7_business_credit_risk_proxy.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "c0fd2853-ed39-407c-8b0c-c126b7f4f41a",
   "metadata": {},
   "source": [
    "# Business credit risk proxy model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "138d69ed-a251-4361-b585-fcb39546d6f2",
   "metadata": {},
   "source": [
    "### *Can we learn to assess business credit risk from the world's greatest record of business performance -- the US stock market?*\n",
    "\n",
    "---\n",
    "\n",
    "## Introduction\n",
    "\n",
    "The other day I was speaking to a friend who works for a web B2B lender. While thinking about their business model, I became curious: *Can one use a stock's recent financials to predict distress in the following 12 months?* If so, often a few features hold most of the predictive power, meaning we could likely build a decent model based on relatively few financial details. That would be very handy to a B2B lender, who could require these details upon application.\n",
    "\n",
    "Benefits to the lender would include:\n",
    "- Large, multi-industry scope.\n",
    "- Broad economic/time scope: These data span the dot com bubble, the GFC, and the COVID era.\n",
    "- Potential for bespoke models or segmentation without recourse to a credit rating agency.\n",
    "- The data are inexpensive.\n",
    "\n",
    "---\n",
    "\n",
    "## Plan\n",
    "\n",
    "This notebook will primarily address the general prediction problem. I'll start by predicting using many features to assess the overall modeling idea. Toward the end, I'll evaluate how skewed the feature importances are towards the top few in the final model.\n",
    "\n",
    "### Target: What to predict?\n",
    "- Interest coverage (EBIT over interest expense) appears to be a popular measure of financial distress. However, I don't like how bond-centric it appears to be.\n",
    "- A ratio of cash flow to current liabilities or current debt probably makes sense. After all, a cash crunch is likely a solid proxy for an inability to pay.\n",
    "- Numerator\n",
    "    - Current debt is the account a 1-year B2B loan would fall under.\n",
    "    - However let's remember that a distressed company will need to prioritize among all its current liabilities -- such as payroll -- not just loans.\n",
    "    - So I think the broader current figure, `current liabilities` (`liabilitiesc` in our data), is the better choice.\n",
    "- Denominator\n",
    "    - Free cash flow (FCF) is more stringent than operating cash flow (NCFO, or net cash flow from operationss).\n",
    "    - However, FCF = NCFO - CapEx, and since CapEx can be volatile, so can FCF.\n",
    "    - Therefore here too, the broader NCFO is probably better.\n",
    "    - I'll run with a `liabilitiesc` / `ncfo`, called \"ratio\" below.\n",
    "        - See [the target table SQL for this notebook](./dbt_sharadar_demo/models/fct_daily_credit_risk_targ.sql) to see how it is measured.\n",
    "        - So `ratio_ttm` is this ratio for trailing twelve months, and `max_ratio_ftm` is the max value of this ratio for a company in the following twelve months.\n",
    "        - You'll also see `ratio_adj`, which deals with the rare negative `liabilitiesc` values.\n",
    "\n",
    "### Features\n",
    "- Financial ratios make companies of different scales comparable.\n",
    "    - They also become comparable across time, obviating the need for inflation adjustment.\n",
    "- I'm going to create many ratios of financial currency figures over a few denominators -- one denominator from each financial statement: revenue, assets, free cash flow.\n",
    "- In particular, I will compute:\n",
    "    - Ratios for trailing twelve months (TTM)\n",
    "    - 5-year ratio average (5Y_AVG).\n",
    "    - Ratio differences (TTM - 5Y_AVG)\n",
    "    - 5-year ratio standard deviation."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8856ec08",
   "metadata": {},
   "source": [
    "---\n",
    "\n",
    "Disclaimer: I am not an investment professional. None of my work within or related to this notebook should be considered investment advice. It is not."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ca803fd2-de97-4344-afb0-8106481bb51f",
   "metadata": {},
   "source": [
    "# Setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "3181c5f9-068c-446f-bd1c-19daab53b141",
   "metadata": {},
   "outputs": [],
   "source": [
    "import duckdb\n",
    "import os\n",
    "import pandas as pd\n",
    "pd.set_option('display.float_format', lambda x: '%.3f' % x)\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "from pathlib import Path\n",
    "import matplotlib.pyplot as plt\n",
    "plt.style.use('dark_background')\n",
    "from itertools import product\n",
    "from datetime import datetime\n",
    "import optuna\n",
    "import xgboost\n",
    "import cupy\n",
    "import cudf\n",
    "from sklearn.metrics import roc_auc_score\n",
    "\n",
    "data_path = Path(os.getenv('DATA_HOME'))\n",
    "duckdb_path = data_path.joinpath('analytics_demo/sharadar.duckdb')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "311b5449-cf62-49ca-a622-3fd06df9ea15",
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings('ignore', category=FutureWarning)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "7e5d4553-6714-436a-8dcc-c2e93f1059e7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "74b7a0de3c88471e99caf30530384616",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "FloatProgress(value=0.0, layout=Layout(width='auto'), style=ProgressStyle(bar_color='black'))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "df_targ: (1499693, 19)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>key_daily</th>\n",
       "      <th>key_dim_fin</th>\n",
       "      <th>ticker</th>\n",
       "      <th>date</th>\n",
       "      <th>mcap_daily</th>\n",
       "      <th>mcap_perc_rank</th>\n",
       "      <th>mcap_perc_rank_rev</th>\n",
       "      <th>ncfo_ttm</th>\n",
       "      <th>liabilitiesc_ttm</th>\n",
       "      <th>ratio_ttm</th>\n",
       "      <th>ratio_adj_ttm</th>\n",
       "      <th>min_ncfo_ftm</th>\n",
       "      <th>max_ncfo_ftm</th>\n",
       "      <th>min_liabilitiesc_ftm</th>\n",
       "      <th>max_liabilitiesc_ftm</th>\n",
       "      <th>min_ratio_ftm</th>\n",
       "      <th>max_ratio_ftm</th>\n",
       "      <th>min_ratio_adj_ftm</th>\n",
       "      <th>max_ratio_adj_ftm</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>AZN_2007-05-03</td>\n",
       "      <td>AZN_ART_2006-12-31_2007-03-27</td>\n",
       "      <td>AZN</td>\n",
       "      <td>2007-05-03</td>\n",
       "      <td>80282700000.000</td>\n",
       "      <td>0.011</td>\n",
       "      <td>0.989</td>\n",
       "      <td>7693000000.000</td>\n",
       "      <td>9447000000.000</td>\n",
       "      <td>1.228</td>\n",
       "      <td>1.228</td>\n",
       "      <td>6672000000.000</td>\n",
       "      <td>7693000000.000</td>\n",
       "      <td>10349000000.000</td>\n",
       "      <td>24933000000.000</td>\n",
       "      <td>1.345</td>\n",
       "      <td>3.344</td>\n",
       "      <td>1.345</td>\n",
       "      <td>3.344</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>OPHM_2003-01-06</td>\n",
       "      <td>OPHM_ART_2002-09-30_2002-11-06</td>\n",
       "      <td>OPHM</td>\n",
       "      <td>2003-01-06</td>\n",
       "      <td>99900000.000</td>\n",
       "      <td>0.547</td>\n",
       "      <td>0.452</td>\n",
       "      <td>-19886311.000</td>\n",
       "      <td>6392642.000</td>\n",
       "      <td>-0.321</td>\n",
       "      <td>0.321</td>\n",
       "      <td>-19886311.000</td>\n",
       "      <td>-19886311.000</td>\n",
       "      <td>6392642.000</td>\n",
       "      <td>6392642.000</td>\n",
       "      <td>-0.321</td>\n",
       "      <td>-0.321</td>\n",
       "      <td>0.321</td>\n",
       "      <td>0.321</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         key_daily                     key_dim_fin ticker       date  \\\n",
       "0   AZN_2007-05-03   AZN_ART_2006-12-31_2007-03-27    AZN 2007-05-03   \n",
       "1  OPHM_2003-01-06  OPHM_ART_2002-09-30_2002-11-06   OPHM 2003-01-06   \n",
       "\n",
       "       mcap_daily  mcap_perc_rank  mcap_perc_rank_rev       ncfo_ttm  \\\n",
       "0 80282700000.000           0.011               0.989 7693000000.000   \n",
       "1    99900000.000           0.547               0.452  -19886311.000   \n",
       "\n",
       "   liabilitiesc_ttm  ratio_ttm  ratio_adj_ttm   min_ncfo_ftm   max_ncfo_ftm  \\\n",
       "0    9447000000.000      1.228          1.228 6672000000.000 7693000000.000   \n",
       "1       6392642.000     -0.321          0.321  -19886311.000  -19886311.000   \n",
       "\n",
       "   min_liabilitiesc_ftm  max_liabilitiesc_ftm  min_ratio_ftm  max_ratio_ftm  \\\n",
       "0       10349000000.000       24933000000.000          1.345          3.344   \n",
       "1           6392642.000           6392642.000         -0.321         -0.321   \n",
       "\n",
       "   min_ratio_adj_ftm  max_ratio_adj_ftm  \n",
       "0              1.345              3.344  \n",
       "1              0.321              0.321  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "df_fin_art_currency: (634776, 73)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>key_dim_fin</th>\n",
       "      <th>accoci</th>\n",
       "      <th>assets</th>\n",
       "      <th>assetsavg</th>\n",
       "      <th>assetsc</th>\n",
       "      <th>assetsnc</th>\n",
       "      <th>capex</th>\n",
       "      <th>cashneq</th>\n",
       "      <th>cashnequsd</th>\n",
       "      <th>cor</th>\n",
       "      <th>...</th>\n",
       "      <th>revenue</th>\n",
       "      <th>revenueusd</th>\n",
       "      <th>rnd</th>\n",
       "      <th>sbcomp</th>\n",
       "      <th>sgna</th>\n",
       "      <th>tangibles</th>\n",
       "      <th>taxassets</th>\n",
       "      <th>taxexp</th>\n",
       "      <th>taxliabilities</th>\n",
       "      <th>workingcapital</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>CCTG_ART_2022-03-31_2023-03-22</td>\n",
       "      <td>-617288.000</td>\n",
       "      <td>16390029.000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>15478272.000</td>\n",
       "      <td>911757.000</td>\n",
       "      <td>-177639.000</td>\n",
       "      <td>5285940.000</td>\n",
       "      <td>5285940.000</td>\n",
       "      <td>19694031.000</td>\n",
       "      <td>...</td>\n",
       "      <td>27169935.000</td>\n",
       "      <td>27169935.000</td>\n",
       "      <td>829024.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>4184951.000</td>\n",
       "      <td>16273120.000</td>\n",
       "      <td>100338.000</td>\n",
       "      <td>399828.000</td>\n",
       "      <td>257246.000</td>\n",
       "      <td>8697363.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>CCTG_ART_2022-09-30_2023-07-28</td>\n",
       "      <td>-1713691.000</td>\n",
       "      <td>18841982.000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>16147238.000</td>\n",
       "      <td>2694744.000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>5901314.000</td>\n",
       "      <td>5901314.000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>18745138.000</td>\n",
       "      <td>88108.000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>399547.000</td>\n",
       "      <td>10306443.000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 73 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                      key_dim_fin       accoci       assets  assetsavg  \\\n",
       "0  CCTG_ART_2022-03-31_2023-03-22  -617288.000 16390029.000        NaN   \n",
       "1  CCTG_ART_2022-09-30_2023-07-28 -1713691.000 18841982.000        NaN   \n",
       "\n",
       "       assetsc    assetsnc       capex     cashneq  cashnequsd          cor  \\\n",
       "0 15478272.000  911757.000 -177639.000 5285940.000 5285940.000 19694031.000   \n",
