12_claim_freq_sev.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "dfb9399d-1bb5-49f2-a9ed-7a1b63419e15",
   "metadata": {},
   "source": [
    "# 12. Claim frequency and severity modeling with GBM"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6c563c8f-6e9e-4b8a-8571-28eef337dd9d",
   "metadata": {},
   "source": [
    "### 12.1. Claim frequency modeling\n",
    "\n",
    "* include the log of exposure as offset\n",
    "* specify the poisson distribution for the target\n",
    "* impose monotonically incresing contraint on ```bm```\n",
    "* perform stochastic gradient boosting with ```max_depth = 3``` and ```learning_rate = 0.1```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "3046d43b-f552-4f26-9f8d-f08da6c902ef",
   "metadata": {},
   "outputs": [],
   "source": [
    "# import relevant libraries\n",
    "from sklearn.model_selection import GridSearchCV,RandomizedSearchCV,train_test_split\n",
    "from matplotlib.colors import ListedColormap\n",
    "from matplotlib import pyplot as plt\n",
    "from matplotlib import animation\n",
    "import xgboost as xgb\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "51d8ee45-7c2a-41ec-8c30-c1e0f557af5e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# load data using separator to correctly parse from .txt\n",
    "df = pd.read_csv(r'.\\data\\PC_data.txt', sep=r\"\\s+\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "4bd09a55-877c-42c8-96a8-0af804f5602f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# column names to smaller case\n",
    "df.columns = map(str.lower, df.columns)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "8c580083-5e93-4c55-88eb-8091d0be8a77",
   "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>id</th>\n",
       "      <th>nclaims</th>\n",
       "      <th>amount</th>\n",
       "      <th>avg</th>\n",
       "      <th>expo</th>\n",
       "      <th>coverage</th>\n",
       "      <th>fuel</th>\n",
       "      <th>use</th>\n",
       "      <th>fleet</th>\n",
       "      <th>sex</th>\n",
       "      <th>ageph</th>\n",
       "      <th>bm</th>\n",
       "      <th>agec</th>\n",
       "      <th>power</th>\n",
       "      <th>pc</th>\n",
       "      <th>town</th>\n",
       "      <th>long</th>\n",
       "      <th>lat</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1618.001036</td>\n",
       "      <td>1618.001036</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>TPL</td>\n",
       "      <td>gasoline</td>\n",
       "      <td>private</td>\n",
       "      <td>N</td>\n",
       "      <td>male</td>\n",
       "      <td>50</td>\n",
       "      <td>5</td>\n",
       "      <td>12</td>\n",
       "      <td>77</td>\n",
       "      <td>1000</td>\n",
       "      <td>BRUSSEL</td>\n",
       "      <td>4.355223</td>\n",
       "      <td>50.845386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>PO</td>\n",
       "      <td>gasoline</td>\n",
       "      <td>private</td>\n",
       "      <td>N</td>\n",
       "      <td>female</td>\n",
       "      <td>64</td>\n",
       "      <td>5</td>\n",
       "      <td>3</td>\n",
       "      <td>66</td>\n",
       "      <td>1000</td>\n",
       "      <td>BRUSSEL</td>\n",
       "      <td>4.355223</td>\n",
       "      <td>50.845386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>TPL</td>\n",
       "      <td>diesel</td>\n",
       "      <td>private</td>\n",
       "      <td>N</td>\n",
