01_data_import.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "4c4452b4-2520-4748-bd6d-2ae484e70b16",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Modify the file 00_setup.py to define input/output file paths on your system\n",
    "# The information in 00_setup.py will be used across notebooks\n",
    "from importlib.machinery import SourceFileLoader\n",
    "setup = SourceFileLoader(\"setup\", \"./00_setup.py\").load_module()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a595d38b-794b-4317-b39f-58da027df783",
   "metadata": {},
   "source": [
    "# Data Import\n",
    " * Import data\n",
    " * Feature processing\n",
    " * Train-test split and mean encoding"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f5cfeb88-9b8e-44e6-91ba-61b54b6fca52",
   "metadata": {},
   "source": [
    "I do not do extensive data exploration but instead use others' insights from various\n",
    "Kaggle projects, especially:\n",
    "* https://www.kaggle.com/code/ricardocolindres/loan-default-prediction-loan-parameter-optimizer"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3a01a85e-5401-4703-a665-361934e75105",
   "metadata": {},
   "source": [
    "I am also not going to use all the features I could.  I want to focus on NAICS, and so I don't do a lot with bank, franchise, or geography."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f3513d1f-ae7b-4bbd-be2f-4bd430d1a318",
   "metadata": {},
   "source": [
    "Because I am focused on NAICS I am going to drop rows with missing NAICS"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6e6da1d7-27bc-40dc-bf59-4062c143c508",
   "metadata": {},
   "source": [
    "*This script takes about 1 minute on my MacBook Air*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "5f2f4da8-290e-4287-ba3c-eef1dfb093f8",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import re, pickle\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "71df4185-bc4e-404b-a2df-6e58259b2381",
   "metadata": {},
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import OneHotEncoder\n",
    "from category_encoders import CountEncoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "6b6f505e-bc8c-41a7-9585-36250a64ec46",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sba_gnn.sba_gnn import sg_plot, sg_blender\n",
    "from sba_gnn.sba_gnn.sg_blender import HierarchicalEncoder # Target encoder"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "55ec3482-660d-4f05-9f80-2086668d6768",
   "metadata": {},
   "source": [
    "## Import Data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "011691b5-499f-4028-9256-0983494975b7",
   "metadata": {},
   "source": [
    "##### Loans Data - See README.md"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "92bd280f-377e-4e47-8f83-331c8eca3768",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/h9/4b0yxscx221gvjqy2cqp_kyh0000gn/T/ipykernel_20792/3842615021.py:1: UserWarning: Could not infer format, so each element will be parsed individually, falling back to `dateutil`. To ensure parsing is consistent and as-expected, please specify a format.\n",
      "  sba_loans = pd.read_csv(setup.input_path,  dtype=setup.input_dtypes, parse_dates = setup.input_dates,\n",
      "/var/folders/h9/4b0yxscx221gvjqy2cqp_kyh0000gn/T/ipykernel_20792/3842615021.py:1: UserWarning: Could not infer format, so each element will be parsed individually, falling back to `dateutil`. To ensure parsing is consistent and as-expected, please specify a format.\n",
      "  sba_loans = pd.read_csv(setup.input_path,  dtype=setup.input_dtypes, parse_dates = setup.input_dates,\n",
      "/var/folders/h9/4b0yxscx221gvjqy2cqp_kyh0000gn/T/ipykernel_20792/3842615021.py:1: UserWarning: Could not infer format, so each element will be parsed individually, falling back to `dateutil`. To ensure parsing is consistent and as-expected, please specify a format.\n",
      "  sba_loans = pd.read_csv(setup.input_path,  dtype=setup.input_dtypes, parse_dates = setup.input_dates,\n"
     ]
    }
   ],
   "source": [
    "sba_loans = pd.read_csv(setup.input_path,  dtype=setup.input_dtypes, parse_dates = setup.input_dates,\n",
    "                        converters = setup.input_converters)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "05161011-6942-46e0-a123-7643d0e93e70",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LoanNr_ChkDgt                object\n",
       "Name                         object\n",
       "City                         object\n",
       "State                        object\n",
       "Zip                          object\n",
       "Bank                         object\n",
       "BankState                    object\n",
       "NAICS                        object\n",
       "ApprovalDate         datetime64[ns]\n",
       "ApprovalFY                   object\n",
       "Term                          int64\n",
       "NoEmp                         int64\n",
       "NewExist                     object\n",
       "CreateJob                     int64\n",
       "RetainedJob                   int64\n",
       "FranchiseCode                object\n",
       "UrbanRural                   object\n",
       "RevLineCr                    object\n",
       "LowDoc                       object\n",
       "ChgOffDate           datetime64[ns]\n",
       "DisbursementDate     datetime64[ns]\n",
       "DisbursementGross           float64\n",
       "BalanceGross                float64\n",
       "MIS_Status                   object\n",
       "ChgOffPrinGr                float64\n",
       "GrAppv                      float64\n",
       "SBA_Appv                    float64\n",
       "dtype: object"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans.dtypes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "c42036a7-4923-47df-bc12-e69c13bcefa7",
   "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>ApprovalDate</th>\n",
       "      <th>ChgOffDate</th>\n",
       "      <th>DisbursementDate</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>899164</td>\n",
       "      <td>162699</td>\n",
       "      <td>896796</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>2001-06-04 11:57:49.414700416</td>\n",
       "      <td>2008-07-02 12:56:30.968597760</td>\n",
       "      <td>2001-09-24 03:38:57.328891136</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>1974-01-09 00:00:00</td>\n",
       "      <td>1988-10-03 00:00:00</td>\n",
       "      <td>1974-01-07 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>1996-11-22 00:00:00</td>\n",
       "      <td>2007-09-04 00:00:00</td>\n",
       "      <td>1997-05-31 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>2002-09-27 00:00:00</td>\n",
       "      <td>2009-12-04 00:00:00</td>\n",
       "      <td>2002-12-31 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>2006-01-20 00:00:00</td>\n",
       "      <td>2011-03-16 00:00:00</td>\n",
       "      <td>2006-03-31 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>2073-12-20 00:00:00</td>\n",
       "      <td>2026-10-22 00:00:00</td>\n",
       "      <td>2073-12-24 00:00:00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        ApprovalDate                     ChgOffDate  \\\n",
       "count                         899164                         162699   \n",
       "mean   2001-06-04 11:57:49.414700416  2008-07-02 12:56:30.968597760   \n",
       "min              1974-01-09 00:00:00            1988-10-03 00:00:00   \n",
       "25%              1996-11-22 00:00:00            2007-09-04 00:00:00   \n",
       "50%              2002-09-27 00:00:00            2009-12-04 00:00:00   \n",
       "75%              2006-01-20 00:00:00            2011-03-16 00:00:00   \n",
       "max              2073-12-20 00:00:00            2026-10-22 00:00:00   \n",
       "\n",
       "                    DisbursementDate  \n",
       "count                         896796  \n",
       "mean   2001-09-24 03:38:57.328891136  \n",
       "min              1974-01-07 00:00:00  \n",
       "25%              1997-05-31 00:00:00  \n",
       "50%              2002-12-31 00:00:00  \n",
       "75%              2006-03-31 00:00:00  \n",
       "max              2073-12-24 00:00:00  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans.select_dtypes(np.datetime64).describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0a808c1c-a3ba-47ea-83d0-3a9effb6d632",
   "metadata": {},
   "source": [
    "##### NAICS map"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "64d438ee-0337-436d-9c24-7b6b384b2d3b",
   "metadata": {},
   "outputs": [],
   "source": [
    "naics_map = pd.read_csv('naics/naics_map.csv', dtype='str', encoding='latin1' ) \\\n",
    "    .rename(columns={'Sector': 'NAICS_sector',\n",
    "                     'Description': 'NAICS_sector_desc'})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "e5bb4b30-0de5-4454-a189-295d0797970e",
   "metadata": {},
   "outputs": [],
   "source": [
    "naics_map[['NAICS_sector', 'NAICS_sector_desc']] = naics_map[['NAICS_sector', 'NAICS_sector_desc']] \\\n",
    "    .apply(lambda x: x.astype('category'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "263feb13-868a-4aeb-b0c1-9ce671681c77",
   "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>NAICS_sector</th>\n",
       "      <th>NAICS_2</th>\n",
       "      <th>NAICS_sector_desc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>11</td>\n",
       "      <td>11</td>\n",
       "      <td>Agriculture, Forestry, Fishing and Hunting</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>21</td>\n",
       "      <td>21</td>\n",
       "      <td>Mining, Quarrying, and Oil and Gas Extraction</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>22</td>\n",
       "      <td>22</td>\n",
       "      <td>Utilities</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>23</td>\n",
       "      <td>23</td>\n",
       "      <td>Construction</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>31-33</td>\n",
       "      <td>31</td>\n",
       "      <td>Manufacturing</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  NAICS_sector NAICS_2                              NAICS_sector_desc\n",
       "0           11      11     Agriculture, Forestry, Fishing and Hunting\n",
       "1           21      21  Mining, Quarrying, and Oil and Gas Extraction\n",
       "2           22      22                                      Utilities\n",
       "3           23      23                                   Construction\n",
       "4        31-33      31                                  Manufacturing"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "naics_map.head(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9e9ff1b1-14e8-4a29-8e49-0c1ea957a556",
   "metadata": {},
   "source": [
    "##### All NAICS Descriptions (2007)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "8f59f87c-6504-4f80-8bfc-27b2b1ac6a2f",
   "metadata": {},
   "outputs": [],
   "source": [
    "naics_desc = pd.read_csv('naics/naics_2007.csv', dtype='str', encoding='latin1' ) \\\n",