       "1 16147238.000 2694744.000         NaN 5901314.000 5901314.000          NaN   \n",
       "\n",
       "   ...      revenue   revenueusd        rnd  sbcomp        sgna    tangibles  \\\n",
       "0  ... 27169935.000 27169935.000 829024.000   0.000 4184951.000 16273120.000   \n",
       "1  ...          NaN          NaN        NaN     NaN         NaN 18745138.000   \n",
       "\n",
       "   taxassets     taxexp  taxliabilities  workingcapital  \n",
       "0 100338.000 399828.000      257246.000     8697363.000  \n",
       "1  88108.000        NaN      399547.000    10306443.000  \n",
       "\n",
       "[2 rows x 73 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "df_fin_ary5y_currency: (1904105, 73)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>key_dim_fin</th>\n",
       "      <th>accoci</th>\n",
       "      <th>assets</th>\n",
       "      <th>assetsavg</th>\n",
       "      <th>assetsc</th>\n",
       "      <th>assetsnc</th>\n",
       "      <th>capex</th>\n",
       "      <th>cashneq</th>\n",
       "      <th>cashnequsd</th>\n",
       "      <th>cor</th>\n",
       "      <th>...</th>\n",
       "      <th>revenue</th>\n",
       "      <th>revenueusd</th>\n",
       "      <th>rnd</th>\n",
       "      <th>sbcomp</th>\n",
       "      <th>sgna</th>\n",
       "      <th>tangibles</th>\n",
       "      <th>taxassets</th>\n",
       "      <th>taxexp</th>\n",
       "      <th>taxliabilities</th>\n",
       "      <th>workingcapital</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>TBACQ_ART_2006-12-31_2007-02-14</td>\n",
       "      <td>-1706000.000</td>\n",
       "      <td>133437000.000</td>\n",
       "      <td>137182500.000</td>\n",
       "      <td>97829000.000</td>\n",
       "      <td>35608000.000</td>\n",
       "      <td>-2716000.000</td>\n",
       "      <td>6506000.000</td>\n",
       "      <td>6506000.000</td>\n",
       "      <td>133706000.000</td>\n",
       "      <td>...</td>\n",
       "      <td>205769000.000</td>\n",
       "      <td>205769000.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>53516000.000</td>\n",
       "      <td>115567000.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>4010000.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>78217000.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>TBACQ_ART_2006-12-31_2007-02-14</td>\n",
       "      <td>-1196000.000</td>\n",
       "      <td>146530000.000</td>\n",
       "      <td>143287000.000</td>\n",
       "      <td>113649000.000</td>\n",
       "      <td>32881000.000</td>\n",
       "      <td>-2444000.000</td>\n",
       "      <td>3814000.000</td>\n",
       "      <td>3814000.000</td>\n",
       "      <td>146399000.000</td>\n",
       "      <td>...</td>\n",
       "      <td>224487000.000</td>\n",
       "      <td>224487000.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>58450000.000</td>\n",
       "      <td>129989000.000</td>\n",
       "      <td>4757000.000</td>\n",
       "      <td>4922000.000</td>\n",
       "      <td>1776000.000</td>\n",
       "      <td>89131000.000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 73 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                       key_dim_fin       accoci        assets     assetsavg  \\\n",
       "0  TBACQ_ART_2006-12-31_2007-02-14 -1706000.000 133437000.000 137182500.000   \n",
       "1  TBACQ_ART_2006-12-31_2007-02-14 -1196000.000 146530000.000 143287000.000   \n",
       "\n",
       "        assetsc     assetsnc        capex     cashneq  cashnequsd  \\\n",
       "0  97829000.000 35608000.000 -2716000.000 6506000.000 6506000.000   \n",
       "1 113649000.000 32881000.000 -2444000.000 3814000.000 3814000.000   \n",
       "\n",
       "            cor  ...       revenue    revenueusd   rnd  sbcomp         sgna  \\\n",
       "0 133706000.000  ... 205769000.000 205769000.000 0.000   0.000 53516000.000   \n",
       "1 146399000.000  ... 224487000.000 224487000.000 0.000   0.000 58450000.000   \n",
       "\n",
       "      tangibles   taxassets      taxexp  taxliabilities  workingcapital  \n",
       "0 115567000.000       0.000 4010000.000           0.000    78217000.000  \n",
       "1 129989000.000 4757000.000 4922000.000     1776000.000    89131000.000  \n",
       "\n",
       "[2 rows x 73 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 5.25 s, sys: 2.38 s, total: 7.63 s\n",
      "Wall time: 7.24 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "con = duckdb.connect(str(duckdb_path))\n",
    "df_targ = con.execute('from fct_daily_credit_risk_targ where mcap_perc_rank_rev > 0.05').fetch_df()\n",
    "df_fin_art_currency = con.execute('from dim_fin_art_currency').fetch_df()\n",
    "df_fin_ary5y_currency = con.execute('from dim_fin_ary5y_currency').fetch_df()\n",
    "con.close()\n",
    "\n",
    "print('df_targ:', df_targ.shape)\n",
    "display(df_targ.head(2))\n",
    "print('df_fin_art_currency:', df_fin_art_currency.shape)\n",
    "display(df_fin_art_currency.head(2))\n",
    "print('df_fin_ary5y_currency:', df_fin_ary5y_currency.shape)\n",
    "display(df_fin_ary5y_currency.head(2))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4cc85ea5-5629-4bed-8cba-c8cbf86b7adf",
   "metadata": {},
   "source": [
    "# Target engineering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "e956259e-bdb6-4588-880b-117e9a60bfd9",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_targ['ratio_adj_chg'] = df_targ.max_ratio_adj_ftm - df_targ.ratio_adj_ttm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "fd18ca2a-bb2a-493b-ad21-b6bb05653ad2",
   "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>cur_liab_mil_ttm</th>\n",
       "      <th>ncfo_mil_ttm</th>\n",
       "      <th>ratio_adj_ttm</th>\n",
       "      <th>ratio_adj_chg</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0.050</th>\n",
       "      <td>1.909</td>\n",
       "      <td>-47.632</td>\n",
       "      <td>0.243</td>\n",
       "      <td>-0.295</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.150</th>\n",
       "      <td>6.709</td>\n",
       "      <td>-7.399</td>\n",
       "      <td>0.573</td>\n",
       "      <td>0.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.250</th>\n",
       "      <td>14.345</td>\n",
       "      <td>-0.217</td>\n",
       "      <td>0.896</td>\n",
       "      <td>0.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.350</th>\n",
       "      <td>28.005</td>\n",
       "      <td>4.791</td>\n",
       "      <td>1.228</td>\n",
       "      <td>0.059</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.450</th>\n",
       "      <td>52.665</td>\n",
       "      <td>14.692</td>\n",
       "      <td>1.603</td>\n",
       "      <td>0.200</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.550</th>\n",
       "      <td>98.040</td>\n",
       "      <td>36.078</td>\n",
       "      <td>2.079</td>\n",
       "      <td>0.447</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.650</th>\n",
       "      <td>183.138</td>\n",
       "      <td>79.502</td>\n",
       "      <td>2.749</td>\n",
       "      <td>0.941</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.750</th>\n",
       "      <td>380.844</td>\n",
       "      <td>174.751</td>\n",
       "      <td>3.889</td>\n",
       "      <td>2.209</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.850</th>\n",
       "      <td>980.473</td>\n",
       "      <td>444.831</td>\n",
       "      <td>6.503</td>\n",
       "      <td>7.151</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.950</th>\n",
       "      <td>5401.000</td>\n",
       "      <td>2442.000</td>\n",
       "      <td>19.547</td>\n",
       "      <td>45.697</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       cur_liab_mil_ttm  ncfo_mil_ttm  ratio_adj_ttm  ratio_adj_chg\n",
       "0.050             1.909       -47.632          0.243         -0.295\n",
       "0.150             6.709        -7.399          0.573          0.000\n",
       "0.250            14.345        -0.217          0.896          0.000\n",
       "0.350            28.005         4.791          1.228          0.059\n",
       "0.450            52.665        14.692          1.603          0.200\n",
       "0.550            98.040        36.078          2.079          0.447\n",
       "0.650           183.138        79.502          2.749          0.941\n",
       "0.750           380.844       174.751          3.889          2.209\n",
       "0.850           980.473       444.831          6.503          7.151\n",
       "0.950          5401.000      2442.000         19.547         45.697"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "quantiles_cut = np.arange(.05, 1, .1)\n",
    "_df_curr_quantiles = (df_targ[['liabilitiesc_ttm', 'ncfo_ttm']]/1e6).quantile(quantiles_cut)\n",
    "_df_curr_quantiles.columns = ['cur_liab_mil_ttm', 'ncfo_mil_ttm']\n",
    "_df_ratio_quantiles = df_targ[['ratio_adj_ttm']].quantile(quantiles_cut)\n",
    "_df_targ_quantiles = df_targ[['ratio_adj_chg']].quantile(quantiles_cut)\n",
    "_df_curr_quantiles = _df_curr_quantiles.merge(_df_ratio_quantiles, how='inner', left_index=True, right_index=True)\n",
    "_df_curr_quantiles = _df_curr_quantiles.merge(_df_targ_quantiles, how='inner', left_index=True, right_index=True)\n",
    "_df_curr_quantiles"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "53eacf1b-f2ae-4994-a8d0-d3496b3e400b",
   "metadata": {},
   "source": [
    "Well it seems like a +2 or so change in `current liabities / operating cash flow` might be pretty bad. Sounds like a pretty rough cash crunch.\n",
    "\n",
    "Consider it: Having double your TTM operating cashflow in current liabilities, at any point, doesn't sound good. Now consider that we're talking about an *increase* of that magnitude. It is a perfect mystery to me why this appears to occur in more than 25% of my samples, and the baselind TTM ratio looks pretty crazy to me too. I'd guess that many of these liabilities are be non-cash-related, such as products to be delivered within the year. However it's probably obvious that I don't know accounting, so I won't delve too deeply into this."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "2efecf93-1b41-4e67-8587-345db9de09df",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.26024482287584544"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_targ['targ'] = (df_targ.ratio_adj_chg >= 2).apply(lambda x: 1 if x else 0)\n",
    "df_targ.loc[df_targ.ratio_adj_chg.isnull(), 'targ'] = np.nan\n",
    "df_targ['targ'].mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b5dec059-8a6c-445b-bc91-63b5047ecb7a",
   "metadata": {},
   "source": [
    "# Feature Engineering"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "c177aae9-2671-4f82-a7da-af7293673a7d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(375894, 805)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>key_dim_fin</th>\n",
       "      <th>taxassets_over_assetsavg_art</th>\n",
       "      <th>taxassets_over_revenue_art</th>\n",
       "      <th>taxassets_over_fcf_art</th>\n",
       "      <th>ncfi_over_assetsavg_art</th>\n",
       "      <th>ncfi_over_revenue_art</th>\n",
       "      <th>ncfi_over_fcf_art</th>\n",
       "      <th>workingcapital_over_assetsavg_art</th>\n",
       "      <th>workingcapital_over_revenue_art</th>\n",
       "      <th>workingcapital_over_fcf_art</th>\n",
       "      <th>...</th>\n",
       "      <th>liabilities_over_fcf_art_minus_liabilities_over_fcf_ary5y_avg</th>\n",
       "      <th>sbcomp_over_assetsavg_art_minus_sbcomp_over_assetsavg_ary5y_avg</th>\n",
       "      <th>sbcomp_over_revenue_art_minus_sbcomp_over_revenue_ary5y_avg</th>\n",
       "      <th>sbcomp_over_fcf_art_minus_sbcomp_over_fcf_ary5y_avg</th>\n",
       "      <th>taxexp_over_assetsavg_art_minus_taxexp_over_assetsavg_ary5y_avg</th>\n",
       "      <th>taxexp_over_revenue_art_minus_taxexp_over_revenue_ary5y_avg</th>\n",
       "      <th>taxexp_over_fcf_art_minus_taxexp_over_fcf_ary5y_avg</th>\n",