       "      <td>male</td>\n",
       "      <td>60</td>\n",
       "      <td>0</td>\n",
       "      <td>10</td>\n",
       "      <td>70</td>\n",
       "      <td>1000</td>\n",
       "      <td>BRUSSEL</td>\n",
       "      <td>4.355223</td>\n",
       "      <td>50.845386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>NaN</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>TPL</td>\n",
       "      <td>gasoline</td>\n",
       "      <td>private</td>\n",
       "      <td>N</td>\n",
       "      <td>male</td>\n",
       "      <td>77</td>\n",
       "      <td>0</td>\n",
       "      <td>15</td>\n",
       "      <td>57</td>\n",
       "      <td>1000</td>\n",
       "      <td>BRUSSEL</td>\n",
       "      <td>4.355223</td>\n",
       "      <td>50.845386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>1</td>\n",
       "      <td>155.974606</td>\n",
       "      <td>155.974606</td>\n",
       "      <td>0.046575</td>\n",
       "      <td>TPL</td>\n",
       "      <td>gasoline</td>\n",
       "      <td>private</td>\n",
       "      <td>N</td>\n",
       "      <td>female</td>\n",
       "      <td>28</td>\n",
       "      <td>9</td>\n",
       "      <td>7</td>\n",
       "      <td>70</td>\n",
       "      <td>1000</td>\n",
       "      <td>BRUSSEL</td>\n",
       "      <td>4.355223</td>\n",
       "      <td>50.845386</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   id  nclaims       amount          avg      expo coverage      fuel  \\\n",
       "0   1        1  1618.001036  1618.001036  1.000000      TPL  gasoline   \n",
       "1   2        0     0.000000          NaN  1.000000       PO  gasoline   \n",
       "2   3        0     0.000000          NaN  1.000000      TPL    diesel   \n",
       "3   4        0     0.000000          NaN  1.000000      TPL  gasoline   \n",
       "4   5        1   155.974606   155.974606  0.046575      TPL  gasoline   \n",
       "\n",
       "       use fleet     sex  ageph  bm  agec  power    pc     town      long  \\\n",
       "0  private     N    male     50   5    12     77  1000  BRUSSEL  4.355223   \n",
       "1  private     N  female     64   5     3     66  1000  BRUSSEL  4.355223   \n",
       "2  private     N    male     60   0    10     70  1000  BRUSSEL  4.355223   \n",
       "3  private     N    male     77   0    15     57  1000  BRUSSEL  4.355223   \n",
       "4  private     N  female     28   9     7     70  1000  BRUSSEL  4.355223   \n",
       "\n",
       "         lat  \n",
       "0  50.845386  \n",
       "1  50.845386  \n",
       "2  50.845386  \n",
       "3  50.845386  \n",
       "4  50.845386  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# rename exp to expo\n",
    "df.rename(columns={'exp':'expo'}, inplace = True); df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "bf0fe268-25ea-4fe7-9948-783c9369ee85",
   "metadata": {},
   "outputs": [],
   "source": [
    "# converting type(str) columns to categorical\n",
    "df['coverage'] = df['coverage'].astype('category').cat.codes\n",
    "df['fleet'] = df['fleet'].astype('category').cat.codes\n",
    "df['sex'] = df['sex'].astype('category').cat.codes\n",
    "df['town'] = df['town'].astype('category').cat.codes\n",
    "df['fuel'] = df['fuel'].astype('category').cat.codes\n",
    "df['use'] = df['use'].astype('category').cat.codes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "2d01ddbd-8598-4b1f-8dcc-2727ea753146",