    "    .rename(columns={'INDEX ITEM DESCRIPTION': 'NAICS07_desc'})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "99e3ed44-ad92-4ca0-87bf-8035153c65a8",
   "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>NAICS07</th>\n",
       "      <th>NAICS07_desc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>18982</th>\n",
       "      <td>811310</td>\n",
       "      <td>Reconditioning shipping barrels and drums</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18499</th>\n",
       "      <td>712130</td>\n",
       "      <td>Conservatories, botanical</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5464</th>\n",
       "      <td>322221</td>\n",
       "      <td>Coating purchased papers for packaging applica...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2583</th>\n",
       "      <td>311320</td>\n",
       "      <td>Syrup, chocolate, made from cacao beans</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      NAICS07                                       NAICS07_desc\n",
       "18982  811310          Reconditioning shipping barrels and drums\n",
       "18499  712130                          Conservatories, botanical\n",
       "5464   322221  Coating purchased papers for packaging applica...\n",
       "2583   311320            Syrup, chocolate, made from cacao beans"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "naics_desc.sample(4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "d9a27425-5c8a-4b5a-bc63-442a1cacbc7a",
   "metadata": {},
   "outputs": [],
   "source": [
    "naics_desc.to_parquet(Path(setup.temp_path).joinpath('01_DATA_naics07_descriptions.parquet'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0d66da81-4a2a-4fb1-a7cc-21ee365fbe94",
   "metadata": {},
   "source": [
    "## Process Target"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "148af406-beb1-41b9-91ef-4286d1ed0c97",
   "metadata": {},
   "source": [
    "##### Create the target feature (0/1) that indicates a default"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "65d46709-9ede-4999-9f10-3f06b4a8caf9",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MIS_Status\n",
       "P I F     739609\n",
       "CHGOFF    157558\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['MIS_Status'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "e2dd041e-e092-4487-a1ae-de765e710a4b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MIS_Status\n",
       "P I F     0.822552\n",
       "CHGOFF    0.175227\n",
       "NaN       0.002221\n",
       "Name: proportion, dtype: float64"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['MIS_Status'].value_counts(normalize=True, dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "d5866b7b-654a-427d-8996-688ba3a9d3e3",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Defaults are fairly common.  About 1/6 of loans default"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "0b7d3959-d712-4585-a6f3-6cd575a27953",
   "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>LoanNr_ChkDgt</th>\n",
       "      <th>Name</th>\n",
       "      <th>City</th>\n",
       "      <th>State</th>\n",
       "      <th>Zip</th>\n",
       "      <th>Bank</th>\n",
       "      <th>BankState</th>\n",
       "      <th>NAICS</th>\n",
       "      <th>ApprovalDate</th>\n",
       "      <th>ApprovalFY</th>\n",
       "      <th>...</th>\n",
       "      <th>RevLineCr</th>\n",
       "      <th>LowDoc</th>\n",
       "      <th>ChgOffDate</th>\n",
       "      <th>DisbursementDate</th>\n",
       "      <th>DisbursementGross</th>\n",
       "      <th>BalanceGross</th>\n",
       "      <th>MIS_Status</th>\n",
       "      <th>ChgOffPrinGr</th>\n",
       "      <th>GrAppv</th>\n",
       "      <th>SBA_Appv</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>223939</th>\n",
       "      <td>2583585001</td>\n",
       "      <td>A KNIGHT'S DREAM</td>\n",
       "      <td>ERIE</td>\n",
       "      <td>PA</td>\n",
       "      <td>16506</td>\n",
       "      <td>CITIZENS BANK NATL ASSOC</td>\n",
       "      <td>RI</td>\n",
       "      <td>452990</td>\n",
       "      <td>2007-07-19</td>\n",
       "      <td>2007</td>\n",
       "      <td>...</td>\n",
       "      <td>Y</td>\n",
       "      <td>N</td>\n",
       "      <td>2008-07-08</td>\n",
       "      <td>2008-01-31</td>\n",
       "      <td>300.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>276.0</td>\n",
       "      <td>300.0</td>\n",
       "      <td>150.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>53650</th>\n",
       "      <td>1411185009</td>\n",
       "      <td>Law Office of Cynthia A. Strom</td>\n",
       "      <td>ASHEVILLE</td>\n",
       "      <td>NC</td>\n",
       "      <td>28803</td>\n",
       "      <td>BRANCH BK. &amp; TR CO</td>\n",
       "      <td>NC</td>\n",
       "      <td>922130</td>\n",
       "      <td>2005-08-12</td>\n",
       "      <td>2005</td>\n",
       "      <td>...</td>\n",
       "      <td>N</td>\n",
       "      <td>N</td>\n",
       "      <td>NaT</td>\n",
       "      <td>2006-01-31</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.0</td>\n",
       "      <td>50000.0</td>\n",
       "      <td>25000.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>195071</th>\n",
       "      <td>2377415006</td>\n",
       "      <td>MAC-MATIC, INC.</td>\n",
       "      <td>PAINESVILLE</td>\n",
       "      <td>OH</td>\n",
       "      <td>44077</td>\n",
       "      <td>CITIZENS BANK NATL ASSOC</td>\n",
       "      <td>RI</td>\n",
       "      <td>332710</td>\n",
       "      <td>2007-04-05</td>\n",
       "      <td>2007</td>\n",
       "      <td>...</td>\n",
       "      <td>Y</td>\n",
       "      <td>N</td>\n",
       "      <td>NaT</td>\n",
       "      <td>2007-05-31</td>\n",
       "      <td>2040.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2000.0</td>\n",
       "      <td>1000.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3 rows × 27 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       LoanNr_ChkDgt                            Name         City State  \\\n",
       "223939    2583585001                A KNIGHT'S DREAM         ERIE    PA   \n",
       "53650     1411185009  Law Office of Cynthia A. Strom    ASHEVILLE    NC   \n",
       "195071    2377415006                 MAC-MATIC, INC.  PAINESVILLE    OH   \n",
       "\n",
       "          Zip                      Bank BankState   NAICS ApprovalDate  \\\n",
       "223939  16506  CITIZENS BANK NATL ASSOC        RI  452990   2007-07-19   \n",
       "53650   28803        BRANCH BK. & TR CO        NC  922130   2005-08-12   \n",
       "195071  44077  CITIZENS BANK NATL ASSOC        RI  332710   2007-04-05   \n",
       "\n",
       "       ApprovalFY  ...  RevLineCr  LowDoc ChgOffDate  DisbursementDate  \\\n",
       "223939       2007  ...          Y       N 2008-07-08        2008-01-31   \n",
       "53650        2005  ...          N       N        NaT        2006-01-31   \n",
       "195071       2007  ...          Y       N        NaT        2007-05-31   \n",
       "\n",
       "        DisbursementGross BalanceGross MIS_Status ChgOffPrinGr   GrAppv  \\\n",
       "223939              300.0          0.0        NaN        276.0    300.0   \n",
       "53650                 0.0          0.0        NaN          0.0  50000.0   \n",
       "195071             2040.0          0.0        NaN          0.0   2000.0   \n",
       "\n",
       "       SBA_Appv  \n",
       "223939    150.0  \n",
       "53650   25000.0  \n",
       "195071   1000.0  \n",
       "\n",
       "[3 rows x 27 columns]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check the nulls\n",
    "sba_loans[sba_loans['MIS_Status'].isna()].sample(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "095ed3a8-9e9e-403d-873d-0358934bb1df",
   "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>ChgOffPrinGr</th>\n",
       "      <th>BalanceGross</th>\n",
       "      <th>ChgOffDate</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>1997.000000</td>\n",
       "      <td>1997.0</td>\n",
       "      <td>261</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>2758.911367</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2003-03-19 06:15:10.344827648</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1988-12-12 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1992-09-15 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2008-05-05 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2010-02-20 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>650543.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2014-08-27 00:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>27531.777037</td>\n",
       "      <td>0.0</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        ChgOffPrinGr  BalanceGross                     ChgOffDate\n",
       "count    1997.000000        1997.0                            261\n",
       "mean     2758.911367           0.0  2003-03-19 06:15:10.344827648\n",
       "min         0.000000           0.0            1988-12-12 00:00:00\n",
       "25%         0.000000           0.0            1992-09-15 00:00:00\n",
       "50%         0.000000           0.0            2008-05-05 00:00:00\n",
       "75%         0.000000           0.0            2010-02-20 00:00:00\n",
       "max    650543.000000           0.0            2014-08-27 00:00:00\n",
       "std     27531.777037           0.0                            NaN"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans[sba_loans['MIS_Status'].isna()][['ChgOffPrinGr', 'BalanceGross', 'ChgOffDate']].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "664a22d7-7198-4366-8334-5019c9c8a636",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ChgOffDate\n",
       "True     1736\n",
       "False     261\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans[sba_loans['MIS_Status'].isna()]['ChgOffDate'].isna().value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "9b90d38b-21f6-449d-8b05-bf529aada245",
   "metadata": {},
   "outputs": [],
   "source": [
    "# The loans with unknown MIS Status are a small count, most don't have charge-off dates but some do\n",
    "# Drop these rows as they are ambiguous and small volume\n",
    "sba_loans = sba_loans[~sba_loans['MIS_Status'].isna()]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "bafa5a82-2ef2-4e89-9f64-2c5c3fe31e2c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create a binary feature for the remaining loans\n",
    "sba_loans['target'] = np.where(sba_loans['MIS_Status'] == 'P I F', 0, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "42677d14-e9ae-45dc-82c8-6e3e0862fcce",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "target\n",
       "0    0.824383\n",
       "1    0.175617\n",
       "Name: proportion, dtype: float64"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['target'].value_counts(dropna=False, normalize=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b18d6996-5869-4d1e-95f6-31296b844ba5",
   "metadata": {},
   "source": [
