, y_pred), '\\n')\n",
    "    print(confusion_matrix(y_test, y_pred), '\\n')\n",
    "    print(classification_report(y_test, y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean accuracy score 0.8669714285714286  \n",
      "\n",
      "[[1727  184    0 ...    0    0    0]\n",
      " [ 299 2066    0 ...    0    0    0]\n",
      " [   6    0    0 ...    0    0    0]\n",
      " ...\n",
      " [   1    0    0 ...    0    0    0]\n",
      " [   0    1    0 ...    0    0    0]\n",
      " [   1    0    0 ...    0    0    0]] \n",
      "\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.82      0.90      0.86      1911\n",
      "          1       0.92      0.87      0.89      2365\n",
      "         11       0.00      0.00      0.00         6\n",
      "         12       0.00      0.00      0.00         4\n",
      "         13       0.00      0.00      0.00         4\n",
      "         14       0.00      0.00      0.00         1\n",
      "         15       0.00      0.00      0.00         4\n",
      "         16       0.00      0.00      0.00         2\n",
      "         17       0.00      0.00      0.00         1\n",
      "         18       0.00      0.00      0.00         3\n",
      "         19       0.00      0.00      0.00         1\n",
      "         20       0.00      0.00      0.00         4\n",
      "         21       0.00      0.00      0.00         1\n",
      "         22       0.00      0.00      0.00         2\n",
      "         23       0.00      0.00      0.00         1\n",
      "         24       0.00      0.00      0.00         3\n",
      "         25       0.00      0.00      0.00         2\n",
      "         26       0.00      0.00      0.00         1\n",
      "         27       0.00      0.00      0.00         2\n",
      "         28       0.00      0.00      0.00         3\n",
      "         29       0.00      0.00      0.00         2\n",
      "         31       0.00      0.00      0.00         1\n",
      "         32       0.00      0.00      0.00         1\n",
      "         33       0.00      0.00      0.00         3\n",
      "         35       0.00      0.00      0.00         1\n",
      "         36       0.00      0.00      0.00         1\n",
      "         37       0.00      0.00      0.00         1\n",
      "         41       0.00      0.00      0.00         1\n",
      "         43       0.00      0.00      0.00         1\n",
      "         44       0.00      0.00      0.00         1\n",
      "         46       0.00      0.00      0.00         2\n",
      "         47       0.00      0.00      0.00         1\n",
      "         48       0.00      0.00      0.00         4\n",
      "         49       0.00      0.00      0.00         1\n",
      "         52       0.00      0.00      0.00         2\n",
      "         55       0.00      0.00      0.00         3\n",
      "         57       0.00      0.00      0.00         1\n",
      "         58       0.00      0.00      0.00         1\n",
      "         60       0.00      0.00      0.00         1\n",
      "         61       0.00      0.00      0.00         2\n",
      "         63       0.00      0.00      0.00         1\n",
      "         64       0.00      0.00      0.00         1\n",
      "         67       0.00      0.00      0.00         1\n",
      "         73       0.00      0.00      0.00         1\n",
      "         75       0.00      0.00      0.00         1\n",
      "         80       0.00      0.00      0.00         2\n",
      "         83       0.00      0.00      0.00         0\n",
      "         85       0.00      0.00      0.00         1\n",
      "         86       0.00      0.00      0.00         1\n",
      "         99       0.00      0.00      0.00         1\n",
      "        100       0.00      0.00      0.00         1\n",
      "        102       0.00      0.00      0.00         1\n",
      "        107       0.00      0.00      0.00         1\n",
      "        109       0.00      0.00      0.00         1\n",
      "        114       0.00      0.00      0.00         1\n",
      "        119       0.00      0.00      0.00         1\n",