   "metadata": {},
   "outputs": [],
   "source": [
    "# cols to retain\n",
    "cols = ['bm','ageph','agec','power','coverage','fuel','sex','fleet','use']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "04628488-312c-49bd-913f-7c52db2bb395",
   "metadata": {},
   "outputs": [],
   "source": [
    "# let's calculate calculate claim frequency and weights\n",
    "y = np.array(df.nclaims/df.expo)\n",
    "w = np.array(df.expo)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "d23943ef-3488-41c4-9aa7-e28066d1485a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# let's rearrange the dataset\n",
    "x = df[cols] # before, we had to convert this to array before fitting xgb. Now, monotone_constraint got restructured and, apparently, tracks the name of columns via dict -> we need to have the column names -> pd.DataFrame"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "7b733bba-3069-4f5b-bf9a-f0af4aaf6b25",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=1,\n",
       "             colsample_bynode=1, colsample_bytree=1, enable_categorical=False,\n",
       "             gamma=0, gpu_id=-1, importance_type=None,\n",
       "             interaction_constraints='', learning_rate=0.1,\n",
       "             max_delta_step=0.699999988, max_depth=3, min_child_weight=1,\n",
       "             missing=nan, monotone_constraints={'bm': 1}, n_estimators=500,\n",
       "             n_jobs=16, num_parallel_tree=1, objective='count:poisson',\n",
       "             predictor='auto', random_state=0, reg_alpha=0, reg_lambda=1,\n",
       "             scale_pos_weight=None, subsample=1, tree_method='exact',\n",
       "             validate_parameters=1, verbosity=None)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "xgb_model = xgb.XGBRegressor(n_estimators = 500, \n",
    "                              objective='count:poisson',\n",
    "                              monotone_constraints = {'bm':1},# the way this is implemented has changed, how? good question ahaha\n",
    "                              max_depth = 3,\n",
    "                              learning_rate = 0.1)\n",
    "\n",
    "xgb_model.fit(x,y,sample_weight = w)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "acf6e046-92fe-4cda-8b15-4a7b5cff3efc",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(163231, 18)"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "ec3d1f16-44fe-45ff-9a00-e7be138d1374",
   "metadata": {},
   "outputs": [],
   "source": [
    "# prediction on y scale\n",
    "pred_on_target_scale = np.round(xgb_model.predict(x), 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "3ded67fb-999b-4951-9523-8dcaac38d095",
   "metadata": {},
   "outputs": [],
   "source": [
    "# prediction on raw log scale\n",
    "raw_pred_on_log_scale = np.round(xgb_model.predict(x, output_margin = True), 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "490a082c-4df3-4950-848d-74a2b40f99b4",
   "metadata": {},
   "outputs": [],
   "source": [
    "# prediction on y scala starting from row log scale\n",
    "pred_on_target_scale_from_raw = np.round(np.exp(xgb_model.predict(x, output_margin = True)),5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "35ec627d-318f-4216-b993-8aa5167eb47f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#checking equality (with tolerance for decimals)\n",