    "## Predictor Features - Basic Processing"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bcc718e2-dea0-45f1-909c-5b9c71dccd69",
   "metadata": {},
   "source": [
    "##### Convert approval year to a date"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "31e56ed8-1f26-4962-9153-e98d3aad1576",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    897167.000000\n",
       "mean       2001.139677\n",
       "std           5.912461\n",
       "min        1966.000000\n",
       "25%        1997.000000\n",
       "50%        2002.000000\n",
       "75%        2006.000000\n",
       "max        2014.000000\n",
       "Name: ApprovalFY, dtype: float64"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['ApprovalFY'] = sba_loans['ApprovalFY'].str[0:4].astype('int')\n",
    "sba_loans['ApprovalFY'].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "47d5b3d3-d1ae-45b5-a712-187a1ecc7065",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "ApprovalFY\n",
       "False    897167\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['ApprovalFY'].isna().value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7335c704-c6b5-4532-8bf1-a6e4095be493",
   "metadata": {},
   "source": [
    "##### Express SBA guaranteed amount as a fraction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "dd41e8b8-ec3e-44be-bfc0-405e12a7079e",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans['guaranteed_fract'] = sba_loans['SBA_Appv']/sba_loans['GrAppv']"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "52a5471d-3bf5-4a55-ae1e-23e7b2445dda",
   "metadata": {},
   "source": [
    "##### Convert Y/N binary columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "26147177-a6c4-4859-bf9f-604a454e2e55",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to convert 'Y','N' columns\n",
    "def convert_yn(ser):\n",
    "    cond = [ser.isna(), ser.str.lower() == 'y', ser.str.lower() == 'n']\n",
    "    choice =[np.nan, 1, 0]\n",
    "    return np.select(cond, choice, np.nan)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "0d4a6190-e8b9-4fe3-98ab-03e8b76b2814",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LowDoc\n",
       "N      780997\n",
       "Y      110171\n",
       "NaN      2578\n",
       "0        1490\n",
       "C         758\n",
       "S         603\n",
       "A         495\n",
       "R          74\n",
       "1           1\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['LowDoc'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "75612cc1-84c8-4b71-acc7-924034569b02",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "RevLineCr\n",
       "N    419252\n",
       "0    257431\n",
       "Y    200660\n",
       "T     15239\n",
       "1        23\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['RevLineCr'].value_counts().head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "889d00f8-ae5a-48e9-b8e0-fcfa642e2e03",
   "metadata": {},
   "outputs": [],
   "source": [
    "# There are some strange values in both fields, but I will follow the documentation and assume that the Y/N mean true or false.\n",
    "# I will let the other values become NA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "9a48efb6-c62c-428c-805c-5c207526dd0d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "RevLineCr\n",
       "0.0    419252\n",
       "NaN    277255\n",
       "1.0    200660\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['RevLineCr'] = convert_yn(sba_loans['RevLineCr'])\n",
    "sba_loans['RevLineCr'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "84533b48-e087-40c0-b6d6-d5e99db3a7ca",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "LowDoc\n",
       "0.0    780997\n",
       "1.0    110171\n",
       "NaN      5999\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['LowDoc'] = convert_yn(sba_loans['LowDoc'])\n",
    "sba_loans['LowDoc'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ea03e207-8ada-4a77-a1c6-5b3aec4f8d43",
   "metadata": {},
   "source": [
    "##### New/existing business"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "2252b3f1-e7c9-4b25-8315-fb1500fe8908",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NewExist\n",
       "1      643446\n",
       "2      252559\n",
       "0        1028\n",
       "NaN       134\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['NewExist'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "b2371a5d-5458-4271-8a0c-2f2f41314a42",
   "metadata": {},
   "outputs": [],
   "source": [
    "cond = [sba_loans['NewExist'].isna(), sba_loans['NewExist'] == '1', sba_loans['NewExist'] == '2']\n",
    "choice = [np.nan, 0, 1]\n",
    "sba_loans['new_business'] = np.select(cond, choice, np.nan)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "3ce2ce93-4a8d-44ba-8cbb-a162ce274165",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "new_business\n",
       "0.0    643446\n",
       "1.0    252559\n",
       "NaN      1162\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['new_business'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "db6edd03-4ac5-4b1a-89a2-a4ffd80ff1bf",
   "metadata": {},
   "source": [
    "##### Urban/Rural"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "59595a8c-1cd0-4ec4-9b54-ea727a660cc0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "UrbanRural\n",
       "1    469281\n",
       "0    322826\n",
       "2    105060\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['UrbanRural'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "b1cd1f53-0af7-490b-8ebb-6c46067c1f54",
   "metadata": {},
   "outputs": [],
   "source": [
    "cond = [sba_loans['UrbanRural'].isna(), sba_loans['UrbanRural'] == '1', sba_loans['UrbanRural'] == '2']\n",
    "choice = [np.nan, 1, 0]\n",
    "sba_loans['urban_flag'] = np.select(cond, choice, np.nan)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e551258b-cb89-4b25-9f1d-2f479f3daf7a",
   "metadata": {},
   "source": [
    "##### Franchise flag\n",
    "Note we have a more detailed code, but just mark rows with no franchise info"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "2fe91805-d394-41b5-95b3-0030212e3260",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "FranchiseCode\n",
       "1        637395\n",
       "0        208040\n",
       "78760      3373\n",
       "68020      1921\n",
       "50564      1034\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['FranchiseCode'].value_counts(dropna=False).head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "b1f98f05",
   "metadata": {},
   "outputs": [],
   "source": [
    "cond = [sba_loans['FranchiseCode'].isna(), sba_loans['FranchiseCode'].isin(['1', '0', '00000', '00001'])]\n",
    "choice = [np.nan, 0]\n",
    "sba_loans['franchise_flag'] = np.select(cond, choice, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "0d4f6b84-aaeb-4eb8-a6c5-2a79308db1f6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "franchise_flag\n",
       "0.0    845435\n",
       "1.0     51732\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['franchise_flag'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cab99b14-ce12-4df8-b74c-2294eff2d7e8",
   "metadata": {},
   "source": [
    "## NAICS\n",
    "I am most interested in this feature for this project.  I am going to add the NAICS sector feature here.  I will use both NAICS and NAICS sector information in my models.  "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c05112e4-c517-4c5c-ad38-80f8dfb7502a",
   "metadata": {},
   "source": [
    "I use 2007 NAICS definitions as this data spans a range of timeframes, with median ~2009; however this one vintage may not perfectly reflect the data. See also https://www.census.gov/naics/?48967"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1f84a819-25f5-4557-a0c6-34a4e2f57b80",
   "metadata": {},
   "source": [
    "Later I will mean encode NAICS (this is my usual approach to this field).  I also will one-hot encode as this is usual, and use a numeric form of NAICS to compare performance.  In this section I map the NAICS sector and create the numeric feature.  After train/test/validation split, I do the encodings."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6af66115-9829-4ba3-9b7d-e9a49d386371",
   "metadata": {},
   "source": [
    "Later I also will drop missing NAICS.  I don't do it here "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "232fa961-261b-44a2-a817-e11f103f6159",
   "metadata": {},
   "source": [
    "##### Descriptives and clean up"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "859a90dd-cd4f-4489-9b6d-d01df1bc99c0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NAICS\n",
       "0         201667\n",
       "722110     27941\n",
       "722211     19435\n",
       "811111     14539\n",
       "621210     14034\n",
       "           ...  \n",
       "212231         1\n",
       "925120         1\n",
       "524192         1\n",
       "311351         1\n",
       "333241         1\n",
       "Name: count, Length: 1312, dtype: int64"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['NAICS'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "b6fe6c65-78d8-499d-bf08-a85e9f6efbc4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NAICS\n",
       "NaN       201667\n",
       "722110     27941\n",
       "722211     19435\n",
       "811111     14539\n",
       "621210     14034\n",
       "           ...  \n",
       "212231         1\n",
       "925120         1\n",
       "524192         1\n",
       "311351         1\n",
       "333241         1\n",
       "Name: count, Length: 1312, dtype: int64"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# First things first those 0's have to go\n",
    "sba_loans['NAICS'] = sba_loans['NAICS'].where(sba_loans['NAICS'] != '0', np.nan)\n",
    "sba_loans['NAICS'].value_counts(dropna=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "8e625a1a-160b-462c-8b2c-16219a1794d0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NAICS\n",
       "6.0    695500\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Length of remaining codes\n",
    "sba_loans['NAICS'].str.len().value_counts()\n",
    "# They all have the right number of digits!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "87e877d6-2ddf-4bb8-87c2-7a2e3a6ccb04",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count\n",
       "False    1278\n",
       "True       33\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Distribution of counts - how many codes have just 1 company (not too many)\n",
    "count_info  = sba_loans['NAICS'].value_counts(dropna=True)\n",
    "(count_info == 1).value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "3719d260-ca6e-4856-af68-2c5a0aaca0df",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NAICS\n",
       "False    695500\n",
       "True     201667\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['NAICS'].isna().value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "3f001ee6-311c-40f3-b3d9-d0ecb7f31eaf",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count     1311.000000\n",
       "mean       530.511060\n",
       "std       1503.077093\n",
       "min          1.000000\n",
       "25%         36.000000\n",