       "      <th>netinccmn_over_assetsavg_art_minus_netinccmn_over_assetsavg_ary5y_avg</th>\n",
       "      <th>netinccmn_over_revenue_art_minus_netinccmn_over_revenue_ary5y_avg</th>\n",
       "      <th>netinccmn_over_fcf_art_minus_netinccmn_over_fcf_ary5y_avg</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>BPTSY_ART_2023-12-31_2024-04-08</td>\n",
       "      <td>0.000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-0.000</td>\n",
       "      <td>0.022</td>\n",
       "      <td>inf</td>\n",
       "      <td>-0.028</td>\n",
       "      <td>-0.207</td>\n",
       "      <td>-inf</td>\n",
       "      <td>0.268</td>\n",
       "      <td>...</td>\n",
       "      <td>0.211</td>\n",
       "      <td>-0.048</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.081</td>\n",
       "      <td>0.000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-0.000</td>\n",
       "      <td>-0.109</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-0.087</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>AABC_ART_2001-12-31_2002-03-07</td>\n",
       "      <td>0.006</td>\n",
       "      <td>0.139</td>\n",
       "      <td>0.534</td>\n",
       "      <td>-0.110</td>\n",
       "      <td>-2.652</td>\n",
       "      <td>-10.178</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>105.526</td>\n",
       "      <td>-0.000</td>\n",
       "      <td>-0.001</td>\n",
       "      <td>-0.003</td>\n",
       "      <td>-0.000</td>\n",
       "      <td>-0.014</td>\n",
       "      <td>0.158</td>\n",
       "      <td>0.001</td>\n",
       "      <td>-0.025</td>\n",
       "      <td>0.351</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>2 rows × 805 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                       key_dim_fin  taxassets_over_assetsavg_art  \\\n",
       "0  BPTSY_ART_2023-12-31_2024-04-08                         0.000   \n",
       "1   AABC_ART_2001-12-31_2002-03-07                         0.006   \n",
       "\n",
       "   taxassets_over_revenue_art  taxassets_over_fcf_art  \\\n",
       "0                         NaN                  -0.000   \n",
       "1                       0.139                   0.534   \n",
       "\n",
       "   ncfi_over_assetsavg_art  ncfi_over_revenue_art  ncfi_over_fcf_art  \\\n",
       "0                    0.022                    inf             -0.028   \n",
       "1                   -0.110                 -2.652            -10.178   \n",
       "\n",
       "   workingcapital_over_assetsavg_art  workingcapital_over_revenue_art  \\\n",
       "0                             -0.207                             -inf   \n",
       "1                                NaN                              NaN   \n",
       "\n",
       "   workingcapital_over_fcf_art  ...  \\\n",
       "0                        0.268  ...   \n",
       "1                          NaN  ...   \n",
       "\n",
       "   liabilities_over_fcf_art_minus_liabilities_over_fcf_ary5y_avg  \\\n",
       "0                                              0.211               \n",
       "1                                            105.526               \n",
       "\n",
       "   sbcomp_over_assetsavg_art_minus_sbcomp_over_assetsavg_ary5y_avg  \\\n",
       "0                                             -0.048                 \n",
       "1                                             -0.000                 \n",
       "\n",
       "   sbcomp_over_revenue_art_minus_sbcomp_over_revenue_ary5y_avg  \\\n",
       "0                                                NaN             \n",
       "1                                             -0.001             \n",
       "\n",
       "   sbcomp_over_fcf_art_minus_sbcomp_over_fcf_ary5y_avg  \\\n",
       "0                                              0.081     \n",
       "1                                             -0.003     \n",
       "\n",
       "   taxexp_over_assetsavg_art_minus_taxexp_over_assetsavg_ary5y_avg  \\\n",
       "0                                              0.000                 \n",
       "1                                             -0.000                 \n",
       "\n",
       "   taxexp_over_revenue_art_minus_taxexp_over_revenue_ary5y_avg  \\\n",
       "0                                                NaN             \n",
       "1                                             -0.014             \n",
       "\n",
       "   taxexp_over_fcf_art_minus_taxexp_over_fcf_ary5y_avg  \\\n",
       "0                                             -0.000     \n",
       "1                                              0.158     \n",
       "\n",
       "   netinccmn_over_assetsavg_art_minus_netinccmn_over_assetsavg_ary5y_avg  \\\n",
       "0                                             -0.109                       \n",
       "1                                              0.001                       \n",
       "\n",
       "   netinccmn_over_revenue_art_minus_netinccmn_over_revenue_ary5y_avg  \\\n",
       "0                                                NaN                   \n",
       "1                                             -0.025                   \n",
       "\n",
       "   netinccmn_over_fcf_art_minus_netinccmn_over_fcf_ary5y_avg  \n",
       "0                                             -0.087          \n",
       "1                                              0.351          \n",
       "\n",
       "[2 rows x 805 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "set_currency_cols = set(df_fin_art_currency.columns.drop(['key_dim_fin', 'marketcap', 'ev']))\n",
    "list_currency_cols = sorted(list(set_currency_cols))\n",
    "set_currency_cols_denom = set(['assetsavg', 'fcf', 'revenue'])\n",
    "set_currency_cols_num = set_currency_cols.difference(set_currency_cols_denom)\n",
    "\n",
    "# For now, just pitch rows with NULL demoniators\n",
    "for c in set_currency_cols_denom:\n",
    "    df_fin_art_currency = df_fin_art_currency[~df_fin_art_currency[c].isnull()].copy()\n",
    "    df_fin_ary5y_currency = df_fin_ary5y_currency[~df_fin_ary5y_currency[c].isnull()].copy()\n",
    "\n",
    "# Create ratios within TTM and 5-year tables\n",
    "df_art_cur_ratios = {'key_dim_fin': df_fin_art_currency.key_dim_fin}\n",
    "df_ary5y_cur_ratios = {'key_dim_fin': df_fin_ary5y_currency.key_dim_fin}\n",
    "df_ary5y_cur_ratio_std = {'key_dim_fin': df_fin_ary5y_currency.key_dim_fin}\n",
    "\n",
    "art_ratio_colnames = []\n",
    "ary5y_ratio_colnames = []\n",
    "\n",
    "for a in set_currency_cols_num:\n",
    "    for b in set_currency_cols_denom:\n",
    "        c = a + '_over_' + b + '_art'\n",
    "        art_ratio_colnames.append(c)\n",
    "        df_art_cur_ratios[c] = df_fin_art_currency[a] / df_fin_art_currency[b]\n",
    "\n",
    "        d = a + '_over_' + b + '_ary5y_avg'\n",
    "        ary5y_ratio_colnames.append(d)\n",
    "        df_ary5y_cur_ratios[d] = df_fin_ary5y_currency[a] / df_fin_ary5y_currency[b]\n",
    "\n",
    "        e = a + '_over_' + b + '_ary5y_std'\n",
    "        df_ary5y_cur_ratio_std[e] = df_fin_ary5y_currency[a] / df_fin_ary5y_currency[b]\n",
    "        \n",
    "df_art_cur_ratios = pd.DataFrame(df_art_cur_ratios)\n",
    "df_ary5y_cur_ratios = pd.DataFrame(df_ary5y_cur_ratios).groupby('key_dim_fin').mean().reset_index()\n",
    "df_ary5y_cur_ratio_std = pd.DataFrame(df_ary5y_cur_ratio_std).groupby('key_dim_fin').std().reset_index()\n",
    "df_ratio_features = df_art_cur_ratios.merge(df_ary5y_cur_ratios, how='inner', on='key_dim_fin')\n",
    "df_ratio_features = df_ratio_features.merge(df_ary5y_cur_ratio_std, how='inner', on='key_dim_fin')\n",
    "\n",
    "# Capture changes between TTM and 5-year features\n",
    "df_feat_change = {'key_dim_fin': df_ratio_features.key_dim_fin}\n",
    "for a, b in zip(art_ratio_colnames, ary5y_ratio_colnames):\n",
    "    df_feat_change[a + '_minus_' + b] = df_ratio_features[a] - df_ratio_features[b]\n",
    "df_feat_change = pd.DataFrame(df_feat_change)\n",
    "df_ratio_features = df_ratio_features.merge(df_feat_change, how='inner', on='key_dim_fin')\n",
    "\n",
    "print(df_ratio_features.shape)\n",
    "display(df_ratio_features.head(2))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5811f7f1-92d2-471a-96ae-ef5b2a499302",
   "metadata": {},
   "source": [
    "# Export"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "14ca2f6c-101a-4f6a-a95e-33c5dfa482b1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(1025905, 807)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>key_daily</th>\n",
       "      <th>date</th>\n",
       "      <th>targ</th>\n",
       "      <th>taxassets_over_assetsavg_art</th>\n",
       "      <th>taxassets_over_revenue_art</th>\n",
       "      <th>taxassets_over_fcf_art</th>\n",
       "      <th>ncfi_over_assetsavg_art</th>\n",
       "      <th>ncfi_over_revenue_art</th>\n",
       "      <th>ncfi_over_fcf_art</th>\n",
       "      <th>workingcapital_over_assetsavg_art</th>\n",
       "      <th>...</th>\n",
       "      <th>liabilities_over_fcf_art_minus_liabilities_over_fcf_ary5y_avg</th>\n",
       "      <th>sbcomp_over_assetsavg_art_minus_sbcomp_over_assetsavg_ary5y_avg</th>\n",
       "      <th>sbcomp_over_revenue_art_minus_sbcomp_over_revenue_ary5y_avg</th>\n",
       "      <th>sbcomp_over_fcf_art_minus_sbcomp_over_fcf_ary5y_avg</th>\n",
       "      <th>taxexp_over_assetsavg_art_minus_taxexp_over_assetsavg_ary5y_avg</th>\n",
       "      <th>taxexp_over_revenue_art_minus_taxexp_over_revenue_ary5y_avg</th>\n",
       "      <th>taxexp_over_fcf_art_minus_taxexp_over_fcf_ary5y_avg</th>\n",
       "      <th>netinccmn_over_assetsavg_art_minus_netinccmn_over_assetsavg_ary5y_avg</th>\n",
       "      <th>netinccmn_over_revenue_art_minus_netinccmn_over_revenue_ary5y_avg</th>\n",
       "      <th>netinccmn_over_fcf_art_minus_netinccmn_over_fcf_ary5y_avg</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>AZN_2007-05-03</td>\n",
       "      <td>2007-05-03</td>\n",
       "      <td>1.000</td>\n",
       "      <td>0.093</td>\n",
       "      <td>0.098</td>\n",
       "      <td>0.373</td>\n",
       "      <td>-0.010</td>\n",
       "      <td>-0.010</td>\n",
       "      <td>-0.039</td>\n",
       "      <td>0.268</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.456</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.022</td>\n",
       "      <td>0.020</td>\n",
       "      <td>0.013</td>\n",
       "      <td>0.042</td>\n",
       "      <td>0.037</td>\n",
       "      <td>-0.031</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>PLL1_2003-01-06</td>\n",
       "      <td>2003-01-06</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>-0.201</td>\n",
       "      <td>-0.271</td>\n",
       "      <td>-3.200</td>\n",
       "      <td>0.261</td>\n",
       "      <td>...</td>\n",
       "      <td>0.227</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>-0.007</td>\n",
       "      <td>-0.007</td>\n",
       "      <td>-0.083</td>\n",
       "      <td>-0.049</td>\n",
       "      <td>-0.058</td>\n",
       "      <td>-0.731</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>PTZ_2003-01-06</td>\n",
       "      <td>2003-01-06</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>-0.158</td>\n",
       "      <td>-0.493</td>\n",
       "      <td>-3.331</td>\n",
       "      <td>0.161</td>\n",
       "      <td>...</td>\n",
       "      <td>1.380</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>-0.007</td>\n",
       "      <td>-0.001</td>\n",
       "      <td>-0.183</td>\n",
       "      <td>-0.027</td>\n",
       "      <td>-0.018</td>\n",
       "      <td>-0.433</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>RDC_2003-01-06</td>\n",
       "      <td>2003-01-06</td>\n",
       "      <td>1.000</td>\n",
       "      <td>0.001</td>\n",
       "      <td>0.003</td>\n",
       "      <td>-0.042</td>\n",
       "      <td>-0.130</td>\n",
       "      <td>-0.426</td>\n",
       "      <td>5.168</td>\n",
       "      <td>0.174</td>\n",