    "np.isclose(pred_on_target_scale, pred_on_target_scale_from_raw, atol=1e-05, equal_nan=False).all() # absolute(a - b) <= atol"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "15396df9-6110-4f08-ac5f-fe06c33df635",
   "metadata": {},
   "outputs": [],
   "source": [
    "test = pd.DataFrame({'obs':y,'pred':pred_on_target_scale})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "afc63b77-e556-4878-9a06-aa8e25dc0ae6",
   "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>obs</th>\n",
       "      <th>pred</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.17022</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.12354</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.10121</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.07280</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>21.470588</td>\n",
       "      <td>0.26871</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>163226</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.16713</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>163227</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.12711</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>163228</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.11546</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>163229</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.11014</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>163230</th>\n",
       "      <td>2.000000</td>\n",
       "      <td>0.17187</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>163231 rows × 2 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              obs     pred\n",
       "0        1.000000  0.17022\n",
       "1        0.000000  0.12354\n",
       "2        0.000000  0.10121\n",
       "3        0.000000  0.07280\n",
       "4       21.470588  0.26871\n",
       "...           ...      ...\n",
       "163226   0.000000  0.16713\n",
       "163227   0.000000  0.12711\n",
       "163228   0.000000  0.11546\n",
       "163229   0.000000  0.11014\n",
       "163230   2.000000  0.17187\n",
       "\n",
       "[163231 rows x 2 columns]"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "ca45bb09-3c23-4920-ae48-d33c09f14c3e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<sklearn.inspection._plot.partial_dependence.PartialDependenceDisplay at 0x1c5dcf0cb08>"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.inspection import PartialDependenceDisplay\n",
    "PartialDependenceDisplay.from_estimator(xgb_model, features = ['bm','ageph'], X=x, feature_names = ['bm','ageph'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "33c534ba-a33f-47c9-a934-c41d36762ed8",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'bm': 37.15327835083008,\n",
       " 'ageph': 3.1714229583740234,\n",
       " 'agec': 2.0373644828796387,\n",
       " 'power': 1.9949947595596313,\n",
       " 'coverage': 2.7172040939331055,\n",
       " 'fuel': 5.319155693054199,\n",
       " 'sex': 2.7582147121429443,\n",
       " 'fleet': 1.5841093063354492,\n",
       " 'use': 1.1127355098724365}"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "xgb_model.get_booster().feature_names = cols\n",