       "50%        126.000000\n",
       "75%        377.000000\n",
       "max      27941.000000\n",
       "Name: count, dtype: float64"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "count_info.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "38e5d3c3-cbdd-400b-8b30-425834da8517",
   "metadata": {},
   "outputs": [],
   "source": [
    "cond  = [count_info == 1, count_info < 10, count_info < 50, count_info < 100,\n",
    "         count_info < 1000, count_info < 10000]\n",
    "choice = ['0001', '0002-0009', '0010-0049', '0050-0099', '0100-1000', '1000-9999']\n",
    "count_cat = pd.Series(np.select(cond, choice, '10k+'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "2832873f-3e2d-42b6-bc42-cfc2580e9a98",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0100-1000    559\n",
       "0010-0049    257\n",
       "0050-0099    191\n",
       "1000-9999    153\n",
       "0002-0009    113\n",
       "0001          33\n",
       "10k+           5\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "count_cat.value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "7cf58634-6875-41ce-b0f7-dd559333e4fa",
   "metadata": {},
   "outputs": [],
   "source": [
    "# There are not so many very rare codes.  But about 50% have < 126 loans.  And there are some very common codes"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "db697475-6dfb-453c-9d0b-2676181594cd",
   "metadata": {},
   "source": [
    "##### Add the NAICS sector information"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "9b1ccddc-f1b1-473b-b961-5a0bcac1385d",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans['NAICS_2'] = sba_loans['NAICS'].str[0:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "aab7ec66-eea1-48ce-8dc0-7f4d0c9c8635",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans = sba_loans \\\n",
    "    .drop(columns=['NAICS_sector', 'NAICS_sector_desc'], errors='ignore') \\\n",
    "    .merge(naics_map, on='NAICS_2', how='left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "96bd6d52-6e42-4e42-924c-676e4949fa4a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NAICS_sector_desc\n",
       "Retail Trade                                                                126975\n",
       "Other Services (except Public Administration)                                72395\n",
       "Professional, Scientific, and Technical Services                             67922\n",
       "Manufacturing                                                                67903\n",
       "Accommodation and Food Services                                              67511\n",
       "Construction                                                                 66492\n",
       "Health Care and Social Assistance                                            55264\n",
       "Wholesale Trade                                                              48673\n",
       "Administrative and Support and Waste Management and Remediation Services     32529\n",
       "Transportation and Warehousing                                               22408\n",
       "Arts, Entertainment, and Recreation                                          14616\n",
       "Real Estate and Rental and Leasing                                           13588\n",
       "Information                                                                  11362\n",
       "Finance and Insurance                                                         9470\n",
       "Agriculture, Forestry, Fishing and Hunting                                    8995\n",
       "Educational Services                                                          6401\n",
       "Mining, Quarrying, and Oil and Gas Extraction                                 1851\n",
       "Utilities                                                                      662\n",
       "Management of Companies and Enterprises                                        256\n",
       "Public Administration (not covered in economic census)                         227\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['NAICS_sector_desc'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "69b7a4b2-9711-4f90-9720-5e8aa889ba09",
   "metadata": {},
   "outputs": [],
   "source": [
    "# There are some slightly rare sectors, e.g. Public Administration"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c58d037f-3570-4679-86ef-7789af9d0183",
   "metadata": {},
   "source": [
    "##### Add the numeric form of the NAICS features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "c5560a5c-b04a-40f9-80c8-e565f1bbefa7",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans['NAICS_num']= sba_loans['NAICS'].astype('float')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "f8c38185-89c1-4052-a0c1-2322c3ff1c8b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    695500.000000\n",
       "mean     514102.946193\n",
       "std      173375.659980\n",
       "min      111110.000000\n",
       "25%      423430.000000\n",
       "50%      512110.000000\n",
       "75%      621399.000000\n",
       "max      928120.000000\n",
       "Name: NAICS_num, dtype: float64"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['NAICS_num'].describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4820a756-110f-4710-8f21-e4b12c95797a",
   "metadata": {
    "tags": []
   },
   "source": [
    "## Age of Loan"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "21406b96-fe24-487b-bf93-9f61419c1f90",
   "metadata": {},
   "source": [
    "Older loans likely have very long terms (and therefore were designated as low risk).  Also loans in place for decades may be unlikely to default now.  In addition, I suspect many such loans will have NULL NAICS codes.  I will explore this."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "5c08a978-10e4-4b60-b7f5-5e424b51ccd7",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans['missing_NAICS'] = np.where(sba_loans['NAICS'].isna(), 1, 0)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb01e0ec-6fa3-4422-953c-5221e36314d7",
   "metadata": {},
   "source": [
    "##### Missingness and loan age"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "67a1bf93-2aaa-41df-ae69-9e73a20de19f",
   "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>median</th>\n",
       "      <th>mean</th>\n",
       "      <th>min</th>\n",
       "      <th>max</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>missing_NAICS</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2004.0</td>\n",
       "      <td>2003.018242</td>\n",
       "      <td>1984</td>\n",
       "      <td>2014</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1995.0</td>\n",
       "      <td>1994.660971</td>\n",
       "      <td>1966</td>\n",
       "      <td>2010</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               median         mean   min   max\n",
       "missing_NAICS                                 \n",
       "0              2004.0  2003.018242  1984  2014\n",
       "1              1995.0  1994.660971  1966  2010"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans.groupby('missing_NAICS')['ApprovalFY'].agg(['median', 'mean', 'min', 'max'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "218d33c0-b720-4160-97ad-9bd292a0c31f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 500x300 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot NAICS missingness\n",
    "fig, ax = plt.subplots(figsize = (5, 3))\n",
    "sba_loans.groupby('ApprovalFY')['missing_NAICS'].agg('mean').plot(ax = ax)\n",
    "ax.set_ylabel('% missing NAICS')\n",
    "plt.gcf().savefig(Path(setup.temp_path).joinpath('01_PLOT_loan_year_NAICS_missing.png'),bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "f137341b-3faa-437b-b10d-04b5eb4153db",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 500x300 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot target rate\n",
    "fig, ax = plt.subplots(figsize = (5, 3))\n",
    "sba_loans.groupby('ApprovalFY')['target'].agg('mean').plot(ax = ax)\n",
    "ax.set_ylabel('target rate')\n",
    "plt.gcf().savefig(Path(setup.temp_path).joinpath('01_PLOT_loan_year_target_rate.png'),bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "77cc6700-c4b9-4421-ba82-42f353064ffe",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 500x300 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot target rate\n",
    "# Plot NAICS missingness\n",
    "fig, ax = plt.subplots(figsize = (5, 3))\n",
    "sba_loans.groupby('ApprovalFY')['target'].agg('count').plot(ax = ax)\n",
    "ax.set_ylabel('loan count')\n",
    "plt.gcf().savefig(Path(setup.temp_path).joinpath('01_PLOT_loan_year_counts.png'),bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7cb97446-4b87-4d0d-a12b-4246c3d66de3",
   "metadata": {},
   "source": [
    "Loans prior to 1990 are rare and also have super high default rates, and often-null NAICS.  I think these loans are probably very different from more modern loans.  I am going to drop these."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0e038133-7de0-4a9d-9552-a178bb715e4e",
   "metadata": {},
   "source": [
    "I probably will not include the approval FY explicitly in my model, because new data could be drawn from future years.  However I expect that the model will still pick up on some over-time trends.  For example, it might correlate a missing NAICS with a higher default rate, due to confounding with age of loan."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0b98b43d-f16e-457e-9893-c7ad8509d34f",
   "metadata": {},
   "source": [
    "At this point I also drop rows with missing NAICS"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "c1adb060-ad00-464b-b014-87a531c50056",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans = sba_loans[(sba_loans['ApprovalFY'] >= 1990) &\n",
    "                     (sba_loans['missing_NAICS'] == 0)] \\\n",
    "    .drop(columns='missing_NAICS')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "59188e42-7a04-48f7-a898-13bb3ba0b799",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    688081.000000\n",
       "mean       2003.172576\n",
       "std           4.750860\n",
       "min        1990.000000\n",
       "25%        2001.000000\n",
       "50%        2004.000000\n",
       "75%        2006.000000\n",
       "max        2014.000000\n",
       "Name: ApprovalFY, dtype: float64"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Drop the \"missing NAICS\" feature\n",
    "sba_loans['ApprovalFY'].describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d8a89ea6-e1f7-46c6-9425-83c263f8b648",
   "metadata": {},
   "source": [
    "##### Save dataset so far"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "34f670cb-b6a6-4414-aaa8-d2991f7c0889",
   "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>ApprovalDate</th>\n",
       "      <th>ApprovalFY</th>\n",
       "      <th>Term</th>\n",
       "      <th>NoEmp</th>\n",
       "      <th>CreateJob</th>\n",
       "      <th>RetainedJob</th>\n",
       "      <th>RevLineCr</th>\n",
       "      <th>LowDoc</th>\n",