      "        124       0.00      0.00      0.00         1\n",
      "        136       0.00      0.00      0.00         1\n",
      "        151       0.00      0.00      0.00         1\n",
      "        199       0.00      0.00      0.00         1\n",
      "        200       0.00      0.00      0.00         1\n",
      "        222       0.00      0.00      0.00         1\n",
      "        272       0.00      0.00      0.00         1\n",
      "\n",
      "avg / total       0.85      0.87      0.86      4375\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/metrics/classification.py:1135: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n",
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/metrics/classification.py:1137: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples.\n",
      "  'recall', 'true', average, warn_for)\n"
     ]
    }
   ],
   "source": [
    "classifications(y_test,y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "cm = confusion_matrix(y_test,y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[1727,  184,    0, ...,    0,    0,    0],\n",
       "       [ 299, 2066,    0, ...,    0,    0,    0],\n",
       "       [   6,    0,    0, ...,    0,    0,    0],\n",
       "       ...,\n",
       "       [   1,    0,    0, ...,    0,    0,    0],\n",
       "       [   0,    1,    0, ...,    0,    0,    0],\n",
       "       [   1,    0,    0, ...,    0,    0,    0]])"
      ]
     },
     "execution_count": 95,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "metadata": {},
   "outputs": [],
   "source": [
    "def roc(probas, true, step=0.01):\n",
    "    \"\"\"\n",
    "    probas should be a numpy array of predict_probas\n",
    "    true is a pandas series of true labels\n",
    "    step is the step size for checking thresholds\n",
    "    \"\"\"\n",
    "    \n",
    "    probas = probas[:,1]  # The output of predict_proba() is an array of the probabilities for every class, but we only want the probabilities for class 1\n",
    "    true = true.values    # We need to convert the class labels from a Pandas Series to a numpy array. We do this using the .values attribute\n",
    "    assert(len(probas) == len(true)) # We're making sure that our probabilities vector is the same length as our true class labesl vector\n",
    "    \n",
    "    TPRs = [] # Setting up empty list of True Positive Rate\n",
    "    FPRs = [] # Setting up empty list of False Positive Rate\n",
    "    \n",
    "    for i in np.arange(0.0,1.0,step): # np.arange allows us to use step sizes that are decimals\n",
    "        preds_class = probas > i # Numpy arrays have a feature called 'broadcasting.' Check the documentation: https://docs.scipy.org/doc/numpy-1.13.0/user/basics.broadcasting.html to see what this does.\n",
    "        TP = 0 \n",
    "        FP = 0\n",
    "        TN = 0\n",
    "        FN = 0\n",
    "        for index in range(len(preds_class)): # We're comparing each prediction with each true value here\n",
    "\n",
    "            if preds_class[index] == 1 and true[index] == 1:\n",
    "                TP += 1\n",
    "            elif preds_class[index] == 1 and true[index] == 0:\n",
    "                FP += 1\n",
    "            elif preds_class[index] == 0 and true[index] == 0:\n",
    "                TN += 1 \n",
    "            elif preds_class[index] == 0 and true[index] == 1:\n",
    "                FN += 1\n",
    "                \n",
    "        TPR = TP/(TP + FN) # Calculating TPR and FPR and appending to our lists\n",
    "        FPR = FP/(FP + TN)\n",
    "        \n",
    "        TPRs.append(TPR)\n",
    "        FPRs.append(FPR)\n",
    "         \n",
    "    plt.rcParams['font.size'] = 14\n",
    "    plt.plot(FPRs, TPRs, color=\"r\")\n",
    "    plt.plot([0,1],[0,1], color='y', lw=2, linestyle='--')\n",
    "    plt.xlim([0.0, 1.0])\n",
    "    plt.ylim([0.0, 1.0])\n",
    "    plt.title('Receiver Operating Characteristic')\n",
    "    plt.xlabel(\"False Positive Rate (1 - Specificity)\")\n",
    "    plt.ylabel(\"True Positive Rate (Sensitivity)\")\n",