    "xgb_model.get_booster().get_score(importance_type=\"gain\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "c314cd7c-71e1-4fc0-99bc-0ef06dc743f4",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:>"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAD4CAYAAADGmmByAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjUuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8/fFQqAAAACXBIWXMAAAsTAAALEwEAmpwYAAAYhElEQVR4nO3dfZzVdZ338debG7mVCZDtQaCOtQgYI2BA3qQRomXu6rZWdENr2i6phbk+9CFbeq2b1XrZtXlTl7rTtYZXcRUZhT2kBAw0dA0YbgdEaVsnAt28QUa8wRX8XH+c79RpnIGB35z5/Ybzfj4e85jf+f5uzuf3e8B5z/f7Ped3FBGYmZkdqh55F2BmZt2bg8TMzDJxkJiZWSYOEjMzy8RBYmZmmfTKu4CudtRRR0VtbW3eZZiZdStr1qx5LiKGtbWu6oKktraWhoaGvMswM+tWJP22vXUe2jIzs0wcJGZmlomDxMzMMnGQmJlZJlU32d64o5naOYvyLsPMrEs13XhuxY7tHomZmWXiIDEzs0wKGySSaiVtyrsOMzPbv8IGiZmZdQ9FD5JekuZJ2iLpR5L6S2qS9M+S1ktqkHSSpMWSfiPpkrwLNjOrNkUPktHA7RExFngRuCy1b4uICcAKYC7wYeBk4J/aOoikWSl0Gva90lzxos3MqknRg+R3EfFIWv4e8J60/NP0uxFYGRG7I+JZ4DVJb2l9kIioj4hJETGpZ/+aihdtZlZNih4krb9QvuXxa+n3G2XLLY+r7rMxZmZ5KnqQHCPplLT8CeDhPIsxM7M3K3qQPAF8TtIWYDBwR871mJlZK4UdBoqIJmBMG6tqy7aZS2myveVx7Zu2NjOziipskFRK3YgaGip4zxkzs2pT9KEtMzMrOAeJmZll4iAxM7NMHCRmZpaJg8TMzDJxkJiZWSYOEjMzy8RBYmZmmThIzMwsEweJmZll4iAxM7NMHCRmZpZJ1d20sXFHM7VzFuVdxkFr8o0mzayg3CMxM7NMChckki6XtEXSvEPYt0nSUZWoy8zM2lbEoa3LgOkRsT3vQszM7MAK1SORdCfwduDnkpolXVW2bpOk2rQ8U9IqSesl/auknjmVbGZW9QoVJBFxCfAU8D7g5ra2kTQWmAGcFhETgH3AJ7uqRjMz+1NFHNo6kDOBdwGrJQH0A57Z3w6SZgGzAHoOGlbp+szMqkqRg2Qvf9pj6pt+C7g7Iv6howeKiHqgHqDP8FHRaRWamVmxhrZaaQJOApB0EnBcav8F8GFJf5bWDZF0bC4VmplZoYNkATBE0mbg88BWgIh4DLgWWCJpI7AUGJ5blWZmVa5wQ1sRUVv28Ox2tpkPzD/AvmZm1gWK3CMxM7NuoHA9kkqrG1FDg+9bZWbWadwjMTOzTBwkZmaWiYPEzMwycZCYmVkmDhIzM8vEQWJmZpk4SMzMLBMHiZmZZeIgMTOzTBwkZmaWiYPEzMwyqbp7bTXuaKZ2zqIue74m39fLzA5z7pGYmVkmDhIzM8ukWweJpKmS7su7DjOzatatg8TMzPJX8SCRtFDSGkmbJc1KbZ+RtFXSKknflvSt1D5M0gJJq9PPaan9eknflfSopF9L+ruypxgo6UeSHpc0T5IqfU5mZvZHXfGurYsjYqekfsBqSYuA64CTgN3AMmBD2vZW4OaIeFjSMcBiYGxadyJwMjAAWJeOAzAReCfwFPAIcBrwcHkBKcBmAfQcNKwiJ2lmVq26Ikgul/ShtHw08CngoYjYCSDpHuD4tH46cEJZp2KQpIFp+d6IeBV4VdJyYAqwC1gVEdvTsdYDtbQKkoioB+oB+gwfFZ18fmZmVa2iQSJpKqVwOCUiXpH0IPA4f+xltNYDODki9rQ6DkDrAGh5/FpZ2z6q8LMxZmZ5qvQcSQ3wQgqRMfxxaOq9kgZL6gVcULb9EmB2ywNJE8rWnS+pr6ShwFRgdYVrNzOzDqh0kNwP9JK0BbgR+BWwA/gasIrSnEYT0Jy2vxyYJGmjpMeAS8qOtRFYno5xQ0Q8VeHazcysAyo6DBQRrwHntG6X1BAR9alH8hNgYdr+OWBGO4fbGBF/0+r4DwIPlj3+fKcUbmZmHZbXfML1kqYDfSkNZy3sqieuG1FDg+9/ZWbWaXIJkoi46iC3v75CpZiZWUb+ZLuZmWXiIDEzs0wcJGZmlomDxMzMMnGQmJlZJg4SMzPLxEFiZmaZOEjMzCwTB4mZmWXiIDEzs0yq7rs7Gnc0Uztn0YE3PERNvo+XmVUZ90jMzCwTB4mZmWXiIDEzs0wcJGZmlkkhg0TSAEmLJG2QtEnSDEnvkvSQpDWSFksaLqlG0hOSRqf9vi/p7/Ku38ysmhT1XVsfAJ6KiHMBJNUAPwfOj4hnJc0AvhoRF0v6PDBX0q3A4Ij4duuDSZoFzALoOWhYl52EmVk1KGqQNAL/Iul/AvcBLwDjgKWSAHoCTwNExFJJHwH+NzC+rYNFRD1QD9Bn+KioePVmZlWkkEESEVslnQR8EPgKsAzYHBGntN5WUg9gLPAKMBjY3pW1mplVu6LOkbwNeCUivgd8HXg3MEzSKWl9b0nvTJv/PbAF+ATwHUm986jZzKxaFbJHAtQBX5f0BvA6cCmwF7gtzZf0Am6RtBf4W2BKROyW9EvgWuAfc6rbzKzqFDJIImIxsLiNVWe00Ta2bL8rK1aUmZm1qZBBUkl1I2po8P2wzMw6TSHnSMzMrPtwkJiZWSYOEjMzy8RBYmZmmThIzMwsEweJmZll4iAxM7NMHCRmZpaJg8TMzDJxkJiZWSYOEjMzy6Tq7rXVuKOZ2jmLKnb8Jt/Hy8yqjHskZmaWiYPEzMwycZCYmVkmhQkSSVU3X2NmdjjoUJBI+htJGyVtkPRdSbWSlqW2X0g6RlKNpN9K6pH2GSDpd+n71d8h6X5JayStkDQmbTNX0p2SVgI3SZoi6VFJ6yT9u6TRabv+kn4o6TFJP5G0UtKktO7stM9aSfdIGliha2VmZm04YC9A0jspfQ/6qRHxnKQhwN3A3RFxt6SLgdsi4q8krQfeCywH/gJYHBGvS6oHLomIX0t6N3A7MC09xch07H2SBgGnR8ReSdOBrwEXAJcBL0TECZLGAetTbUel2qZHxMuSrgGuBL7c6hxmAbMAeg4adoiXyszM2tKR4aRpwD0R8RxAROyUdArw12n9d4Gb0vJ8YAalIPkYcHvqIZwK3COp5Zh9yo5/T0TsS8s1wN2SRgEB9E7t7wFuTc+/SdLG1H4ycALwSDr2EcCjrU8gIuqBeoA+w0dFB87ZzMw6qLPnJX4KfC31Wt4FLAMGALsiYkI7+7xctnwDsDwiPiSpFnjwAM8nYGlEfDxL0WZmdug6MkeyDPiIpKEAKST+nVKPA+CTwAqAiHgJWE2p93BfROyLiBeBJyV9JO0vSePbea4aYEda/nRZ+yPAR9P+JwB1qf1XwGmS/jytGyDp+A6ck5mZdZIDBklEbAa+CjwkaQPwDWA2cFEaYvoU8IWyXeYDM9PvFp8EPpP23wyc387T3QT8s6R1/Glv6XZgmKTHgK+kYzRHxLOUAuf7qZZHgTEHOiczM+s8iij+lIGknkDviNgj6R3AA8DoiPjvgz1Wn+GjYviFt3R2iX/gW6SY2eFI0pqImNTWuu7y2Y3+wHJJvSnNi1x2KCECUDeihga/2JuZdZpuESQRsRtoMwnNzCxfhflku5mZdU8OEjMzy8RBYmZmmThIzMwsEweJmZll4iAxM7NMHCRmZpaJg8TMzDJxkJiZWSYOEjMzy6Rb3CKlMzXuaKZ2zqLMx/HNGc3MStwjMTOzTBwkZmaWiYPEzMwycZCYmVkmuQaJpIWS1kjaLGlWavuMpK2SVkn6tqRvpfZhkhZIWp1+TkvtAyV9R1KjpI2SLsjznMzMqk3e79q6OCJ2SuoHrJa0CLgOOAnYDSwDNqRtbwVujoiHJR0DLAbGpu2bI6IOQNLg1k+SQmoWQM9Bwyp8SmZm1SXvILlc0ofS8tHAp4CHImIngKR7gOPT+unACZJa9h0kaWBq/1hLY0S80PpJIqIeqIfSd7ZX4DzMzKpWbkEiaSqlEDglIl6R9CDwOKVeRlt6ACdHxJ5Wx6lglWZmdiB5zpHUAC+kEBkDnAwMAN4rabCkXkD5fMcSYHbLA0kT0uJS4HNl7W8a2jIzs8rJM0juB3pJ2gLcCPwK2AF8DVgFPAI0Ac1p+8uBSWlC/THgktT+FWCwpE2SNgDv67pTMDOz3Ia2IuI14JzW7ZIaIqI+9Uh+AixM2z8HzGjjOC8BF1a2WjMza0/ek+1tuV7SdKAvpeGshZ158LoRNTT4PllmZp2mcEESEVflXYOZmXWcP9luZmaZOEjMzCwTB4mZmWXiIDEzs0wcJGZmlomDxMzMMnGQmJlZJg4SMzPLxEFiZmaZOEjMzCyTwt0ipdIadzRTO2dR5uM0+X5dZmaAeyRmZpaRg8TMzDJxkJiZWSaHZZCkL8UyM7MuUPEgkVQr6XFJ8yRtkfQjSf0lnSlpnaRGSXdJ6iNpsqQfp/3Ol/SqpCMk9ZX0n6n9HZLul7RG0or0fe9ImivpTkkrgZsqfV5mZlbSVT2S0cDtETEWeBG4EpgLzIiIOkrvHrsUWAdMSPucDmwCJgPvBlam9npgdkS8C7gKuL3seUYCp0bEleVPLmmWpAZJDfteacbMzDpPVwXJ7yLikbT8PeBM4MmI2Jra7gbOiIi9wG8kjQWmAN8AzqAUKiskDQROBe6RtB74V2B42fPcExH7Wj95RNRHxKSImNSzf00FTs/MrHp11VxCtHq8Cxjazra/BM4BXgceoNRz6QlcTSn4dkXEhHb2fTljnWZmdpC6qkdyjKRT0vIngAagVtKfp7ZPAQ+l5RXAFcCjEfEspcAZDWyKiBeBJyV9BEAl47voHMzMrA1dFSRPAJ+TtAUYDNwMXERpiKoReAO4M227EngrpZ4JwEagMSJaejWfBD4jaQOwGTi/a07BzMza0lVDW3sjYmartl8AE1tvGBGvAn3KHs9qtf5J4ANt7PfpTqnUzMwOStV93qJuRA0Nvk+WmVmnqXiQREQTMK7Sz2NmZvk4LD/ZbmZmXcdBYmZmmThIzMwsEweJmZll4iAxM7NMHCRmZpaJg8TMzDJxkJiZWSYOEjMzy8RBYmZmmThIzMwsk6q7aWPjjmZq5yw66P2afKNHM7M2uUdiZmaZ5B4kki6XtEXSDknfOsRjTJD0wc6uzczMDiz3IAEuA84CvpThGBMAB4mZWQ5yDRJJdwJvB35O6St4W9qHSVogaXX6OS21D5B0l6RVktZJOl/SEcCXgRmS1kuakcvJmJlVqVyDJCIuAZ4C3ge8ULbqVuDmiJgMXAD8n9T+JWBZRExJ+3wd6A38D2B+REyIiPldVb+ZmRX3XVvTgRMktTweJGkgcDZwnqSrUntf4JgDHUzSLGAWQM9Bwzq/WjOzKlbUIOkBnBwRe8obVUqWCyLiiVbt797fwSKiHqgH6DN8VHRyrWZmVa0Ik+1tWQLMbnkgaUJaXAzMToGCpImpfTdwZFcWaGZmJUUNksuBSZI2SnoMuCS130BpTmSjpM3pMcBySkNhnmw3M+tiuQ9tRURtWpybfoiI54A3BUJEvAp8to32ncDkStVoZmbtK2qPxMzMuonceyRdrW5EDQ2+b5aZWadxj8TMzDJxkJiZWSYOEjMzy8RBYmZmmThIzMwsEweJmZllUnVv/zUz66jXX3+d7du3s2fPngNvfJjo27cvI0eOpHfv3h3ex0FiZtaO7du3c+SRR1JbW0vZ3cgPWxHB888/z/bt2znuuOM6vJ+HtszM2rFnzx6GDh1aFSECIImhQ4cedA/MQWJmth/VEiItDuV8HSRmZpZJ1c2RNO5opnbOogNu1+T7cZlZKx157TgYh8vrjHskZmZVYO/evRU7toPEzKygXn75Zc4991zGjx/PuHHjmD9/PqtXr+bUU09l/PjxTJkyhd27d7Nnzx4uuugi6urqmDhxIsuXLwdg7ty5nHfeeUybNo0zzzyTl19+mYsvvpgpU6YwceJE7r333k6ps+qGtszMuov777+ft73tbSxaVBpSa25uZuLEicyfP5/Jkyfz4osv0q9fP2699VYk0djYyOOPP87ZZ5/N1q1bAVi7di0bN25kyJAhfPGLX2TatGncdddd7Nq1iylTpjB9+nQGDBiQqU73SMzMCqquro6lS5dyzTXXsGLFCrZt28bw4cOZPLn0hbCDBg2iV69ePPzww8ycOROAMWPGcOyxx/4hSM466yyGDBkCwJIlS7jxxhuZMGECU6dOZc+ePWzbti1znYXskUiqBe6LiHHp8VXAQGAnpe9v3ws8FhEfkzQA+CYwjtL3uV8fEZ3TXzMzy9Hxxx/P2rVr+dnPfsa1117LtGnTDvoY5b2NiGDBggWMHj26M8vsdj2SOcDEiDiRUqAAfAlYFhFTgPcBX0/h8geSZklqkNSw75Xmrq3YzOwQPfXUU/Tv35+ZM2dy9dVXs3LlSp5++mlWr14NwO7du9m7dy+nn3468+bNA2Dr1q1s27atzbB4//vfzze/+U0iAoB169Z1Sp2F7JHsx0ZgnqSFwMLUdjZwXuq1APQFjgG2tOwUEfVAPUCf4aOiq4o1s8NLV79dt7GxkauvvpoePXrQu3dv7rjjDiKC2bNn8+qrr9KvXz8eeOABLrvsMi699FLq6uro1asXc+fOpU+fPm863nXXXccVV1zBiSeeyBtvvMFxxx3Hfffdl7lOtSRTkUgaCSyJiBPS42sphd4NwBnAXwLnAHXASuATEfFER47dZ/ioGH7hLQfc7nB5f7eZHbotW7YwduzYvMvocm2dt6Q1ETGpre2LOrT1e+DPJA2V1Af4C0q1Hh0Ry4FrgBpK8yaLgdlKn+uXNDGnms3MqlIhh7Yi4nVJXwZWATuAx4GewPck1QACbouIXZJuAG4BNkrqATxJKXjMzKwLFDJIACLiNuC2Dmz3KvDZyldkZtUoIqrqxo2HMt1R2CCplLoRNTR4/sPMOqBv3748//zzVXMr+ZbvI+nbt+9B7Vd1QWJm1lEjR45k+/btPPvss3mX0mVaviHxYDhIzMza0bt374P6psBqVdR3bZmZWTfhIDEzs0wcJGZmlkkhP9leSZJ2Ax36FHwBHAU8l3cRHeRaK8O1VoZrPXjHRsSwtlZU42T7E+19zL9oJDW41s7nWivDtVZGd6jVQ1tmZpaJg8TMzDKpxiCpz7uAg+BaK8O1VoZrrYzC11p1k+1mZta5qrFHYmZmnchBYmZmmVRVkEj6gKQnJP2HpDl517M/kpokNUpaL6kh73rKSbpL0jOSNpW1DZG0VNKv0+/BedbYop1ar5e0I13b9ZI+mGeNLSQdLWm5pMckbZb0hdReqGu7nzoLd10l9ZW0StKGVOs/pfbjJK1MrwXzJR1R4FrnSnqy7LpOyLnUN6maORJJPYGtwFnAdmA18PGIeCzXwtohqQmYFBFF+CDSn5B0BvAS8H8jYlxquwnYGRE3ppAeHBHX5FlnqqutWq8HXoqI/5Vnba1JGg4Mj4i1ko4E1gB/BXyaAl3b/dT5UQp2XdM3pw6IiJck9QYeBr4AXAn8OCJ+IOlOYENE3FHQWi8B7ouIH+VZ3/5UU49kCvAfEfGfEfHfwA+A83OuqVuKiF8CO1s1nw/cnZbvpvTCkrt2ai2kiHg6Itam5d3AFmAEBbu2+6mzcKLkpfSwd/oJYBrQ8sKc+zWF/dZaeNUUJCOA35U93k5B//EnASyRtEbSrLyL6YC3RsTTafm/gLfmWUwHfF7SxjT0VYhhuHKSaoGJwEoKfG1b1QkFvK6SekpaDzwDLAV+A+yKiL1pk8K8FrSuNSJarutX03W9WVKf/CpsWzUFSXfznog4CTgH+FwaoukWojReWuS/pO4A3gFMAJ4G/iXXalqRNBBYAFwRES+WryvStW2jzkJe14jYFxETgJGURibG5FtR+1rXKmkc8A+Uap4MDAFyHzJurZqCZAdwdNnjkamtkCJiR/r9DPATSv8Biuz3aey8ZQz9mZzraVdE/D79h30D+DYFurZpbHwBMC8ifpyaC3dt26qzyNcVICJ2AcuBU4C3SGq512DhXgvKav1AGkqMiHgN+A4Fu65QXUGyGhiV3q1xBPAx4Kc519QmSQPSJCaSBgBnA5v2v1fufgpcmJYvBO7NsZb9anlRTj5EQa5tmmz9N2BLRHyjbFWhrm17dRbxukoaJuktabkfpTfbbKH0Iv3htFnu1xTarfXxsj8iRGkuJ/fr2lrVvGsLIL0d8RagJ3BXRHw134raJuntlHohULpD8/8rUq2Svg9MpXR7698D/wgsBH4IHAP8FvhoROQ+yd1OrVMpDb8E0AR8tmwOIjeS3gOsABqBN1LzFynNPxTm2u6nzo9TsOsq6URKk+k9Kf3h/MOI+HL6P/YDSkNF64CZ6S/+3Oyn1mXAMEDAeuCSskn5QqiqIDEzs85XTUNbZmZWAQ4SMzPLxEFiZmaZOEjMzCwTB4mZmWXiIDEzs0wcJGZmlsn/Bz9/GUhdX1bbAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# feature importance\n",
    "feature_importance = xgb_model.get_booster().get_score(importance_type='gain')\n",
    "keys = list(feature_importance.keys())\n",
    "values = list(feature_importance.values())\n",
    "\n",
    "data = pd.DataFrame(data=values, index=keys, columns=[\"score\"]).sort_values(by = \"score\", ascending=True)\n",
    "data.plot(kind='barh')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "23710c5a-ba0f-4172-b7b3-fc669e7c114c",
   "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>Tree</th>\n",
       "      <th>Node</th>\n",
       "      <th>ID</th>\n",
       "      <th>Feature</th>\n",
       "      <th>Split</th>\n",
       "      <th>Yes</th>\n",
       "      <th>No</th>\n",
       "      <th>Missing</th>\n",
       "      <th>Gain</th>\n",
       "      <th>Cover</th>\n",
       "      <th>Category</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0-0</td>\n",
       "      <td>bm</td>\n",
       "      <td>7.0</td>\n",
       "      <td>0-1</td>\n",
       "      <td>0-2</td>\n",
       "      <td>0-1</td>\n",
       "      <td>255.927734</td>\n",
       "      <td>146231.156000</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0-1</td>\n",
       "      <td>bm</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0-3</td>\n",
       "      <td>0-4</td>\n",
       "      <td>0-3</td>\n",
       "      <td>40.210938</td>\n",
       "      <td>116025.688000</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0-2</td>\n",
       "      <td>bm</td>\n",
       "      <td>11.0</td>\n",
       "      <td>0-5</td>\n",
       "      <td>0-6</td>\n",
       "      <td>0-5</td>\n",
       "      <td>32.356445</td>\n",
       "      <td>30205.470700</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0-3</td>\n",
       "      <td>ageph</td>\n",
       "      <td>56.0</td>\n",