       "      <th>ChgOffDate</th>\n",
       "      <th>DisbursementDate</th>\n",
       "      <th>...</th>\n",
       "      <th>BalanceGross</th>\n",
       "      <th>ChgOffPrinGr</th>\n",
       "      <th>GrAppv</th>\n",
       "      <th>SBA_Appv</th>\n",
       "      <th>target</th>\n",
       "      <th>guaranteed_fract</th>\n",
       "      <th>new_business</th>\n",
       "      <th>urban_flag</th>\n",
       "      <th>franchise_flag</th>\n",
       "      <th>NAICS_num</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>688081</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>457016.000000</td>\n",
       "      <td>683029.000000</td>\n",
       "      <td>144040</td>\n",
       "      <td>686207</td>\n",
       "      <td>...</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>6.880810e+05</td>\n",
       "      <td>6.880810e+05</td>\n",
       "      <td>6.880810e+05</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>687231.000000</td>\n",
       "      <td>551445.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "      <td>688081.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>2003-06-08 19:41:48.344801920</td>\n",
       "      <td>2003.172576</td>\n",
       "      <td>102.479897</td>\n",
       "      <td>9.715318</td>\n",
       "      <td>2.169425</td>\n",
       "      <td>5.136857</td>\n",
       "      <td>0.425541</td>\n",
       "      <td>0.098445</td>\n",
       "      <td>2009-10-13 17:19:37.272979712</td>\n",
       "      <td>2003-09-17 10:43:04.219630208</td>\n",
       "      <td>...</td>\n",
       "      <td>3.906393</td>\n",
       "      <td>1.463758e+04</td>\n",
       "      <td>1.824504e+05</td>\n",
       "      <td>1.398423e+05</td>\n",
       "      <td>0.203011</td>\n",
       "      <td>0.678458</td>\n",
       "      <td>0.278279</td>\n",
       "      <td>0.816127</td>\n",
       "      <td>0.050324</td>\n",
       "      <td>514486.574621</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>1989-10-03 00:00:00</td>\n",
       "      <td>1990.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1990-12-14 00:00:00</td>\n",
       "      <td>1987-04-01 00:00:00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.000000e+03</td>\n",
       "      <td>5.000000e+02</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.028140</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>111110.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>2001-02-27 00:00:00</td>\n",
       "      <td>2001.000000</td>\n",
       "      <td>60.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2008-06-23 00:00:00</td>\n",
       "      <td>2001-05-31 00:00:00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>3.000000e+04</td>\n",
       "      <td>1.700000e+04</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>423450.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>2004-06-09 00:00:00</td>\n",
       "      <td>2004.000000</td>\n",
       "      <td>84.000000</td>\n",
       "      <td>4.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2010-03-06 00:00:00</td>\n",
       "      <td>2004-09-15 00:00:00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>7.500000e+04</td>\n",
       "      <td>5.000000e+04</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.750000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>512131.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>2006-09-15 00:00:00</td>\n",
       "      <td>2006.000000</td>\n",
       "      <td>120.000000</td>\n",
       "      <td>9.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>5.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2011-05-18 00:00:00</td>\n",
       "      <td>2006-11-30 00:00:00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.000000e+05</td>\n",
       "      <td>1.500000e+05</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.850000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>621399.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>2014-06-25 00:00:00</td>\n",
       "      <td>2014.000000</td>\n",
       "      <td>569.000000</td>\n",
       "      <td>9999.000000</td>\n",
       "      <td>5621.000000</td>\n",
       "      <td>9500.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>2014-08-29 00:00:00</td>\n",
       "      <td>2028-06-18 00:00:00</td>\n",
       "      <td>...</td>\n",
       "      <td>996262.000000</td>\n",
       "      <td>3.512596e+06</td>\n",
       "      <td>5.472000e+06</td>\n",
       "      <td>5.472000e+06</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>928120.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>NaN</td>\n",
       "      <td>4.750860</td>\n",
       "      <td>74.482744</td>\n",
       "      <td>58.359100</td>\n",
       "      <td>14.965085</td>\n",
       "      <td>20.635561</td>\n",
       "      <td>0.494425</td>\n",
       "      <td>0.297916</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>1648.593116</td>\n",
       "      <td>6.799730e+04</td>\n",
       "      <td>2.911322e+05</td>\n",
       "      <td>2.379942e+05</td>\n",
       "      <td>0.402241</td>\n",
       "      <td>0.175191</td>\n",
       "      <td>0.448152</td>\n",
       "      <td>0.387381</td>\n",
       "      <td>0.218613</td>\n",
       "      <td>173198.290425</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 21 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                        ApprovalDate     ApprovalFY           Term  \\\n",
       "count                         688081  688081.000000  688081.000000   \n",
       "mean   2003-06-08 19:41:48.344801920    2003.172576     102.479897   \n",
       "min              1989-10-03 00:00:00    1990.000000       0.000000   \n",
       "25%              2001-02-27 00:00:00    2001.000000      60.000000   \n",
       "50%              2004-06-09 00:00:00    2004.000000      84.000000   \n",
       "75%              2006-09-15 00:00:00    2006.000000     120.000000   \n",
       "max              2014-06-25 00:00:00    2014.000000     569.000000   \n",
       "std                              NaN       4.750860      74.482744   \n",
       "\n",
       "               NoEmp      CreateJob    RetainedJob      RevLineCr  \\\n",
       "count  688081.000000  688081.000000  688081.000000  457016.000000   \n",
       "mean        9.715318       2.169425       5.136857       0.425541   \n",
       "min         0.000000       0.000000       0.000000       0.000000   \n",
       "25%         2.000000       0.000000       0.000000       0.000000   \n",
       "50%         4.000000       0.000000       2.000000       0.000000   \n",
       "75%         9.000000       2.000000       5.000000       1.000000   \n",
       "max      9999.000000    5621.000000    9500.000000       1.000000   \n",
       "std        58.359100      14.965085      20.635561       0.494425   \n",
       "\n",
       "              LowDoc                     ChgOffDate  \\\n",
       "count  683029.000000                         144040   \n",
       "mean        0.098445  2009-10-13 17:19:37.272979712   \n",
       "min         0.000000            1990-12-14 00:00:00   \n",
       "25%         0.000000            2008-06-23 00:00:00   \n",
       "50%         0.000000            2010-03-06 00:00:00   \n",
       "75%         0.000000            2011-05-18 00:00:00   \n",
       "max         1.000000            2014-08-29 00:00:00   \n",
       "std         0.297916                            NaN   \n",
       "\n",
       "                    DisbursementDate  ...   BalanceGross  ChgOffPrinGr  \\\n",
       "count                         686207  ...  688081.000000  6.880810e+05   \n",
       "mean   2003-09-17 10:43:04.219630208  ...       3.906393  1.463758e+04   \n",
       "min              1987-04-01 00:00:00  ...       0.000000  0.000000e+00   \n",
       "25%              2001-05-31 00:00:00  ...       0.000000  0.000000e+00   \n",
       "50%              2004-09-15 00:00:00  ...       0.000000  0.000000e+00   \n",
       "75%              2006-11-30 00:00:00  ...       0.000000  0.000000e+00   \n",
       "max              2028-06-18 00:00:00  ...  996262.000000  3.512596e+06   \n",
       "std                              NaN  ...    1648.593116  6.799730e+04   \n",
       "\n",
       "             GrAppv      SBA_Appv         target  guaranteed_fract  \\\n",
       "count  6.880810e+05  6.880810e+05  688081.000000     688081.000000   \n",
       "mean   1.824504e+05  1.398423e+05       0.203011          0.678458   \n",
       "min    1.000000e+03  5.000000e+02       0.000000          0.028140   \n",
       "25%    3.000000e+04  1.700000e+04       0.000000          0.500000   \n",
       "50%    7.500000e+04  5.000000e+04       0.000000          0.750000   \n",
       "75%    2.000000e+05  1.500000e+05       0.000000          0.850000   \n",
       "max    5.472000e+06  5.472000e+06       1.000000          1.000000   \n",
       "std    2.911322e+05  2.379942e+05       0.402241          0.175191   \n",
       "\n",
       "        new_business     urban_flag  franchise_flag      NAICS_num  \n",
       "count  687231.000000  551445.000000   688081.000000  688081.000000  \n",
       "mean        0.278279       0.816127        0.050324  514486.574621  \n",
       "min         0.000000       0.000000        0.000000  111110.000000  \n",
       "25%         0.000000       1.000000        0.000000  423450.000000  \n",
       "50%         0.000000       1.000000        0.000000  512131.000000  \n",
       "75%         1.000000       1.000000        0.000000  621399.000000  \n",
       "max         1.000000       1.000000        1.000000  928120.000000  \n",
       "std         0.448152       0.387381        0.218613  173198.290425  \n",
       "\n",
       "[8 rows x 21 columns]"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "678a7cde-1b80-4721-aae8-45399ea56fbc",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans.to_parquet(Path(setup.temp_path).joinpath('01_DATA_import_filtered.parquet'))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cbdc33e5-3650-4818-bee7-7c7685e268e3",
   "metadata": {},
   "source": [
    "## Train/Test Split\n",
    "I will do mean and one hot encoding, but need to split first to avoid data leakage.  I will do a 70/15/15 split.  "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0b81f32c-b5d3-4292-941b-35be7f02102d",
   "metadata": {},
   "source": [
    "I will put all cases for a few NAICS into the test data set, to test the effect of encodings on unseen codes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "55374304-2675-4436-892b-14299f150644",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1311, 1)"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans[['NAICS']].drop_duplicates().shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "b3f224b5-f31b-4110-86da-038c4133476c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(131, 1)"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "holdout_naics = sba_loans[['NAICS']].drop_duplicates().sample(frac=0.10, random_state =2432)\n",