       "      <td>...</td>\n",
       "      <td>-20.057</td>\n",
       "      <td>-0.000</td>\n",
       "      <td>0.002</td>\n",
       "      <td>-0.184</td>\n",
       "      <td>-0.004</td>\n",
       "      <td>0.019</td>\n",
       "      <td>-2.387</td>\n",
       "      <td>-0.006</td>\n",
       "      <td>0.039</td>\n",
       "      <td>-4.330</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>SMSI_2003-01-06</td>\n",
       "      <td>2003-01-06</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>-0.002</td>\n",
       "      <td>-0.001</td>\n",
       "      <td>-0.117</td>\n",
       "      <td>0.456</td>\n",
       "      <td>...</td>\n",
       "      <td>17.074</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.000</td>\n",
       "      <td>-0.130</td>\n",
       "      <td>-0.096</td>\n",
       "      <td>-11.885</td>\n",
       "      <td>0.222</td>\n",
       "      <td>0.312</td>\n",
       "      <td>-6.787</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 807 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "         key_daily       date  targ  taxassets_over_assetsavg_art  \\\n",
       "0   AZN_2007-05-03 2007-05-03 1.000                         0.093   \n",
       "1  PLL1_2003-01-06 2003-01-06 0.000                         0.000   \n",
       "2   PTZ_2003-01-06 2003-01-06 0.000                         0.000   \n",
       "3   RDC_2003-01-06 2003-01-06 1.000                         0.001   \n",
       "4  SMSI_2003-01-06 2003-01-06 0.000                         0.000   \n",
       "\n",
       "   taxassets_over_revenue_art  taxassets_over_fcf_art  \\\n",
       "0                       0.098                   0.373   \n",
       "1                       0.000                   0.000   \n",
       "2                       0.000                   0.000   \n",
       "3                       0.003                  -0.042   \n",
       "4                       0.000                   0.000   \n",
       "\n",
       "   ncfi_over_assetsavg_art  ncfi_over_revenue_art  ncfi_over_fcf_art  \\\n",
       "0                   -0.010                 -0.010             -0.039   \n",
       "1                   -0.201                 -0.271             -3.200   \n",
       "2                   -0.158                 -0.493             -3.331   \n",
       "3                   -0.130                 -0.426              5.168   \n",
       "4                   -0.002                 -0.001             -0.117   \n",
       "\n",
       "   workingcapital_over_assetsavg_art  ...  \\\n",
       "0                              0.268  ...   \n",
       "1                              0.261  ...   \n",
       "2                              0.161  ...   \n",
       "3                              0.174  ...   \n",
       "4                              0.456  ...   \n",
       "\n",
       "   liabilities_over_fcf_art_minus_liabilities_over_fcf_ary5y_avg  \\\n",
       "0                                             -0.456               \n",
       "1                                              0.227               \n",
       "2                                              1.380               \n",
       "3                                            -20.057               \n",
       "4                                             17.074               \n",
       "\n",
       "   sbcomp_over_assetsavg_art_minus_sbcomp_over_assetsavg_ary5y_avg  \\\n",
       "0                                              0.000                 \n",
       "1                                              0.000                 \n",
       "2                                              0.000                 \n",
       "3                                             -0.000                 \n",
       "4                                              0.000                 \n",
       "\n",
       "   sbcomp_over_revenue_art_minus_sbcomp_over_revenue_ary5y_avg  \\\n",
       "0                                              0.000             \n",
       "1                                              0.000             \n",
       "2                                              0.000             \n",
       "3                                              0.002             \n",
       "4                                              0.000             \n",
       "\n",
       "   sbcomp_over_fcf_art_minus_sbcomp_over_fcf_ary5y_avg  \\\n",
       "0                                              0.000     \n",
       "1                                              0.000     \n",
       "2                                              0.000     \n",
       "3                                             -0.184     \n",
       "4                                              0.000     \n",
       "\n",
       "   taxexp_over_assetsavg_art_minus_taxexp_over_assetsavg_ary5y_avg  \\\n",
       "0                                              0.022                 \n",
       "1                                             -0.007                 \n",
       "2                                             -0.007                 \n",
       "3                                             -0.004                 \n",
       "4                                             -0.130                 \n",
       "\n",
       "   taxexp_over_revenue_art_minus_taxexp_over_revenue_ary5y_avg  \\\n",
       "0                                              0.020             \n",
       "1                                             -0.007             \n",
       "2                                             -0.001             \n",
       "3                                              0.019             \n",
       "4                                             -0.096             \n",
       "\n",
       "   taxexp_over_fcf_art_minus_taxexp_over_fcf_ary5y_avg  \\\n",
       "0                                              0.013     \n",
       "1                                             -0.083     \n",
       "2                                             -0.183     \n",
       "3                                             -2.387     \n",
       "4                                            -11.885     \n",
       "\n",
       "   netinccmn_over_assetsavg_art_minus_netinccmn_over_assetsavg_ary5y_avg  \\\n",
       "0                                              0.042                       \n",
       "1                                             -0.049                       \n",
       "2                                             -0.027                       \n",
       "3                                             -0.006                       \n",
       "4                                              0.222                       \n",
       "\n",
       "   netinccmn_over_revenue_art_minus_netinccmn_over_revenue_ary5y_avg  \\\n",
       "0                                              0.037                   \n",
       "1                                             -0.058                   \n",
       "2                                             -0.018                   \n",
       "3                                              0.039                   \n",
       "4                                              0.312                   \n",
       "\n",
       "   netinccmn_over_fcf_art_minus_netinccmn_over_fcf_ary5y_avg  \n",
       "0                                             -0.031          \n",
       "1                                             -0.731          \n",
       "2                                             -0.433          \n",
       "3                                             -4.330          \n",
       "4                                             -6.787          \n",
       "\n",
       "[5 rows x 807 columns]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = df_targ[['key_daily', 'date', 'targ', 'key_dim_fin']].copy()\n",
    "df = df.merge(df_ratio_features, how='inner', on='key_dim_fin').drop(columns='key_dim_fin')\n",
    "print(df.shape)\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "837bbc20-064a-4811-b7ec-3320f835b5ac",
   "metadata": {},
   "outputs": [],
   "source": [
    "# reduce size\n",
    "list_cols_feat = df.columns.drop(['key_daily', 'date', 'targ']).tolist()\n",
    "df[list_cols_feat] = df[list_cols_feat].astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "3ac52926-5faf-4267-83b4-15c078cd440e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "3ef2b2af56dc4a068064f200a7a437b6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "FloatProgress(value=0.0, layout=Layout(width='auto'), style=ProgressStyle(bar_color='black'))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "con = duckdb.connect(str(duckdb_path))\n",
    "con.sql(\"CREATE OR REPLACE TABLE df_cash_crunch AS SELECT * FROM df\")\n",
    "con.close()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e86938e2-e4a7-4e22-8038-b4199d294881",
   "metadata": {},
   "source": [
    "# EDA & Final Prep"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a7c0285a-bd17-4836-a6fc-2301895f78ab",
   "metadata": {},
   "source": [
    "I'm going to use an expanding window for tuning, training through year X, skipping year X+1 (since our target is 1-year), and predicting several years forward to lower the model count during tuning.\n",
    "\n",
    "![Time Series ML Folds](https://i.imgur.com/OfHh4qF.png)\n",
    "\n",
    "Tasks:\n",
    "- Impute +/- inf.\n",
    "- Drop features with too many nulls.\n",
    "- Find a \"starting year\" for the first training window (\"TW1\").\n",
    "- To speed up tuning, use TW1 to build a correlation matrix and drop highly-correlated features."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "1d35d8cf-87b4-4ebf-a274-1e1cf88fcb87",
   "metadata": {},
   "outputs": [],
   "source": [
    "# I'll be using XGBoost, which can handle nulls. Since this is a demo, I'll simply set inf to null.\n",
    "df = df.replace([-np.inf, np.inf], np.nan)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "ca395a62-3623-4509-8d34-f3103e880f43",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cols_feat = df.columns.drop(['key_daily', 'date', 'targ']).tolist()\n",
    "\n",
    "df_frac_nullity = df[cols_feat].isnull().mean()\n",
    "df_frac_nullity.hist(bins=20)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8fa430d4-cd81-4919-8bc6-61a14dd88b7e",
   "metadata": {},
   "source": [
    "OK none of these are that bad. I'll keep them all."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "3e046d55-dffb-471f-b612-6f2dfd3c0b52",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: xlabel='date'>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "date\n",
       "1998       55\n",
       "1999      864\n",
       "2000     1043\n",
       "2001    13417\n",
       "2002    37604\n",
       "2003    48777\n",
       "2004    50688\n",
       "2005    51083\n",
       "2006    50014\n",
       "2007    49284\n",
       "2008    48478\n",
       "Name: count, dtype: int64"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Compute daily and yearly row counts\n",
    "display(df.date.value_counts().sort_index().plot())\n",
    "display(df.date.dt.year.value_counts().sort_index().head(11))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "586a40ad-88ee-43f6-91d4-a082d9861951",
   "metadata": {},
   "source": [
    "Let's extend TW1 through 2007, which will give it 5 years with full market data. Below I create train/predict folds in a way that gives prediction coverage from 2009 trough last year without the need to train one model for each year."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "464a1880-0498-4328-b48c-8b99606ad6e1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[..., 2007] --> [2009, 2010, 2011, 2012]\n",
      "[..., 2011] --> [2013, 2014, 2015, 2016]\n",
      "[..., 2015] --> [2017, 2018, 2019, 2020]\n",
      "[..., 2019] --> [2021, 2022, 2023, 2024]\n"
     ]
    }
   ],
   "source": [
    "# Store training window partition\n",
    "first_window_final_year = 2007\n",
    "years_btw_windows = 4\n",
    "\n",