    "    plt.show();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Decision Tree Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/model_selection/_split.py:605: Warning: The least populated class in y has only 1 members, which is too few. The minimum number of members in any class cannot be less than n_splits=3.\n",
      "  % (min_groups, self.n_splits)), Warning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.7468190476190476\n",
      "{'max_depth': 7, 'max_features': 5}\n"
     ]
    }
   ],
   "source": [
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "dt_c = DecisionTreeClassifier(max_depth=5)\n",
    "\n",
    "\n",
    "params = {\n",
    "    'max_depth':[1, 2, 3, 5, 7, 10],\n",
    "    'max_features':[1, 2,  3, 5]\n",
    "}\n",
    "\n",
    "\n",
    "gs_dt = GridSearchCV(estimator = dt_c, param_grid = params)\n",
    "gs_dt.fit(X_train, y_train)\n",
    "print(gs_dt.best_score_)\n",
    "print(gs_dt.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['class_weight', 'criterion', 'max_depth', 'max_features', 'max_leaf_nodes', 'min_impurity_decrease', 'min_impurity_split', 'min_samples_leaf', 'min_samples_split', 'min_weight_fraction_leaf', 'presort', 'random_state', 'splitter'])"
      ]
     },
     "execution_count": 98,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dt_c.get_params().keys()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Random Forest Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/model_selection/_split.py:605: Warning: The least populated class in y has only 1 members, which is too few. The minimum number of members in any class cannot be less than n_splits=3.\n",
      "  % (min_groups, self.n_splits)), Warning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.856\n",
      "{'max_depth': 4, 'max_features': 7, 'n_estimators': 7}\n"
     ]
    }
   ],
   "source": [
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "rf_c = RandomForestClassifier()\n",
    "\n",
    "\n",
    "params = {\n",
    "    'n_estimators': [3,5,7,10], \n",
    "    'max_features': [3, 5, 7],\n",
    "    'max_depth': [1, 2 , 3, 4]\n",
    "}\n",
    "gs_rf = GridSearchCV(rf_c, param_grid=params)\n",
    "gs_rf.fit(X_train, y_train)\n",
    "print(gs_rf.best_score_)\n",
    "print(gs_rf.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['bootstrap', 'class_weight', 'criterion', 'max_depth', 'max_features', 'max_leaf_nodes', 'min_impurity_decrease', 'min_impurity_split', 'min_samples_leaf', 'min_samples_split', 'min_weight_fraction_leaf', 'n_estimators', 'n_jobs', 'oob_score', 'random_state', 'verbose', 'warm_start'])"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rf_c.get_params().keys()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### AdaBoost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/sklearn/model_selection/_split.py:605: Warning: The least populated class in y has only 1 members, which is too few. The minimum number of members in any class cannot be less than n_splits=3.\n",
      "  % (min_groups, self.n_splits)), Warning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8552380952380952\n",
      "{'n_estimators': 6}\n"
     ]
    }
   ],
   "source": [
    "from sklearn.ensemble import AdaBoostClassifier\n",
    "ada_r = AdaBoostClassifier()\n",
    "\n",
    "params = {\n",
    "    'n_estimators':range(1,7)\n",
    "}\n",
    "gs_ada = GridSearchCV(ada_r, param_grid = params)\n",
    "gs_ada.fit(X_train, y_train)\n",
    "print(gs_ada.best_score_)\n",
    "print(gs_ada.best_params_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dict_keys(['cv', 'error_score', 'estimator__algorithm', 'estimator__base_estimator', 'estimator__learning_rate', 'estimator__n_estimators', 'estimator__random_state', 'estimator', 'fit_params', 'iid', 'n_jobs', 'param_grid', 'pre_dispatch', 'refit', 'return_train_score', 'scoring', 'verbose'])"
      ]
     },
     "execution_count": 103,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gs_ada.get_params().keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install tensorflow"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### CNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {},