    "holdout_naics.to_csv(Path(setup.temp_path).joinpath('01_REPORT_holdout_naics.csv'))\n",
    "holdout_naics.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "90e5df47-35ef-41ad-a1f8-8b300998d370",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_merge = sba_loans[['LoanNr_ChkDgt', 'NAICS']].reset_index() \\\n",
    "    .merge(holdout_naics, on='NAICS', how='left', indicator=True) \\\n",
    "    .set_index('index')\n",
    "sba_holdout_naics_index = sba_merge[sba_merge['_merge'] == 'both'].index\n",
    "sba_allow_naics_index = sba_merge[sba_merge['_merge'] == 'left_only'].index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "93af0301-2865-475a-a50c-ada398719151",
   "metadata": {},
   "outputs": [],
   "source": [
    "[train_index, other_index] = train_test_split(sba_allow_naics_index, \n",
    "                                              train_size=0.70, random_state=2343)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "77f3ec90-9c95-4cf8-90f5-0a5472b9995a",
   "metadata": {},
   "outputs": [],
   "source": [
    "[val_index, test_split_ind] = train_test_split(other_index, \n",
    "                                               train_size=0.50, random_state=2343)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "78527c54-4f37-452d-974b-6bfaf512eebe",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_index = test_split_ind.union(sba_holdout_naics_index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "843d709f-8077-4d02-b6c6-c64bd3e632ea",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train index: (436120,), validation index: (93454,), test index: (158507,)\n",
      "test regular split: (93455,), NAICS holdout:(65052,)\n",
      "total: 688081\n"
     ]
    }
   ],
   "source": [
    "print(f'train index: {train_index.shape}, validation index: {val_index.shape}, test index: {test_index.shape}')\n",
    "print(f'test regular split: {test_split_ind.shape}, NAICS holdout:{sba_holdout_naics_index.shape}')\n",
    "print(f'total: {len(train_index) + len(val_index) + len(test_index)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "542728da-a16c-4ac4-8ac3-e6312d0d122a",
   "metadata": {},
   "outputs": [],
   "source": [
    "#sba_loans_save = sba_loans.copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "id": "054eaed4-7f3b-4065-885a-27e7a471f62c",
   "metadata": {},
   "outputs": [],
   "source": [
    "#cond = [sba_loans.index\n",
    "cond= [sba_loans.index.isin(train_index), sba_loans.index.isin(val_index)]\n",
    "choice=['train', 'val']\n",
    "sba_loans['dset'] = np.select(cond, choice, 'test')\n",
    "sba_loans['dset_naics_holdout'] = np.where(sba_loans.index.isin(sba_holdout_naics_index), 1, 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "id": "633fe387-2cc8-4960-add9-7792885c9977",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "dset\n",
       "train    436120\n",
       "test     158507\n",
       "val       93454\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['dset'].value_counts()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b8233624-b8a9-4e02-ad16-e8b9da8da37b",
   "metadata": {},
   "source": [
    "## Target (Mean) Encoding"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9eed231b-981a-46be-b011-25dce6f99820",
   "metadata": {},
   "source": [
    "##### Single-feature mean encoders.  \n",
    "I don't want to overfit, and so lower volume categories must be smoothed.  "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d2f625cd-b253-487f-8a01-a285456aec0f",
   "metadata": {},
   "source": [
    "If only one code is input, my Hierarchical Blender is a \"normal\" target encoder, which blends the overall mean with the NAICS mean.  The blending is sinusoidal and paramaeterized."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "259bc03d-c706-4b15-a72c-6c7fbf5a0528",
   "metadata": {},
   "source": [
    "I will create features for the NAICS hierarchy and blend them *individually* first"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "id": "d19bcdaf-9eb7-4b58-aeef-e8d79e518136",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans['NAICS_5'] = sba_loans['NAICS'].str[0:5] # \"Industries\"\n",
    "sba_loans['NAICS_4'] = sba_loans['NAICS'].str[0:4] # \"Industry Groups\"\n",
    "sba_loans['NAICS_3'] = sba_loans['NAICS'].str[0:3] # \"Subsectors\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "id": "af5436dc-c756-4bef-95e0-65c17f97d962",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NAICS\n",
       "0    456017\n",
       "1    101119\n",
       "2     46504\n",
       "9     34750\n",
       "8     18739\n",
       "3     17087\n",
       "5      6030\n",
       "4      4542\n",
       "6      2417\n",
       "7       876\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Are there distinct industries?\n",
    "sba_loans['NAICS'].str[-1].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "2abaa4fd-78df-41ba-a342-dcb56ff904c1",
   "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>NAICS</th>\n",
       "      <th>NAICS_2</th>\n",
       "      <th>NAICS_sector</th>\n",
       "      <th>NAICS_sector_desc</th>\n",
       "      <th>NAICS_num</th>\n",
       "      <th>NAICS_5</th>\n",
       "      <th>NAICS_4</th>\n",
       "      <th>NAICS_3</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>451120</td>\n",
       "      <td>45</td>\n",
       "      <td>44-45</td>\n",
       "      <td>Retail Trade</td>\n",
       "      <td>451120.0</td>\n",
       "      <td>45112</td>\n",
       "      <td>4511</td>\n",
       "      <td>451</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>722410</td>\n",
       "      <td>72</td>\n",
       "      <td>72</td>\n",
       "      <td>Accommodation and Food Services</td>\n",
       "      <td>722410.0</td>\n",
       "      <td>72241</td>\n",
       "      <td>7224</td>\n",
       "      <td>722</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    NAICS NAICS_2 NAICS_sector                NAICS_sector_desc  NAICS_num  \\\n",
       "0  451120      45        44-45                     Retail Trade   451120.0   \n",
       "1  722410      72           72  Accommodation and Food Services   722410.0   \n",
       "\n",
       "  NAICS_5 NAICS_4 NAICS_3  \n",
       "0   45112    4511     451  \n",
       "1   72241    7224     722  "
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans.filter(like='NAICS').head(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "id": "9ffbcb3c-8211-479e-a8ab-86af216be8fd",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Function to implement target encoding with or without hiearchical blending.\n",
    "# No hierachy = just pass one feature at a time (single item list)\n",
    "# I will save just the lowest level enoder\n",
    "def do_blend(data, feature = ['NAICS'], name_out = 'menc',\n",
    "            save_enc = False, save_path = None):\n",
    "    train_data = data[data['dset'] == 'train']\n",
    "    hier_enc = HierarchicalEncoder()\n",
    "    hier_enc.fit(train_data[feature],train_data['target'])\n",
    "    \n",
    "    if save_enc:\n",
    "        with open(save_path, 'wb') as fout:\n",
    "            pickle.dump(hier_enc, fout)\n",
    "    return hier_enc.transform(data[feature]) \\\n",
    "        .rename(name_out)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "id": "76129dcd-0eff-494a-89b8-1bb811ec22b5",
   "metadata": {},
   "outputs": [],
   "source": [
    "naics_menc_list = ['NAICS', 'NAICS_5', 'NAICS_4', 'NAICS_3', 'NAICS_sector']\n",
    "naics_menc_name = ['menc_' + c for c in naics_menc_list]\n",
    "# Save only the full mean encoder\n",
    "save_hier = [True if n == 'menc_NAICS' else False for n in naics_menc_name]\n",
    "save_hier_name = [Path(setup.temp_path).joinpath('01_DATA_mean_encoder_single.pkl') \\\n",
    "                      if n == 'menc_NAICS' else None for n in naics_menc_name]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "id": "39506c40-1561-4122-a59b-6d4a24b7042e",
   "metadata": {},
   "outputs": [],
   "source": [
    "menc_single_df = pd.concat([do_blend(sba_loans, [naics_menc_list[i]], \n",
    "                                     naics_menc_name[i], save_hier[i], save_hier_name[i]) \\\n",
    "                                for i in range(len(naics_menc_list))],\n",
    "                           axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "id": "1a6316b6-c287-4a63-8e13-bf3162e7b7dc",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans.drop(columns=naics_menc_name, errors='ignore')\n",
    "sba_loans = pd.concat([sba_loans, menc_single_df], axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ad8c099b-a0ec-41cc-a115-d08bc998f924",
   "metadata": {},
   "source": [
    "##### Hierarchical Blending\n",
    "Use all the levels at once to get a single feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "id": "9a74dd07-dbf6-40ec-a8cd-9986114326f9",
   "metadata": {},
   "outputs": [],
   "source": [
    "menc_hier = do_blend(sba_loans, naics_menc_list, name_out = 'mhier_NAICS')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "id": "7d77147c-e75f-430a-999a-7ed6581e1ad2",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans.drop(columns=['mhier_NAICS'], errors='ignore')\n",
    "sba_loans = pd.concat([sba_loans, menc_hier], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "id": "7a0115bf-7fda-44de-9317-e724a9bc3ad8",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(Path(setup.temp_path).joinpath('01_DATA_mean_encoder_hier.pkl'), 'wb') as fout:\n",
    "    pickle.dump(menc_hier, fout)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "02aa774a-a5f6-42b2-9388-76e7faa5dbae",
   "metadata": {},
   "source": [
    "##### Correlations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "id": "001d2234-a9ce-44a1-a04c-e0e0a05534bc",
   "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>menc_NAICS</th>\n",
       "      <th>menc_NAICS_5</th>\n",
       "      <th>menc_NAICS_4</th>\n",
       "      <th>menc_NAICS_3</th>\n",
       "      <th>menc_NAICS_sector</th>\n",
       "      <th>mhier_NAICS</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>menc_NAICS</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.953888</td>\n",
       "      <td>0.861789</td>\n",
       "      <td>0.741466</td>\n",
       "      <td>0.445599</td>\n",
       "      <td>0.962039</td>\n",
       "      <td>0.227233</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>menc_NAICS_5</th>\n",
       "      <td>0.953888</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.899522</td>\n",
       "      <td>0.771791</td>\n",
       "      <td>0.475434</td>\n",
       "      <td>0.948546</td>\n",
       "      <td>0.222323</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>menc_NAICS_4</th>\n",
       "      <td>0.861789</td>\n",
       "      <td>0.899522</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.853746</td>\n",
       "      <td>0.511701</td>\n",
       "      <td>0.890962</td>\n",
       "      <td>0.208747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>menc_NAICS_3</th>\n",
       "      <td>0.741466</td>\n",