    "last_year = df.date.max().year\n",
    "train_window_limits = list(np.arange(first_window_final_year, last_year - 3, years_btw_windows))\n",
    "pred_window_lists = [list(range(x + 2, x + years_btw_windows + 2)) for x in train_window_limits]\n",
    "\n",
    "for ty, py in zip(train_window_limits, pred_window_lists):\n",
    "    print('[..., {}]'.format(ty), '-->', py)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "d42e77db-f5bb-44e8-a9c3-935985821c2a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Will drop 397 of 804 features.\n",
      "CPU times: user 15.9 s, sys: 372 ms, total: 16.2 s\n",
      "Wall time: 16.2 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# For variable pairs with this absolute correlation or greater, keep only one.\n",
    "max_corr = 0.95\n",
    "\n",
    "# Sample within the first train window then compute correlation matrix.\n",
    "tw1_limit = min(train_window_limits)\n",
    "X_corr = df.loc[df.date.dt.year <= tw1_limit, cols_feat].sample(n=10000).corr()\n",
    "\n",
    "# Convert to upper triangle then find list of columns with strong correlation.\n",
    "X_corr = X_corr.where(np.triu(np.ones(X_corr.shape), k=1).astype(bool))\n",
    "correlated_feat_drop = [c for c in X_corr.columns if any(X_corr[c].abs() > max_corr)]\n",
    "print(\"Will drop {} of {} features.\".format(str(len(correlated_feat_drop)), str(len(cols_feat))))\n",
    "\n",
    "cols_feat_keep_low_corr = list(set(X_corr.columns).difference(set(correlated_feat_drop)))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7dc2a624-bd96-48c7-a989-f4f9d17bb588",
   "metadata": {},
   "source": [
    "# Tune ML\n",
    "\n",
    "For hyperparameter tuning, rather than using just a grid *or* an optimizer, I like to force a basic grid search *then* run an optimizer. Otherwise they seem prone to getting stuck at local optima. Optuna's `enqueue_trial()` allows for just this."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "d62b43ba-727b-4b7a-bd45-7ca1bc53a77c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(792508, 410)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cudf_cash_crunch = cudf.from_pandas(df.loc[~df.targ.isnull(), ['key_daily', 'date', 'targ'] + cols_feat_keep_low_corr])\n",
    "cudf_cash_crunch.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "40db8e27-8942-45ac-ae0f-267e93304112",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create boolean masks, which I call selectors, for train and prediction windows.\n",
    "train_window_selectors = [cudf_cash_crunch.date.dt.year <= cutoff_yr for cutoff_yr in train_window_limits]\n",
    "prediction_window_selectors = [cudf_cash_crunch.date.dt.year.isin(pwl) for pwl in pred_window_lists]\n",
    "window_selectors = list(zip(train_window_selectors, prediction_window_selectors))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "0fab714f-88e3-42bc-9377-0f6bfaa15e45",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "96 models in grid.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>n_estimators</th>\n",
       "      <th>learning_rate</th>\n",
       "      <th>max_depth</th>\n",
       "      <th>colsample_bytree</th>\n",
       "      <th>subsample</th>\n",
       "      <th>min_child_weight</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>25</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "      <td>0.100</td>\n",
       "      <td>0.100</td>\n",
       "      <td>50</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>25</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "      <td>0.100</td>\n",
       "      <td>0.100</td>\n",
       "      <td>500</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>25</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "      <td>0.100</td>\n",
       "      <td>0.100</td>\n",
       "      <td>5000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>25</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "      <td>0.100</td>\n",
       "      <td>0.500</td>\n",
       "      <td>50</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>25</td>\n",
       "      <td>0.001</td>\n",
       "      <td>3</td>\n",
       "      <td>0.100</td>\n",
       "      <td>0.500</td>\n",
       "      <td>500</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   n_estimators  learning_rate  max_depth  colsample_bytree  subsample  \\\n",
       "0            25          0.001          3             0.100      0.100   \n",
       "1            25          0.001          3             0.100      0.100   \n",
       "2            25          0.001          3             0.100      0.100   \n",
       "3            25          0.001          3             0.100      0.500   \n",
       "4            25          0.001          3             0.100      0.500   \n",
       "\n",
       "   min_child_weight  \n",
       "0                50  \n",
       "1               500  \n",
       "2              5000  \n",
       "3                50  \n",
       "4               500  "
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "param_labels = ['n_estimators', 'learning_rate', 'max_depth', 'colsample_bytree', 'subsample', 'min_child_weight']\n",
    "\n",
    "grid_n_estimators = [25, 250]\n",
    "grid_learning_rate = [0.001, 0.01]\n",
    "grid_max_depth = [3, 10]\n",
    "grid_colsample_bytree = [0.1, 0.5]\n",
    "grid_subsample = grid_colsample_bytree\n",
    "grid_min_child_weight = [50, 500, 5000]\n",
    "\n",
    "grid = product(grid_n_estimators, grid_learning_rate, grid_max_depth, grid_colsample_bytree, grid_subsample, grid_min_child_weight)\n",
    "grid = [{param_labels[i]: g[i] for i in range(len(param_labels))} for g in grid]\n",
    "print(len(grid), 'models in grid.')\n",
    "pd.DataFrame(grid).head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "44e19835-e450-4062-b179-6f42740e8a3b",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-06-26 12:34:01,319] A new study created in memory with name: no-name-dfb9858c-96bf-4ef0-9f71-6c73948ed90c\n"
     ]
    }
   ],
   "source": [
    "study = optuna.create_study(direction='maximize')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "41319739-65bd-4633-a80e-7c89204c2b44",
   "metadata": {},
   "outputs": [],
   "source": [
    "for _t in grid:\n",
    "    study.enqueue_trial(_t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "97045cce-d8ca-4b44-a53b-6a9fc1ba6cd4",
   "metadata": {},
   "outputs": [],
   "source": [
    "def run_trial(trial):\n",
    "    global cudf_cash_crunch\n",
    "    cudf_cash_crunch['pred'] = cupy.nan\n",
    "    \n",
    "    model_params = {\n",
    "        'device': 'cuda',  # For xgboost>=2.0.0\n",
    "        'tree_method': 'hist',  # For xgboost>=2.0.0\n",
    "        'objective': 'binary:logistic',\n",
    "        'n_estimators': trial.suggest_int('n_estimators', 25, 500),\n",
    "        'learning_rate': trial.suggest_float('learning_rate', 0.00001, 10),\n",
    "        'max_depth': trial.suggest_int('max_depth', 1, 15),\n",
    "        'colsample_bytree': trial.suggest_float('colsample_bytree', 0.01,1.0),\n",
    "        'subsample': trial.suggest_float('subsample', 0.01, 1.0),\n",
    "        'min_child_weight': trial.suggest_int('min_child_weight', 1, 100000)  # leaf sample cnt\n",
    "    }\n",
    "   \n",
    "    for sel_train, sel_pred in window_selectors:\n",
    "        clf = xgboost.XGBClassifier(**model_params)\n",
    "        clf.fit(X=cudf_cash_crunch.loc[sel_train, cols_feat_keep_low_corr], y=cudf_cash_crunch.loc[sel_train, 'targ'])\n",
    "        cudf_cash_crunch.loc[sel_pred, 'pred'] = clf.predict_proba(X=cudf_cash_crunch.loc[sel_pred, cols_feat_keep_low_corr])[:, 1]\n",
    "\n",
    "    df_score = cudf_cash_crunch.loc[~cudf_cash_crunch.pred.isnull(), ['targ', 'pred']].to_pandas()\n",
    "    return roc_auc_score(df_score.targ, df_score.pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e33d154e-6d3d-4340-b0b5-0cab6c349b5a",
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "study.optimize(run_trial, 200)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "07af12d2-abb2-4c66-8500-4df704cd4c65",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style type=\"text/css\">\n",
       "#T_880ca_row0_col0, #T_880ca_row3_col6, #T_880ca_row4_col6, #T_880ca_row5_col3, #T_880ca_row6_col3, #T_880ca_row7_col3, #T_880ca_row9_col3, #T_880ca_row9_col5, #T_880ca_row10_col3, #T_880ca_row11_col2, #T_880ca_row11_col3, #T_880ca_row12_col1, #T_880ca_row12_col3, #T_880ca_row14_col3, #T_880ca_row15_col1, #T_880ca_row15_col3, #T_880ca_row18_col3, #T_880ca_row18_col4 {\n",
       "  background-color: #b40426;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row0_col1, #T_880ca_row1_col1, #T_880ca_row13_col1, #T_880ca_row15_col4 {\n",
       "  background-color: #d1dae9;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row0_col2, #T_880ca_row0_col3, #T_880ca_row0_col4, #T_880ca_row0_col5, #T_880ca_row1_col2, #T_880ca_row1_col3, #T_880ca_row1_col4, #T_880ca_row1_col5, #T_880ca_row1_col6, #T_880ca_row2_col1, #T_880ca_row2_col2, #T_880ca_row2_col3, #T_880ca_row2_col4, #T_880ca_row2_col5, #T_880ca_row13_col2, #T_880ca_row13_col3, #T_880ca_row13_col5, #T_880ca_row19_col0 {\n",
       "  background-color: #3b4cc0;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row0_col6, #T_880ca_row2_col6, #T_880ca_row13_col6 {\n",
       "  background-color: #cedaeb;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row1_col0 {\n",
       "  background-color: #e4d9d2;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row2_col0 {\n",
       "  background-color: #d9dce1;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row3_col0 {\n",
       "  background-color: #a5c3fe;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row3_col1, #T_880ca_row5_col5 {\n",
       "  background-color: #d75445;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row3_col2 {\n",
       "  background-color: #f6a283;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row3_col3, #T_880ca_row4_col3, #T_880ca_row8_col3, #T_880ca_row16_col3, #T_880ca_row17_col3, #T_880ca_row19_col3 {\n",
       "  background-color: #ee8468;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row3_col4, #T_880ca_row9_col2, #T_880ca_row14_col2 {\n",
       "  background-color: #e57058;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row3_col5, #T_880ca_row15_col5 {\n",
       "  background-color: #dd5f4b;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row4_col0 {\n",
       "  background-color: #9fbfff;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row4_col1 {\n",
       "  background-color: #e8765c;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row4_col2 {\n",
       "  background-color: #f5a081;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row4_col4, #T_880ca_row7_col1, #T_880ca_row9_col1, #T_880ca_row9_col6 {\n",
       "  background-color: #c83836;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row4_col5 {\n",
       "  background-color: #df634e;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row5_col0 {\n",
       "  background-color: #93b5fe;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row5_col1, #T_880ca_row10_col2 {\n",
       "  background-color: #cf453c;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row5_col2 {\n",
       "  background-color: #dc5d4a;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row5_col4 {\n",
       "  background-color: #f7b396;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row5_col6 {\n",
       "  background-color: #d65244;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row6_col0 {\n",
       "  background-color: #8badfd;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row6_col1 {\n",
       "  background-color: #e97a5f;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row6_col2 {\n",
       "  background-color: #f08a6c;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row6_col4 {\n",
       "  background-color: #ecd3c5;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row6_col5 {\n",
       "  background-color: #bd1f2d;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row6_col6 {\n",