   "outputs": [],
   "source": [
    "ss = StandardScaler()\n",
    "X_train = ss.fit_transform(X_train)\n",
    "X_test = ss.transform(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 13125 samples, validate on 4375 samples\n",
      "Epoch 1/30\n",
      "13125/13125 [==============================] - 3s 265us/step - loss: -15.2629 - acc: 0.6035 - val_loss: -16.4067 - val_acc: 0.6473\n",
      "Epoch 2/30\n",
      "13125/13125 [==============================] - 2s 128us/step - loss: -16.8374 - acc: 0.6841 - val_loss: -16.3960 - val_acc: 0.6891\n",
      "Epoch 3/30\n",
      "13125/13125 [==============================] - 2s 126us/step - loss: -16.9694 - acc: 0.7259 - val_loss: -16.5211 - val_acc: 0.7161\n",
      "Epoch 4/30\n",
      "13125/13125 [==============================] - 2s 134us/step - loss: -17.0732 - acc: 0.7387 - val_loss: -16.3704 - val_acc: 0.7488\n",
      "Epoch 5/30\n",
      "13125/13125 [==============================] - 2s 125us/step - loss: -17.1284 - acc: 0.7622 - val_loss: -16.3839 - val_acc: 0.7166\n",
      "Epoch 6/30\n",
      "13125/13125 [==============================] - 2s 135us/step - loss: -17.0763 - acc: 0.7595 - val_loss: -16.3532 - val_acc: 0.7456\n",
      "Epoch 7/30\n",
      "13125/13125 [==============================] - 2s 128us/step - loss: -17.1946 - acc: 0.7783 - val_loss: -16.3263 - val_acc: 0.7422\n",
      "Epoch 8/30\n",
      "13125/13125 [==============================] - 2s 125us/step - loss: -17.2266 - acc: 0.7855 - val_loss: -16.5380 - val_acc: 0.7447\n",
      "Epoch 9/30\n",
      "13125/13125 [==============================] - 2s 131us/step - loss: -17.2201 - acc: 0.7879 - val_loss: -16.6140 - val_acc: 0.7781\n",
      "Epoch 10/30\n",
      "13125/13125 [==============================] - 2s 126us/step - loss: -17.2562 - acc: 0.7928 - val_loss: -16.6024 - val_acc: 0.7838\n",
      "Epoch 11/30\n",
      "13125/13125 [==============================] - 2s 120us/step - loss: -17.2869 - acc: 0.8029 - val_loss: -16.6419 - val_acc: 0.7831\n",
      "Epoch 12/30\n",
      "13125/13125 [==============================] - 2s 139us/step - loss: -17.3267 - acc: 0.8093 - val_loss: -16.3867 - val_acc: 0.7806\n",
      "Epoch 13/30\n",
      "13125/13125 [==============================] - 2s 129us/step - loss: -17.3513 - acc: 0.8146 - val_loss: -16.2958 - val_acc: 0.7815\n",
      "Epoch 14/30\n",
      "13125/13125 [==============================] - 2s 129us/step - loss: -17.3727 - acc: 0.8146 - val_loss: -16.5178 - val_acc: 0.7584\n",
      "Epoch 15/30\n",
      "13125/13125 [==============================] - 2s 127us/step - loss: -17.3562 - acc: 0.8153 - val_loss: -16.3608 - val_acc: 0.6686\n",
      "Epoch 16/30\n",
      "13125/13125 [==============================] - 2s 135us/step - loss: -17.3671 - acc: 0.8167 - val_loss: -16.4392 - val_acc: 0.7927\n",
      "Epoch 17/30\n",
      "13125/13125 [==============================] - 2s 139us/step - loss: -17.2646 - acc: 0.8082 - val_loss: -16.4812 - val_acc: 0.7915\n",
      "Epoch 18/30\n",
      "13125/13125 [==============================] - 2s 122us/step - loss: -17.3674 - acc: 0.8210 - val_loss: -16.4678 - val_acc: 0.7893\n",
      "Epoch 19/30\n",
      "13125/13125 [==============================] - 2s 128us/step - loss: -17.4255 - acc: 0.8325 - val_loss: -16.4440 - val_acc: 0.8041\n",
      "Epoch 20/30\n",
      "13125/13125 [==============================] - 2s 129us/step - loss: -17.4305 - acc: 0.8314 - val_loss: -16.3358 - val_acc: 0.7849\n",
      "Epoch 21/30\n",
      "13125/13125 [==============================] - 2s 123us/step - loss: -17.4488 - acc: 0.8375 - val_loss: -16.2968 - val_acc: 0.8080\n",
      "Epoch 22/30\n",
      "13125/13125 [==============================] - 2s 142us/step - loss: -17.4535 - acc: 0.8398 - val_loss: -16.2952 - val_acc: 0.8046\n",
      "Epoch 23/30\n",
      "13125/13125 [==============================] - 2s 128us/step - loss: -17.4587 - acc: 0.8427 - val_loss: -16.2697 - val_acc: 0.7920\n",
      "Epoch 24/30\n",