       "      <td>0.771791</td>\n",
       "      <td>0.853746</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.585696</td>\n",
       "      <td>0.775213</td>\n",
       "      <td>0.182358</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>menc_NAICS_sector</th>\n",
       "      <td>0.445599</td>\n",
       "      <td>0.475434</td>\n",
       "      <td>0.511701</td>\n",
       "      <td>0.585696</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.454516</td>\n",
       "      <td>0.107757</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mhier_NAICS</th>\n",
       "      <td>0.962039</td>\n",
       "      <td>0.948546</td>\n",
       "      <td>0.890962</td>\n",
       "      <td>0.775213</td>\n",
       "      <td>0.454516</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.230828</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>target</th>\n",
       "      <td>0.227233</td>\n",
       "      <td>0.222323</td>\n",
       "      <td>0.208747</td>\n",
       "      <td>0.182358</td>\n",
       "      <td>0.107757</td>\n",
       "      <td>0.230828</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   menc_NAICS  menc_NAICS_5  menc_NAICS_4  menc_NAICS_3  \\\n",
       "menc_NAICS           1.000000      0.953888      0.861789      0.741466   \n",
       "menc_NAICS_5         0.953888      1.000000      0.899522      0.771791   \n",
       "menc_NAICS_4         0.861789      0.899522      1.000000      0.853746   \n",
       "menc_NAICS_3         0.741466      0.771791      0.853746      1.000000   \n",
       "menc_NAICS_sector    0.445599      0.475434      0.511701      0.585696   \n",
       "mhier_NAICS          0.962039      0.948546      0.890962      0.775213   \n",
       "target               0.227233      0.222323      0.208747      0.182358   \n",
       "\n",
       "                   menc_NAICS_sector  mhier_NAICS    target  \n",
       "menc_NAICS                  0.445599     0.962039  0.227233  \n",
       "menc_NAICS_5                0.475434     0.948546  0.222323  \n",
       "menc_NAICS_4                0.511701     0.890962  0.208747  \n",
       "menc_NAICS_3                0.585696     0.775213  0.182358  \n",
       "menc_NAICS_sector           1.000000     0.454516  0.107757  \n",
       "mhier_NAICS                 0.454516     1.000000  0.230828  \n",
       "target                      0.107757     0.230828  1.000000  "
      ]
     },
     "execution_count": 84,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "menc_corr = sba_loans[naics_menc_name + ['mhier_NAICS', 'target']].corr()\n",
    "menc_corr.to_csv(Path(setup.temp_path).joinpath('01_REPORT_correlations_menc_single.csv'))\n",
    "menc_corr"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7d90e79a-4cb5-43d7-9e11-9493f285b6dc",
   "metadata": {},
   "source": [
    "### One Hot Encoding\n",
    "NAICS sector only"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "id": "e2513685-cc7e-4c19-aea4-ff6bcb2387f0",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_df = sba_loans[sba_loans['dset'] == 'train']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "id": "85113b82-9ed8-41a2-8ffd-6208f909b88b",
   "metadata": {},
   "outputs": [],
   "source": [
    "naics_onehot_encoder = OneHotEncoder(sparse_output=False, handle_unknown='infrequent_if_exist',\n",
    "                                  min_frequency = 0.05)  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "id": "56a447b1-7310-4b9e-93b1-5df6fe8b9522",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>#sk-container-id-1 {color: black;}#sk-container-id-1 pre{padding: 0;}#sk-container-id-1 div.sk-toggleable {background-color: white;}#sk-container-id-1 label.sk-toggleable__label {cursor: pointer;display: block;width: 100%;margin-bottom: 0;padding: 0.3em;box-sizing: border-box;text-align: center;}#sk-container-id-1 label.sk-toggleable__label-arrow:before {content: \"▸\";float: left;margin-right: 0.25em;color: #696969;}#sk-container-id-1 label.sk-toggleable__label-arrow:hover:before {color: black;}#sk-container-id-1 div.sk-estimator:hover label.sk-toggleable__label-arrow:before {color: black;}#sk-container-id-1 div.sk-toggleable__content {max-height: 0;max-width: 0;overflow: hidden;text-align: left;background-color: #f0f8ff;}#sk-container-id-1 div.sk-toggleable__content pre {margin: 0.2em;color: black;border-radius: 0.25em;background-color: #f0f8ff;}#sk-container-id-1 input.sk-toggleable__control:checked~div.sk-toggleable__content {max-height: 200px;max-width: 100%;overflow: auto;}#sk-container-id-1 input.sk-toggleable__control:checked~label.sk-toggleable__label-arrow:before {content: \"▾\";}#sk-container-id-1 div.sk-estimator input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 div.sk-label input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 input.sk-hidden--visually {border: 0;clip: rect(1px 1px 1px 1px);clip: rect(1px, 1px, 1px, 1px);height: 1px;margin: -1px;overflow: hidden;padding: 0;position: absolute;width: 1px;}#sk-container-id-1 div.sk-estimator {font-family: monospace;background-color: #f0f8ff;border: 1px dotted black;border-radius: 0.25em;box-sizing: border-box;margin-bottom: 0.5em;}#sk-container-id-1 div.sk-estimator:hover {background-color: #d4ebff;}#sk-container-id-1 div.sk-parallel-item::after {content: \"\";width: 100%;border-bottom: 1px solid gray;flex-grow: 1;}#sk-container-id-1 div.sk-label:hover label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-1 div.sk-serial::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: 0;}#sk-container-id-1 div.sk-serial {display: flex;flex-direction: column;align-items: center;background-color: white;padding-right: 0.2em;padding-left: 0.2em;position: relative;}#sk-container-id-1 div.sk-item {position: relative;z-index: 1;}#sk-container-id-1 div.sk-parallel {display: flex;align-items: stretch;justify-content: center;background-color: white;position: relative;}#sk-container-id-1 div.sk-item::before, #sk-container-id-1 div.sk-parallel-item::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: -1;}#sk-container-id-1 div.sk-parallel-item {display: flex;flex-direction: column;z-index: 1;position: relative;background-color: white;}#sk-container-id-1 div.sk-parallel-item:first-child::after {align-self: flex-end;width: 50%;}#sk-container-id-1 div.sk-parallel-item:last-child::after {align-self: flex-start;width: 50%;}#sk-container-id-1 div.sk-parallel-item:only-child::after {width: 0;}#sk-container-id-1 div.sk-dashed-wrapped {border: 1px dashed gray;margin: 0 0.4em 0.5em 0.4em;box-sizing: border-box;padding-bottom: 0.4em;background-color: white;}#sk-container-id-1 div.sk-label label {font-family: monospace;font-weight: bold;display: inline-block;line-height: 1.2em;}#sk-container-id-1 div.sk-label-container {text-align: center;}#sk-container-id-1 div.sk-container {/* jupyter's `normalize.less` sets `[hidden] { display: none; }` but bootstrap.min.css set `[hidden] { display: none !important; }` so we also need the `!important` here to be able to override the default hidden behavior on the sphinx rendered scikit-learn.org. See: https://github.com/scikit-learn/scikit-learn/issues/21755 */display: inline-block !important;position: relative;}#sk-container-id-1 div.sk-text-repr-fallback {display: none;}</style><div id=\"sk-container-id-1\" class=\"sk-top-container\"><div class=\"sk-text-repr-fallback\"><pre>OneHotEncoder(handle_unknown=&#x27;infrequent_if_exist&#x27;, min_frequency=0.05,\n",
       "              sparse_output=False)</pre><b>In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. <br />On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.</b></div><div class=\"sk-container\" hidden><div class=\"sk-item\"><div class=\"sk-estimator sk-toggleable\"><input class=\"sk-toggleable__control sk-hidden--visually\" id=\"sk-estimator-id-1\" type=\"checkbox\" checked><label for=\"sk-estimator-id-1\" class=\"sk-toggleable__label sk-toggleable__label-arrow\">OneHotEncoder</label><div class=\"sk-toggleable__content\"><pre>OneHotEncoder(handle_unknown=&#x27;infrequent_if_exist&#x27;, min_frequency=0.05,\n",
       "              sparse_output=False)</pre></div></div></div></div></div>"
      ],
      "text/plain": [
       "OneHotEncoder(handle_unknown='infrequent_if_exist', min_frequency=0.05,\n",
       "              sparse_output=False)"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "naics_onehot_encoder.fit(train_df[['NAICS_sector_desc']])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "4a42165f-e096-4503-a8c2-951444ec2cfd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['NAICS_sector_desc'], dtype=object)"
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "naics_onehot_encoder.feature_names_in_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "id": "a0f917f4-826f-42e5-bba2-e1ca1e65ca25",
   "metadata": {},
   "outputs": [],
   "source": [
    "def do_transform_onehot(data, enc = naics_onehot_encoder,\n",
    "                       prefix = 'NS__'):\n",
    "    this_feat = enc.feature_names_in_[0]\n",
    "    onehot_df = naics_sec_onehot = pd.DataFrame(enc.transform(data[[this_feat]]))\n",
    "    onehot_df.set_index(data.index, inplace=True)\n",
    "    onehot_df.columns = [x.replace(this_feat,prefix) for x in enc.get_feature_names_out()]\n",
    "    \n",
    "    return pd.concat([data.drop(columns=onehot_df.columns, errors='ignore'),\n",
    "                      onehot_df], axis=1)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "id": "b797d58c-1bbe-48f2-bb60-ec9bcb37e51d",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans = do_transform_onehot(sba_loans)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "91a30c34-a5b9-4b76-ade2-9d1aefd078fc",
   "metadata": {},
   "source": [
    "## Count Encoding\n",
    "Also count encode for NAICS."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "id": "ad2a8ed7-80ca-4613-8eea-bb1e9a916d16",
   "metadata": {},
   "outputs": [],
   "source": [