       "  background-color: #eb7d62;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row7_col0 {\n",
       "  background-color: #84a7fc;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row7_col2 {\n",
       "  background-color: #e0654f;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row7_col4 {\n",
       "  background-color: #f7ad90;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row7_col5, #T_880ca_row8_col2 {\n",
       "  background-color: #e7745b;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row7_col6 {\n",
       "  background-color: #f29274;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row8_col0 {\n",
       "  background-color: #7da0f9;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row8_col1, #T_880ca_row14_col1 {\n",
       "  background-color: #de614d;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row8_col4 {\n",
       "  background-color: #f5c2aa;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row8_col5, #T_880ca_row10_col5, #T_880ca_row14_col4, #T_880ca_row16_col1 {\n",
       "  background-color: #ba162b;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row8_col6 {\n",
       "  background-color: #ca3b37;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row9_col0 {\n",
       "  background-color: #6c8ff1;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row9_col4 {\n",
       "  background-color: #ead5c9;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row10_col0 {\n",
       "  background-color: #6788ee;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row10_col1 {\n",
       "  background-color: #b70d28;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row10_col4, #T_880ca_row17_col2 {\n",
       "  background-color: #c43032;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row10_col6, #T_880ca_row12_col2 {\n",
       "  background-color: #c73635;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row11_col0 {\n",
       "  background-color: #6687ed;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row11_col1, #T_880ca_row18_col1 {\n",
       "  background-color: #cb3e38;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row11_col4, #T_880ca_row17_col5 {\n",
       "  background-color: #f7ac8e;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row11_col5, #T_880ca_row14_col5 {\n",
       "  background-color: #d85646;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row11_col6, #T_880ca_row15_col2, #T_880ca_row16_col2, #T_880ca_row18_col2 {\n",
       "  background-color: #e26952;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row12_col0 {\n",
       "  background-color: #6384eb;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row12_col4 {\n",
       "  background-color: #f7b497;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row12_col5 {\n",
       "  background-color: #cd423b;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row12_col6 {\n",
       "  background-color: #f7a688;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row13_col0 {\n",
       "  background-color: #6282ea;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row13_col4 {\n",
       "  background-color: #4a63d3;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row14_col0 {\n",
       "  background-color: #6180e9;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row14_col6 {\n",
       "  background-color: #d0473d;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row15_col0 {\n",
       "  background-color: #5673e0;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row15_col6 {\n",
       "  background-color: #f7bca1;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row16_col0 {\n",
       "  background-color: #4e68d8;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row16_col4 {\n",
       "  background-color: #e8d6cc;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row16_col5, #T_880ca_row16_col6 {\n",
       "  background-color: #b8122a;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row17_col0, #T_880ca_row18_col0 {\n",
       "  background-color: #4b64d5;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row17_col1 {\n",
       "  background-color: #f7aa8c;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row17_col4 {\n",
       "  background-color: #adc9fd;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row17_col6 {\n",
       "  background-color: #c6d6f1;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row18_col5 {\n",
       "  background-color: #c32e31;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row18_col6 {\n",
       "  background-color: #e36c55;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row19_col1 {\n",
       "  background-color: #f59f80;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row19_col2 {\n",
       "  background-color: #be242e;\n",
       "  color: #f1f1f1;\n",
       "}\n",
       "#T_880ca_row19_col4 {\n",
       "  background-color: #aec9fc;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row19_col5 {\n",
       "  background-color: #f4c5ad;\n",
       "  color: #000000;\n",
       "}\n",
       "#T_880ca_row19_col6 {\n",
       "  background-color: #a6c4fe;\n",
       "  color: #000000;\n",
       "}\n",
       "</style>\n",
       "<table id=\"T_880ca\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th class=\"blank level0\" >&nbsp;</th>\n",
       "      <th id=\"T_880ca_level0_col0\" class=\"col_heading level0 col0\" >roc_auc</th>\n",
       "      <th id=\"T_880ca_level0_col1\" class=\"col_heading level0 col1\" >params_colsample_bytree</th>\n",
       "      <th id=\"T_880ca_level0_col2\" class=\"col_heading level0 col2\" >params_learning_rate</th>\n",
       "      <th id=\"T_880ca_level0_col3\" class=\"col_heading level0 col3\" >params_max_depth</th>\n",
       "      <th id=\"T_880ca_level0_col4\" class=\"col_heading level0 col4\" >params_min_child_weight</th>\n",
       "      <th id=\"T_880ca_level0_col5\" class=\"col_heading level0 col5\" >params_n_estimators</th>\n",
       "      <th id=\"T_880ca_level0_col6\" class=\"col_heading level0 col6\" >params_subsample</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row0\" class=\"row_heading level0 row0\" >93</th>\n",
       "      <td id=\"T_880ca_row0_col0\" class=\"data row0 col0\" >0.80828</td>\n",
       "      <td id=\"T_880ca_row0_col1\" class=\"data row0 col1\" >0.50000</td>\n",
       "      <td id=\"T_880ca_row0_col2\" class=\"data row0 col2\" >0.01000</td>\n",
       "      <td id=\"T_880ca_row0_col3\" class=\"data row0 col3\" >10</td>\n",
       "      <td id=\"T_880ca_row0_col4\" class=\"data row0 col4\" >50</td>\n",
       "      <td id=\"T_880ca_row0_col5\" class=\"data row0 col5\" >250</td>\n",
       "      <td id=\"T_880ca_row0_col6\" class=\"data row0 col6\" >0.50000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row1\" class=\"row_heading level0 row1\" >90</th>\n",
       "      <td id=\"T_880ca_row1_col0\" class=\"data row1 col0\" >0.80421</td>\n",
       "      <td id=\"T_880ca_row1_col1\" class=\"data row1 col1\" >0.50000</td>\n",
       "      <td id=\"T_880ca_row1_col2\" class=\"data row1 col2\" >0.01000</td>\n",
       "      <td id=\"T_880ca_row1_col3\" class=\"data row1 col3\" >10</td>\n",
       "      <td id=\"T_880ca_row1_col4\" class=\"data row1 col4\" >50</td>\n",
       "      <td id=\"T_880ca_row1_col5\" class=\"data row1 col5\" >250</td>\n",
       "      <td id=\"T_880ca_row1_col6\" class=\"data row1 col6\" >0.10000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row2\" class=\"row_heading level0 row2\" >87</th>\n",
       "      <td id=\"T_880ca_row2_col0\" class=\"data row2 col0\" >0.80383</td>\n",
       "      <td id=\"T_880ca_row2_col1\" class=\"data row2 col1\" >0.10000</td>\n",
       "      <td id=\"T_880ca_row2_col2\" class=\"data row2 col2\" >0.01000</td>\n",
       "      <td id=\"T_880ca_row2_col3\" class=\"data row2 col3\" >10</td>\n",
       "      <td id=\"T_880ca_row2_col4\" class=\"data row2 col4\" >50</td>\n",
       "      <td id=\"T_880ca_row2_col5\" class=\"data row2 col5\" >250</td>\n",
       "      <td id=\"T_880ca_row2_col6\" class=\"data row2 col6\" >0.50000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row3\" class=\"row_heading level0 row3\" >197</th>\n",
       "      <td id=\"T_880ca_row3_col0\" class=\"data row3 col0\" >0.80237</td>\n",
       "      <td id=\"T_880ca_row3_col1\" class=\"data row3 col1\" >0.88820</td>\n",
       "      <td id=\"T_880ca_row3_col2\" class=\"data row3 col2\" >0.19489</td>\n",
       "      <td id=\"T_880ca_row3_col3\" class=\"data row3 col3\" >14</td>\n",
       "      <td id=\"T_880ca_row3_col4\" class=\"data row3 col4\" >7242</td>\n",
       "      <td id=\"T_880ca_row3_col5\" class=\"data row3 col5\" >456</td>\n",
       "      <td id=\"T_880ca_row3_col6\" class=\"data row3 col6\" >0.99143</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row4\" class=\"row_heading level0 row4\" >195</th>\n",
       "      <td id=\"T_880ca_row4_col0\" class=\"data row4 col0\" >0.80225</td>\n",
       "      <td id=\"T_880ca_row4_col1\" class=\"data row4 col1\" >0.83032</td>\n",
       "      <td id=\"T_880ca_row4_col2\" class=\"data row4 col2\" >0.19648</td>\n",
       "      <td id=\"T_880ca_row4_col3\" class=\"data row4 col3\" >14</td>\n",
       "      <td id=\"T_880ca_row4_col4\" class=\"data row4 col4\" >8112</td>\n",
       "      <td id=\"T_880ca_row4_col5\" class=\"data row4 col5\" >454</td>\n",
       "      <td id=\"T_880ca_row4_col6\" class=\"data row4 col6\" >0.99080</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row5\" class=\"row_heading level0 row5\" >176</th>\n",
       "      <td id=\"T_880ca_row5_col0\" class=\"data row5 col0\" >0.80195</td>\n",
       "      <td id=\"T_880ca_row5_col1\" class=\"data row5 col1\" >0.91284</td>\n",
       "      <td id=\"T_880ca_row5_col2\" class=\"data row5 col2\" >0.23378</td>\n",
       "      <td id=\"T_880ca_row5_col3\" class=\"data row5 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row5_col4\" class=\"data row5 col4\" >5870</td>\n",
       "      <td id=\"T_880ca_row5_col5\" class=\"data row5 col5\" >461</td>\n",
       "      <td id=\"T_880ca_row5_col6\" class=\"data row5 col6\" >0.90256</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row6\" class=\"row_heading level0 row6\" >194</th>\n",
       "      <td id=\"T_880ca_row6_col0\" class=\"data row6 col0\" >0.80174</td>\n",
       "      <td id=\"T_880ca_row6_col1\" class=\"data row6 col1\" >0.82435</td>\n",
       "      <td id=\"T_880ca_row6_col2\" class=\"data row6 col2\" >0.21031</td>\n",
       "      <td id=\"T_880ca_row6_col3\" class=\"data row6 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row6_col4\" class=\"data row6 col4\" >4884</td>\n",
       "      <td id=\"T_880ca_row6_col5\" class=\"data row6 col5\" >479</td>\n",
       "      <td id=\"T_880ca_row6_col6\" class=\"data row6 col6\" >0.82434</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row7\" class=\"row_heading level0 row7\" >167</th>\n",
       "      <td id=\"T_880ca_row7_col0\" class=\"data row7 col0\" >0.80157</td>\n",
       "      <td id=\"T_880ca_row7_col1\" class=\"data row7 col1\" >0.93168</td>\n",
       "      <td id=\"T_880ca_row7_col2\" class=\"data row7 col2\" >0.22984</td>\n",
       "      <td id=\"T_880ca_row7_col3\" class=\"data row7 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row7_col4\" class=\"data row7 col4\" >6020</td>\n",
       "      <td id=\"T_880ca_row7_col5\" class=\"data row7 col5\" >446</td>\n",
       "      <td id=\"T_880ca_row7_col6\" class=\"data row7 col6\" >0.78511</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row8\" class=\"row_heading level0 row8\" >189</th>\n",
       "      <td id=\"T_880ca_row8_col0\" class=\"data row8 col0\" >0.80139</td>\n",
       "      <td id=\"T_880ca_row8_col1\" class=\"data row8 col1\" >0.86875</td>\n",
       "      <td id=\"T_880ca_row8_col2\" class=\"data row8 col2\" >0.22160</td>\n",
       "      <td id=\"T_880ca_row8_col3\" class=\"data row8 col3\" >14</td>\n",