      "13125/13125 [==============================] - 2s 142us/step - loss: -17.4748 - acc: 0.8434 - val_loss: -16.2885 - val_acc: 0.8103\n",
      "Epoch 25/30\n",
      "13125/13125 [==============================] - 2s 121us/step - loss: -17.4788 - acc: 0.8451 - val_loss: -16.3645 - val_acc: 0.7970\n",
      "Epoch 26/30\n",
      "13125/13125 [==============================] - 2s 123us/step - loss: -17.4662 - acc: 0.8465 - val_loss: -16.2410 - val_acc: 0.8080\n",
      "Epoch 27/30\n",
      "13125/13125 [==============================] - 2s 127us/step - loss: -17.4973 - acc: 0.8474 - val_loss: -16.2572 - val_acc: 0.8103\n",
      "Epoch 28/30\n",
      "13125/13125 [==============================] - 2s 136us/step - loss: -17.4675 - acc: 0.8488 - val_loss: -16.3507 - val_acc: 0.7440\n",
      "Epoch 29/30\n",
      "13125/13125 [==============================] - 2s 142us/step - loss: -17.5026 - acc: 0.8526 - val_loss: -16.2108 - val_acc: 0.8053\n",
      "Epoch 30/30\n",
      "13125/13125 [==============================] - 2s 123us/step - loss: -17.5276 - acc: 0.8587 - val_loss: -16.1949 - val_acc: 0.8053\n"
     ]
    }
   ],
   "source": [
    "from keras import regularizers\n",
    "import numpy as np\n",
    "from keras.layers import Dropout\n",
    "from keras.callbacks import EarlyStopping\n",
    "\n",
    "np.random.seed(42)\n",
    "\n",
    "model= Sequential()\n",
    "\n",
    "n_input = X_train.shape[1]\n",
    "n_hidden = n_input\n",
    "n_output = 1\n",
    "#first layer defines the shape of the input which is X_train.shape\n",
    "#then add a layer defining the activation function - relu is on of the most popular\n",
    "#activation - should the neuron be fired or not - relu is quite efficient.\n",
    "model.add(Dense(n_hidden, input_dim=n_input, activation='relu'))\n",
    "#A dense layer is a classic fully connected neural network layer : each input node is connected to each output node.\n",
    "#for this layer use the sigmoid activation function\n",
    "#sigmoid will assign a propability to each tag (total output = 1) since its a classification problem\n",
    "model.add(Dense(n_output, activation='sigmoid'))\n",
    "\n",
    "#now we compile the model\n",
    "#optimizer is the function used to search for the right weights\n",
    "#here we are using Adaptive Moment Estimation (Adam) that also uses adaptive learning rates\n",
    "#The loss function, also called the objective function is the evaluation of the model used by the optimizer \n",
    "# to navigate the weight space; only accuracy is supported at the moment for metrics.\n",
    "model.compile(loss='binary_crossentropy', optimizer='adam', \n",
    "                metrics=['acc'])\n",
    "\n",
    "#early_stop = EarlyStopping(monitor='val_loss', min_delta=0) \n",
    "\n",
    "\n",
    "history = model.fit(X_train, y_train, validation_data=(X_test, y_test),\n",
    "                      epochs=30, batch_size=None)\n",
    "                     # callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x1a447060b8>"
      ]
     },
     "execution_count": 108,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/kirosgebremariam/anaconda3/envs/dsi/lib/python3.6/site-packages/matplotlib/figure.py:2267: UserWarning: This figure includes Axes that are not compatible with tight_layout, so results might be incorrect.\n",
      "  warnings.warn(\"This figure includes Axes that are not compatible \"\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "image/png": {
       "height": 280,
       "width": 424
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_loss = history.history['loss']\n",
    "test_loss = history.history['val_loss']\n",
    "plt.plot(train_loss, label='Training loss')\n",
    "plt.plot(test_loss, label='Testing loss')\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.8586666666666667, 0.8052571428162711)"
      ]
     },
     "execution_count": 109,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "history.history['acc'][-1], history.history['val_acc'][-1]"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:dsi]",
   "language": "python",
   "name": "conda-env-dsi-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}