    "count_enc = CountEncoder(normalize=True, handle_unknown=0,handle_missing = 'return_nan')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "id": "df7533a8-39b3-4fc1-844b-265d3974492e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>#sk-container-id-2 {color: black;}#sk-container-id-2 pre{padding: 0;}#sk-container-id-2 div.sk-toggleable {background-color: white;}#sk-container-id-2 label.sk-toggleable__label {cursor: pointer;display: block;width: 100%;margin-bottom: 0;padding: 0.3em;box-sizing: border-box;text-align: center;}#sk-container-id-2 label.sk-toggleable__label-arrow:before {content: \"▸\";float: left;margin-right: 0.25em;color: #696969;}#sk-container-id-2 label.sk-toggleable__label-arrow:hover:before {color: black;}#sk-container-id-2 div.sk-estimator:hover label.sk-toggleable__label-arrow:before {color: black;}#sk-container-id-2 div.sk-toggleable__content {max-height: 0;max-width: 0;overflow: hidden;text-align: left;background-color: #f0f8ff;}#sk-container-id-2 div.sk-toggleable__content pre {margin: 0.2em;color: black;border-radius: 0.25em;background-color: #f0f8ff;}#sk-container-id-2 input.sk-toggleable__control:checked~div.sk-toggleable__content {max-height: 200px;max-width: 100%;overflow: auto;}#sk-container-id-2 input.sk-toggleable__control:checked~label.sk-toggleable__label-arrow:before {content: \"▾\";}#sk-container-id-2 div.sk-estimator input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-2 div.sk-label input.sk-toggleable__control:checked~label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-2 input.sk-hidden--visually {border: 0;clip: rect(1px 1px 1px 1px);clip: rect(1px, 1px, 1px, 1px);height: 1px;margin: -1px;overflow: hidden;padding: 0;position: absolute;width: 1px;}#sk-container-id-2 div.sk-estimator {font-family: monospace;background-color: #f0f8ff;border: 1px dotted black;border-radius: 0.25em;box-sizing: border-box;margin-bottom: 0.5em;}#sk-container-id-2 div.sk-estimator:hover {background-color: #d4ebff;}#sk-container-id-2 div.sk-parallel-item::after {content: \"\";width: 100%;border-bottom: 1px solid gray;flex-grow: 1;}#sk-container-id-2 div.sk-label:hover label.sk-toggleable__label {background-color: #d4ebff;}#sk-container-id-2 div.sk-serial::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: 0;}#sk-container-id-2 div.sk-serial {display: flex;flex-direction: column;align-items: center;background-color: white;padding-right: 0.2em;padding-left: 0.2em;position: relative;}#sk-container-id-2 div.sk-item {position: relative;z-index: 1;}#sk-container-id-2 div.sk-parallel {display: flex;align-items: stretch;justify-content: center;background-color: white;position: relative;}#sk-container-id-2 div.sk-item::before, #sk-container-id-2 div.sk-parallel-item::before {content: \"\";position: absolute;border-left: 1px solid gray;box-sizing: border-box;top: 0;bottom: 0;left: 50%;z-index: -1;}#sk-container-id-2 div.sk-parallel-item {display: flex;flex-direction: column;z-index: 1;position: relative;background-color: white;}#sk-container-id-2 div.sk-parallel-item:first-child::after {align-self: flex-end;width: 50%;}#sk-container-id-2 div.sk-parallel-item:last-child::after {align-self: flex-start;width: 50%;}#sk-container-id-2 div.sk-parallel-item:only-child::after {width: 0;}#sk-container-id-2 div.sk-dashed-wrapped {border: 1px dashed gray;margin: 0 0.4em 0.5em 0.4em;box-sizing: border-box;padding-bottom: 0.4em;background-color: white;}#sk-container-id-2 div.sk-label label {font-family: monospace;font-weight: bold;display: inline-block;line-height: 1.2em;}#sk-container-id-2 div.sk-label-container {text-align: center;}#sk-container-id-2 div.sk-container {/* jupyter's `normalize.less` sets `[hidden] { display: none; }` but bootstrap.min.css set `[hidden] { display: none !important; }` so we also need the `!important` here to be able to override the default hidden behavior on the sphinx rendered scikit-learn.org. See: https://github.com/scikit-learn/scikit-learn/issues/21755 */display: inline-block !important;position: relative;}#sk-container-id-2 div.sk-text-repr-fallback {display: none;}</style><div id=\"sk-container-id-2\" class=\"sk-top-container\"><div class=\"sk-text-repr-fallback\"><pre>CountEncoder(cols=[&#x27;NAICS&#x27;], combine_min_nan_groups=True,\n",
       "             handle_missing=&#x27;return_nan&#x27;, handle_unknown=0, normalize=True)</pre><b>In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook. <br />On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.</b></div><div class=\"sk-container\" hidden><div class=\"sk-item\"><div class=\"sk-estimator sk-toggleable\"><input class=\"sk-toggleable__control sk-hidden--visually\" id=\"sk-estimator-id-2\" type=\"checkbox\" checked><label for=\"sk-estimator-id-2\" class=\"sk-toggleable__label sk-toggleable__label-arrow\">CountEncoder</label><div class=\"sk-toggleable__content\"><pre>CountEncoder(cols=[&#x27;NAICS&#x27;], combine_min_nan_groups=True,\n",
       "             handle_missing=&#x27;return_nan&#x27;, handle_unknown=0, normalize=True)</pre></div></div></div></div></div>"
      ],
      "text/plain": [
       "CountEncoder(cols=['NAICS'], combine_min_nan_groups=True,\n",
       "             handle_missing='return_nan', handle_unknown=0, normalize=True)"
      ]
     },
     "execution_count": 92,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "count_enc.fit(train_df[['NAICS']])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "id": "c7c9cc03-8510-45c2-aa27-73f30dd1e659",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans['cenc_NAICS']  = count_enc.transform(sba_loans[['NAICS']])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "id": "eee94bbb-f1ad-4ce5-9017-b91ef1a4210e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    688081.000000\n",
       "mean          0.007265\n",
       "std           0.010443\n",
       "min           0.000000\n",
       "25%           0.000718\n",
       "50%           0.003056\n",
       "75%           0.009309\n",
       "max           0.044873\n",
       "Name: cenc_NAICS, dtype: float64"
      ]
     },
     "execution_count": 94,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['cenc_NAICS'].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "id": "359d4447-a9c5-404f-ac66-6c14522b95d0",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "cenc_NAICS\n",
       "False    688081\n",
       "Name: count, dtype: int64"
      ]
     },
     "execution_count": 95,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans['cenc_NAICS'].isna().value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "id": "0b4c5b08-10b1-4994-a5f9-885d978913d1",
   "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>NAICS</th>\n",
       "      <th>NAICS_sector_desc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>451120</td>\n",
       "      <td>Retail Trade</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>621210</td>\n",
       "      <td>Health Care and Social Assistance</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>811118</td>\n",
       "      <td>Other Services (except Public Administration)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>721310</td>\n",
       "      <td>Accommodation and Food Services</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>811111</td>\n",
       "      <td>Other Services (except Public Administration)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "     NAICS                              NAICS_sector_desc\n",
       "0   451120                                   Retail Trade\n",
       "2   621210              Health Care and Social Assistance\n",
       "7   811118  Other Services (except Public Administration)\n",
       "8   721310                Accommodation and Food Services\n",
       "10  811111  Other Services (except Public Administration)"
      ]
     },
     "execution_count": 96,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df.head(5) \\\n",
    "    [['NAICS', 'NAICS_sector_desc'] + [c for c in train_df.columns if c.startswith('NS__')]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "id": "5d9f707e-868f-4764-9c04-427ca8f9e2f3",
   "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>NAICS</th>\n",
       "      <th>NAICS_sector_desc</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>53</th>\n",
       "      <td>621210</td>\n",
       "      <td>Health Care and Social Assistance</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>71</th>\n",
       "      <td>621210</td>\n",
       "      <td>Health Care and Social Assistance</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>106</th>\n",
       "      <td>451220</td>\n",
       "      <td>Retail Trade</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>115</th>\n",
       "      <td>441222</td>\n",
       "      <td>Retail Trade</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>119</th>\n",
       "      <td>451120</td>\n",
       "      <td>Retail Trade</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      NAICS                  NAICS_sector_desc\n",
       "53   621210  Health Care and Social Assistance\n",
       "71   621210  Health Care and Social Assistance\n",
       "106  451220                       Retail Trade\n",
       "115  441222                       Retail Trade\n",
       "119  451120                       Retail Trade"
      ]
     },
     "execution_count": 97,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans[sba_loans['dset'] == 'val'].head(5) \\\n",
    "    [['NAICS', 'NAICS_sector_desc'] + [c for c in train_df.columns if c.startswith('NS__')]]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ad14245a-ab3d-46fd-ae6c-3a404a3da2fa",
   "metadata": {},
   "source": [
    "## Save Transformed Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "id": "55a50dfe-0f60-46ad-8b76-fac3801207ee",
   "metadata": {},
   "outputs": [],
   "source": [
    "sba_loans.to_parquet(Path(setup.temp_path).joinpath('01_DATA_transformed.parquet'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "id": "2e77c1e0-7501-4817-859b-04550f074466",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dataset shape: (688081, 58), unique indices: (688081, 58)\n"
     ]
    }
   ],
   "source": [
    "print(f\"dataset shape: {sba_loans.shape}, unique indices: {sba_loans.drop_duplicates('LoanNr_ChkDgt').shape}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "id": "5a0b8e1f-080a-4348-b2f0-9de010db6522",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['LoanNr_ChkDgt', 'Name', 'City', 'State', 'Zip', 'Bank', 'BankState',\n",
       "       'NAICS', 'ApprovalDate', 'ApprovalFY', 'Term', 'NoEmp', 'NewExist',\n",
       "       'CreateJob', 'RetainedJob', 'FranchiseCode', 'UrbanRural', 'RevLineCr',\n",
       "       'LowDoc', 'ChgOffDate', 'DisbursementDate', 'DisbursementGross',\n",
       "       'BalanceGross', 'MIS_Status', 'ChgOffPrinGr', 'GrAppv', 'SBA_Appv',\n",
       "       'target', 'guaranteed_fract', 'new_business', 'urban_flag',\n",
       "       'franchise_flag', 'NAICS_2', 'NAICS_sector', 'NAICS_sector_desc',\n",
       "       'NAICS_num', 'dset', 'dset_naics_holdout', 'NAICS_5', 'NAICS_4',\n",
       "       'NAICS_3', 'menc_NAICS', 'menc_NAICS_5', 'menc_NAICS_4', 'menc_NAICS_3',\n",
       "       'menc_NAICS_sector', 'mhier_NAICS',\n",
       "       'NS___Accommodation and Food Services',\n",
       "       'NS___Administrative and Support and Waste Management and Remediation Services',\n",
       "       'NS___Construction', 'NS___Health Care and Social Assistance',\n",
       "       'NS___Manufacturing',\n",
       "       'NS___Other Services (except Public Administration)',\n",
       "       'NS___Professional, Scientific, and Technical Services',\n",
       "       'NS___Retail Trade', 'NS___Wholesale Trade', 'NS___infrequent_sklearn',\n",
       "       'cenc_NAICS'],\n",
       "      dtype='object')"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sba_loans.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d7fbed41-77c8-4392-8739-12b2cf6c39f5",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "tf_p39",
   "language": "python",
   "name": "tf_p39"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.18"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}