       "      <td id=\"T_880ca_row8_col4\" class=\"data row8 col4\" >5437</td>\n",
       "      <td id=\"T_880ca_row8_col5\" class=\"data row8 col5\" >481</td>\n",
       "      <td id=\"T_880ca_row8_col6\" class=\"data row8 col6\" >0.93680</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row9\" class=\"row_heading level0 row9\" >187</th>\n",
       "      <td id=\"T_880ca_row9_col0\" class=\"data row9 col0\" >0.80101</td>\n",
       "      <td id=\"T_880ca_row9_col1\" class=\"data row9 col1\" >0.92904</td>\n",
       "      <td id=\"T_880ca_row9_col2\" class=\"data row9 col2\" >0.22382</td>\n",
       "      <td id=\"T_880ca_row9_col3\" class=\"data row9 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row9_col4\" class=\"data row9 col4\" >4799</td>\n",
       "      <td id=\"T_880ca_row9_col5\" class=\"data row9 col5\" >485</td>\n",
       "      <td id=\"T_880ca_row9_col6\" class=\"data row9 col6\" >0.94217</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row10\" class=\"row_heading level0 row10\" >184</th>\n",
       "      <td id=\"T_880ca_row10_col0\" class=\"data row10 col0\" >0.80084</td>\n",
       "      <td id=\"T_880ca_row10_col1\" class=\"data row10 col1\" >0.97208</td>\n",
       "      <td id=\"T_880ca_row10_col2\" class=\"data row10 col2\" >0.24419</td>\n",
       "      <td id=\"T_880ca_row10_col3\" class=\"data row10 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row10_col4\" class=\"data row10 col4\" >8236</td>\n",
       "      <td id=\"T_880ca_row10_col5\" class=\"data row10 col5\" >481</td>\n",
       "      <td id=\"T_880ca_row10_col6\" class=\"data row10 col6\" >0.94473</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row11\" class=\"row_heading level0 row11\" >172</th>\n",
       "      <td id=\"T_880ca_row11_col0\" class=\"data row11 col0\" >0.80082</td>\n",
       "      <td id=\"T_880ca_row11_col1\" class=\"data row11 col1\" >0.92319</td>\n",
       "      <td id=\"T_880ca_row11_col2\" class=\"data row11 col2\" >0.26398</td>\n",
       "      <td id=\"T_880ca_row11_col3\" class=\"data row11 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row11_col4\" class=\"data row11 col4\" >6059</td>\n",
       "      <td id=\"T_880ca_row11_col5\" class=\"data row11 col5\" >460</td>\n",
       "      <td id=\"T_880ca_row11_col6\" class=\"data row11 col6\" >0.86317</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row12\" class=\"row_heading level0 row12\" >164</th>\n",
       "      <td id=\"T_880ca_row12_col0\" class=\"data row12 col0\" >0.80076</td>\n",
       "      <td id=\"T_880ca_row12_col1\" class=\"data row12 col1\" >0.97947</td>\n",
       "      <td id=\"T_880ca_row12_col2\" class=\"data row12 col2\" >0.25079</td>\n",
       "      <td id=\"T_880ca_row12_col3\" class=\"data row12 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row12_col4\" class=\"data row12 col4\" >5863</td>\n",
       "      <td id=\"T_880ca_row12_col5\" class=\"data row12 col5\" >468</td>\n",
       "      <td id=\"T_880ca_row12_col6\" class=\"data row12 col6\" >0.74038</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row13\" class=\"row_heading level0 row13\" >94</th>\n",
       "      <td id=\"T_880ca_row13_col0\" class=\"data row13 col0\" >0.80072</td>\n",
       "      <td id=\"T_880ca_row13_col1\" class=\"data row13 col1\" >0.50000</td>\n",
       "      <td id=\"T_880ca_row13_col2\" class=\"data row13 col2\" >0.01000</td>\n",
       "      <td id=\"T_880ca_row13_col3\" class=\"data row13 col3\" >10</td>\n",
       "      <td id=\"T_880ca_row13_col4\" class=\"data row13 col4\" >500</td>\n",
       "      <td id=\"T_880ca_row13_col5\" class=\"data row13 col5\" >250</td>\n",
       "      <td id=\"T_880ca_row13_col6\" class=\"data row13 col6\" >0.50000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row14\" class=\"row_heading level0 row14\" >178</th>\n",
       "      <td id=\"T_880ca_row14_col0\" class=\"data row14 col0\" >0.80069</td>\n",
       "      <td id=\"T_880ca_row14_col1\" class=\"data row14 col1\" >0.86847</td>\n",
       "      <td id=\"T_880ca_row14_col2\" class=\"data row14 col2\" >0.22334</td>\n",
       "      <td id=\"T_880ca_row14_col3\" class=\"data row14 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row14_col4\" class=\"data row14 col4\" >8476</td>\n",
       "      <td id=\"T_880ca_row14_col5\" class=\"data row14 col5\" >460</td>\n",
       "      <td id=\"T_880ca_row14_col6\" class=\"data row14 col6\" >0.91886</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row15\" class=\"row_heading level0 row15\" >161</th>\n",
       "      <td id=\"T_880ca_row15_col0\" class=\"data row15 col0\" >0.80043</td>\n",
       "      <td id=\"T_880ca_row15_col1\" class=\"data row15 col1\" >0.98032</td>\n",
       "      <td id=\"T_880ca_row15_col2\" class=\"data row15 col2\" >0.22820</td>\n",
       "      <td id=\"T_880ca_row15_col3\" class=\"data row15 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row15_col4\" class=\"data row15 col4\" >3955</td>\n",
       "      <td id=\"T_880ca_row15_col5\" class=\"data row15 col5\" >456</td>\n",
       "      <td id=\"T_880ca_row15_col6\" class=\"data row15 col6\" >0.68397</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row16\" class=\"row_heading level0 row16\" >192</th>\n",
       "      <td id=\"T_880ca_row16_col0\" class=\"data row16 col0\" >0.80020</td>\n",
       "      <td id=\"T_880ca_row16_col1\" class=\"data row16 col1\" >0.96565</td>\n",
       "      <td id=\"T_880ca_row16_col2\" class=\"data row16 col2\" >0.22741</td>\n",
       "      <td id=\"T_880ca_row16_col3\" class=\"data row16 col3\" >14</td>\n",
       "      <td id=\"T_880ca_row16_col4\" class=\"data row16 col4\" >4728</td>\n",
       "      <td id=\"T_880ca_row16_col5\" class=\"data row16 col5\" >482</td>\n",
       "      <td id=\"T_880ca_row16_col6\" class=\"data row16 col6\" >0.97773</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row17\" class=\"row_heading level0 row17\" >145</th>\n",
       "      <td id=\"T_880ca_row17_col0\" class=\"data row17 col0\" >0.80012</td>\n",
       "      <td id=\"T_880ca_row17_col1\" class=\"data row17 col1\" >0.72079</td>\n",
       "      <td id=\"T_880ca_row17_col2\" class=\"data row17 col2\" >0.25217</td>\n",
       "      <td id=\"T_880ca_row17_col3\" class=\"data row17 col3\" >14</td>\n",
       "      <td id=\"T_880ca_row17_col4\" class=\"data row17 col4\" >2964</td>\n",
       "      <td id=\"T_880ca_row17_col5\" class=\"data row17 col5\" >415</td>\n",
       "      <td id=\"T_880ca_row17_col6\" class=\"data row17 col6\" >0.47304</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row18\" class=\"row_heading level0 row18\" >182</th>\n",
       "      <td id=\"T_880ca_row18_col0\" class=\"data row18 col0\" >0.80011</td>\n",
       "      <td id=\"T_880ca_row18_col1\" class=\"data row18 col1\" >0.92383</td>\n",
       "      <td id=\"T_880ca_row18_col2\" class=\"data row18 col2\" >0.22744</td>\n",
       "      <td id=\"T_880ca_row18_col3\" class=\"data row18 col3\" >15</td>\n",
       "      <td id=\"T_880ca_row18_col4\" class=\"data row18 col4\" >8612</td>\n",
       "      <td id=\"T_880ca_row18_col5\" class=\"data row18 col5\" >475</td>\n",
       "      <td id=\"T_880ca_row18_col6\" class=\"data row18 col6\" >0.85856</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th id=\"T_880ca_level0_row19\" class=\"row_heading level0 row19\" >143</th>\n",
       "      <td id=\"T_880ca_row19_col0\" class=\"data row19 col0\" >0.79962</td>\n",
       "      <td id=\"T_880ca_row19_col1\" class=\"data row19 col1\" >0.74984</td>\n",
       "      <td id=\"T_880ca_row19_col2\" class=\"data row19 col2\" >0.25684</td>\n",
       "      <td id=\"T_880ca_row19_col3\" class=\"data row19 col3\" >14</td>\n",
       "      <td id=\"T_880ca_row19_col4\" class=\"data row19 col4\" >3007</td>\n",
       "      <td id=\"T_880ca_row19_col5\" class=\"data row19 col5\" >396</td>\n",
       "      <td id=\"T_880ca_row19_col6\" class=\"data row19 col6\" >0.38898</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n"
      ],
      "text/plain": [
       "<pandas.io.formats.style.Styler at 0x74d4bc688e50>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_hyperp = study.trials_dataframe().rename(columns={'value': 'roc_auc'})\n",
    "cols_hyperp = ['roc_auc'] + df_hyperp.columns[df_hyperp.columns.str.contains('params_')].tolist()\n",
    "df_hyperp[cols_hyperp].sort_values('roc_auc', ascending=False).head(20).style.background_gradient(cmap='coolwarm').format(precision=5)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5cbbf3db-9afc-431b-9f75-767170ea9be5",
   "metadata": {},
   "source": [
    "# Assess best model\n",
    "\n",
    "I'll take the best hyperparams and just glance at performance, since my main question has been answered: yes, the prediction problem has legs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "faf36b23-9e68-4168-a635-0c94e3a54f29",
   "metadata": {},
   "outputs": [],
   "source": [
    "best_params = {'device': 'cuda',\n",
    "               'tree_method': 'hist',\n",
    "               'objective': 'binary:logistic',\n",
    "               'colsample_bytree': 0.5,\n",
    "               'learning_rate': 0.01,\n",
    "               'max_depth': 10,\n",
    "               'min_child_weight': 50,\n",
    "               'n_estimators': 250,\n",
    "               'subsample': 0.5\n",
    "               }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "1e596de3-42f9-4975-9b36-3057c699315d",
   "metadata": {},
   "outputs": [],
   "source": [
    "cudf_cash_crunch['pred'] = cupy.nan\n",
    "\n",
    "for cutoff_year in list(range(first_window_final_year, last_year - 1)):\n",
    "    sel_train = cudf_cash_crunch.date.dt.year <= cutoff_year # oops\n",
    "    sel_pred = cudf_cash_crunch.date.dt.year == cutoff_year + 1\n",
    "    if (sel_train.sum() > 0) and (sel_pred.sum() > 0):\n",
    "        clf = xgboost.XGBClassifier(**best_params)\n",
    "        clf.fit(X=cudf_cash_crunch.loc[sel_train, cols_feat_keep_low_corr], y=cudf_cash_crunch.loc[sel_train, 'targ'])\n",
    "        cudf_cash_crunch.loc[sel_pred, 'pred'] = clf.predict_proba(X=cudf_cash_crunch.loc[sel_pred, cols_feat_keep_low_corr])[:, 1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "ce4c5d1a-5bfe-458b-98be-5b3d8b0073ba",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(545352, 6)\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>key_daily</th>\n",
       "      <th>date</th>\n",
       "      <th>targ</th>\n",
       "      <th>pred</th>\n",
       "      <th>ratio_adj_chg</th>\n",
       "      <th>pred2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>SCI_2018-05-03</td>\n",
       "      <td>2018-05-03</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.090</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>ENSV_2009-11-04</td>\n",
       "      <td>2009-11-04</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.279</td>\n",
       "      <td>-10.391</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>SFL_2012-09-06</td>\n",
       "      <td>2012-09-06</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.184</td>\n",
       "      <td>1.218</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>SFL_2011-02-03</td>\n",
       "      <td>2011-02-03</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.228</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>AGU_2016-01-06</td>\n",
       "      <td>2016-01-06</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0.189</td>\n",
       "      <td>0.000</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         key_daily       date  targ  pred  ratio_adj_chg  pred2\n",
       "0   SCI_2018-05-03 2018-05-03 0.000 0.090          0.000      0\n",
       "1  ENSV_2009-11-04 2009-11-04 0.000 0.279        -10.391      0\n",
       "2   SFL_2012-09-06 2012-09-06 0.000 0.184          1.218      0\n",
       "3   SFL_2011-02-03 2011-02-03 0.000 0.228          0.000      0\n",
       "4   AGU_2016-01-06 2016-01-06 0.000 0.189          0.000      0"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "_sel = (~cudf_cash_crunch.pred.isnull()) & (~cudf_cash_crunch.targ.isnull())\n",
    "df_score = cudf_cash_crunch.loc[_sel, ['key_daily', 'date', 'targ', 'pred']].to_pandas()\n",
    "df_score = df_score.merge(df_targ[['key_daily', 'ratio_adj_chg']], how='inner', on='key_daily')\n",
    "df_score['pred2'] = df_score.pred.apply(lambda x: 1 if x > 0.5 else 0)\n",
    "print(df_score.shape)\n",
    "df_score.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "c9562ff1-d69d-47df-866f-710849a593e3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x74d4bc832850>"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay, precision_score, recall_score\n",
    "ConfusionMatrixDisplay.from_predictions(df_score.targ, df_score.pred2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "b94084be-dd5d-4006-8928-2fa0a33997db",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.7303228 , 0.03444198],\n",
       "       [0.16830781, 0.06692742]])"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confusion_matrix(df_score.targ, df_score.pred2)/len(df_score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "c3b88a88-705f-4079-bd3d-7beea3d8b5b4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.6602329872291162, 0.2845127293703132)"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "precision_score(df_score.targ, df_score.pred2), recall_score(df_score.targ, df_score.pred2)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6af19d0c-4d90-4a1f-af7a-9da70130bfde",
   "metadata": {},
   "source": [
    "That recall is pretty dismal, though admittedly that's for a 0.5 threshold. ROC AUC should tell a more complete tale."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "49612259-945c-41a2-9ba2-c6768e273937",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_curve, RocCurveDisplay, auc\n",
    "\n",
    "fpr, tpr, thresholds = roc_curve(df_score.targ, df_score.pred)\n",
    "auc_score = auc(fpr, tpr)\n",
    "plt.plot(fpr, tpr)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('ROC AUC = {:.4f}'.format(auc_score))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "57c25e92-2f74-4345-a675-b705c043597d",
   "metadata": {},
   "source": [
    "Score through time? Meaning, as each subsequent models trained, did it improve with more historical stock data?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "aaf0614a-f8e5-4b8b-9d41-ed1878d809fa",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "min   2008-01-04\n",
       "max   2022-12-05\n",
       "Name: date, dtype: datetime64[ns]"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_score.date.agg(['min', 'max'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "5b68cf7a-4328-4db4-aba6-94ed2938b780",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "list_prediction_years = list(range(df_score.date.min().year, df_score.date.max().year + 1))\n",
    "list_scores_yearly = []\n",
    "for y in list_prediction_years:\n",
    "    sel_pred_yr = df_score.date.dt.year == y\n",
    "    fpr, tpr, thresholds = roc_curve(df_score.loc[sel_pred_yr, 'targ'], df_score.loc[sel_pred_yr, 'pred'])\n",
    "    list_scores_yearly.append(auc(fpr, tpr))\n",
    "\n",
    "pd.DataFrame({'roc_auc_score': list_scores_yearly}, index=list_prediction_years).plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9a5dfdb9-a19e-4814-9a87-894da62bf722",
   "metadata": {},
   "source": [
    "There may indeed be a jagged upward trend in performance for subsequent models, with a massive 4% drop in AUC for the pandemic era. I suppose that speaks to model risk a bit.\n",
    "\n",
    "Let's also peek at the predicted probabilities against the underlying continuous target with the help of LOWESS (scatterplot smoother)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "851c151c-f19a-48ce-9f26-16e439dadea6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Axes: xlabel='pred', ylabel='ratio_adj_chg'>"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.regplot(data=df_score.sample(frac=0.1), x='pred', y='ratio_adj_chg', scatter=False, lowess=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f742efb2-3291-4ff8-8cdb-54da4ee8884f",
   "metadata": {},
   "source": [
    "I also wonder what features are prominent here for the final-year model, which was trained on most of the instances."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "aa1d2e25-9fea-44d7-93af-d4d8ee1886cc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(407, 3)\n",
      "21 features contribute 49.2% of cumulative importance\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>feat</th>\n",
       "      <th>imp</th>\n",
       "      <th>imp_c</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>netinccmnusd_over_revenue_art</td>\n",
       "      <td>0.094</td>\n",
       "      <td>0.094</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>ncfo_over_assetsavg_art</td>\n",
       "      <td>0.080</td>\n",
       "      <td>0.173</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>netinccmnusd_over_revenue_ary5y_avg</td>\n",
       "      <td>0.045</td>\n",
       "      <td>0.218</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>liabilitiesc_over_fcf_ary5y_std</td>\n",
       "      <td>0.029</td>\n",
       "      <td>0.247</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>revenueusd_over_fcf_ary5y_std</td>\n",
       "      <td>0.025</td>\n",
       "      <td>0.272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>ebitda_over_fcf_art</td>\n",
       "      <td>0.022</td>\n",
       "      <td>0.294</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>taxexp_over_revenue_art</td>\n",
       "      <td>0.019</td>\n",
       "      <td>0.313</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>assetsnc_over_assetsavg_ary5y_avg</td>\n",
       "      <td>0.018</td>\n",
       "      <td>0.331</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>capex_over_revenue_ary5y_avg</td>\n",
       "      <td>0.016</td>\n",
       "      <td>0.347</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>assetsnc_over_assetsavg_art</td>\n",
       "      <td>0.015</td>\n",
       "      <td>0.362</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>netinccmnusd_over_assetsavg_art</td>\n",
       "      <td>0.013</td>\n",
       "      <td>0.375</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>revenueusd_over_assetsavg_art</td>\n",
       "      <td>0.013</td>\n",
       "      <td>0.388</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>liabilitiesc_over_fcf_art</td>\n",
       "      <td>0.013</td>\n",
       "      <td>0.401</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>liabilitiesc_over_assetsavg_art</td>\n",
       "      <td>0.013</td>\n",
       "      <td>0.413</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>revenueusd_over_fcf_art</td>\n",
       "      <td>0.013</td>\n",
       "      <td>0.426</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>assetsc_over_assetsavg_ary5y_avg</td>\n",
       "      <td>0.012</td>\n",
       "      <td>0.438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>capex_over_fcf_art</td>\n",
       "      <td>0.012</td>\n",
       "      <td>0.450</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>workingcapital_over_fcf_ary5y_std</td>\n",
       "      <td>0.012</td>\n",
       "      <td>0.462</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>receivables_over_assetsavg_ary5y_std</td>\n",
       "      <td>0.012</td>\n",
       "      <td>0.474</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>receivables_over_assetsavg_ary5y_avg</td>\n",
       "      <td>0.009</td>\n",
       "      <td>0.483</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                    feat   imp  imp_c\n",
       "0          netinccmnusd_over_revenue_art 0.094  0.094\n",
       "1                ncfo_over_assetsavg_art 0.080  0.173\n",
       "2    netinccmnusd_over_revenue_ary5y_avg 0.045  0.218\n",
       "3        liabilitiesc_over_fcf_ary5y_std 0.029  0.247\n",
       "4          revenueusd_over_fcf_ary5y_std 0.025  0.272\n",
       "5                    ebitda_over_fcf_art 0.022  0.294\n",
       "6                taxexp_over_revenue_art 0.019  0.313\n",
       "7      assetsnc_over_assetsavg_ary5y_avg 0.018  0.331\n",
       "8           capex_over_revenue_ary5y_avg 0.016  0.347\n",
       "9            assetsnc_over_assetsavg_art 0.015  0.362\n",
       "10       netinccmnusd_over_assetsavg_art 0.013  0.375\n",
       "11         revenueusd_over_assetsavg_art 0.013  0.388\n",
       "12             liabilitiesc_over_fcf_art 0.013  0.401\n",
       "13       liabilitiesc_over_assetsavg_art 0.013  0.413\n",
       "14               revenueusd_over_fcf_art 0.013  0.426\n",
       "15      assetsc_over_assetsavg_ary5y_avg 0.012  0.438\n",
       "16                    capex_over_fcf_art 0.012  0.450\n",
       "17     workingcapital_over_fcf_ary5y_std 0.012  0.462\n",
       "18  receivables_over_assetsavg_ary5y_std 0.012  0.474\n",
       "19  receivables_over_assetsavg_ary5y_avg 0.009  0.483"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<Axes: >"
      ]
     },
     "execution_count": 88,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "clf.importance_type = 'total_gain'\n",
    "cumul_importance_thresh = 0.5\n",
    "\n",
    "df_fi = pd.DataFrame({'feat': cols_feat_keep_low_corr, 'imp': clf.feature_importances_})\n",
    "df_fi = df_fi.sort_values('imp', ascending=False).reset_index(drop=True)\n",
    "df_fi['imp_c'] = df_fi.imp.cumsum()\n",
    "important_features = df_fi.loc[df_fi.imp_c <= cumul_importance_thresh, 'feat']\n",
    "threshold_contrib = df_fi.loc[df_fi.feat.isin(important_features), 'imp_c'].max()\n",
    "\n",
    "print(df_fi.shape)\n",
    "print(len(important_features), 'features contribute {}% of cumulative importance'.format((100.*threshold_contrib).round(1)))\n",
    "df_fi.to_csv(data_path.joinpath(\"analytics_demo\").joinpath('df_fi.csv'), index=None)\n",
    "display(df_fi.head(20))\n",
    "df_fi.loc[:, 'imp_c'].plot.line()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ec04f5b6-f19b-4a60-87d2-106b514ea85c",
   "metadata": {},
   "source": [
    "# Notes and To-Do's"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "986ae5f1-a71e-4a27-a722-17b418deeeab",
   "metadata": {},
   "source": [
    "- The above is what I hoped for: A couple dozen features pulling most of the weight.\n",
    "    - I'm surprised to see profit margin (`netinccmnusd_over_revenue_art`) doing 9.4% of the predictive work for a cash flow target.\n",
    "        - Combined with its 5-year average, it does nearly 14% of the work!\n",
    "    - Unsurprisingly, the other half dozen top features involve cash flow somehow.\n",
    "- I failed to create a lagged target as a feature; it just slipped my mind.\n",
    "    - Then again, I expect that some of the top features (like `liabilitiesc_over_fcf_ary5y_std`) carry related information.\n",
    "    - Still, I suspect I could add, say, 0.05 to the AUC score with a lagged target feature.\n",
    "- I was in a hurry and chose `binary:logistic` as XGB's objective function rather than `reg:logistic`, but really it would've been wiser to either think carefully about this, or tune it as a hyperparam.\n",
    "    - Note: A quick search indicates that `binary:logistic` scores a split on accuracy, while `reg:logistic` uses RMSE somehow.\n",
    "- Given the importance of catching risky borrowers, we likely need to understand recall behavior better.\n",
    "- Once we were happy with our feature set and model performance, we'd want to lower dimensionality via feature combination into latent dimensions, rather than simply removing correlated features or relying on supervised models (feature importance, SHAP files). Potential techniques include:\n",
    "    - PCA\n",
    "    - Autoencoder\n",
    "    - Feature network\n",
    "- With luck, studying latent dimensions will enable the design of robust features from a small handful of financial details, perfect for a business credit application."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}