       "      <td>0-7</td>\n",
       "      <td>0-8</td>\n",
       "      <td>0-7</td>\n",
       "      <td>13.670898</td>\n",
       "      <td>81035.437500</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>0-4</td>\n",
       "      <td>fuel</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0-9</td>\n",
       "      <td>0-10</td>\n",
       "      <td>0-9</td>\n",
       "      <td>6.787109</td>\n",
       "      <td>34990.250000</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6831</th>\n",
       "      <td>499</td>\n",
       "      <td>10</td>\n",
       "      <td>499-10</td>\n",
       "      <td>Leaf</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.055307</td>\n",
       "      <td>1.378483</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6832</th>\n",
       "      <td>499</td>\n",
       "      <td>11</td>\n",
       "      <td>499-11</td>\n",
       "      <td>Leaf</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.022926</td>\n",
       "      <td>21.727396</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6833</th>\n",
       "      <td>499</td>\n",
       "      <td>12</td>\n",
       "      <td>499-12</td>\n",
       "      <td>Leaf</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-0.000382</td>\n",
       "      <td>8127.612300</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6834</th>\n",
       "      <td>499</td>\n",
       "      <td>13</td>\n",
       "      <td>499-13</td>\n",
       "      <td>Leaf</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.000595</td>\n",
       "      <td>10542.572300</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6835</th>\n",
       "      <td>499</td>\n",
       "      <td>14</td>\n",
       "      <td>499-14</td>\n",
       "      <td>Leaf</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-0.000179</td>\n",
       "      <td>21982.957000</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>6836 rows × 11 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "      Tree  Node      ID Feature  Split  Yes    No Missing        Gain  \\\n",
       "0        0     0     0-0      bm    7.0  0-1   0-2     0-1  255.927734   \n",
       "1        0     1     0-1      bm    2.0  0-3   0-4     0-3   40.210938   \n",
       "2        0     2     0-2      bm   11.0  0-5   0-6     0-5   32.356445   \n",
       "3        0     3     0-3   ageph   56.0  0-7   0-8     0-7   13.670898   \n",
       "4        0     4     0-4    fuel    1.0  0-9  0-10     0-9    6.787109   \n",
       "...    ...   ...     ...     ...    ...  ...   ...     ...         ...   \n",
       "6831   499    10  499-10    Leaf    NaN  NaN   NaN     NaN    0.055307   \n",
       "6832   499    11  499-11    Leaf    NaN  NaN   NaN     NaN    0.022926   \n",
       "6833   499    12  499-12    Leaf    NaN  NaN   NaN     NaN   -0.000382   \n",
       "6834   499    13  499-13    Leaf    NaN  NaN   NaN     NaN    0.000595   \n",
       "6835   499    14  499-14    Leaf    NaN  NaN   NaN     NaN   -0.000179   \n",
       "\n",
       "              Cover  Category  \n",
       "0     146231.156000       NaN  \n",
       "1     116025.688000       NaN  \n",
       "2      30205.470700       NaN  \n",
       "3      81035.437500       NaN  \n",
       "4      34990.250000       NaN  \n",
       "...             ...       ...  \n",
       "6831       1.378483       NaN  \n",
       "6832      21.727396       NaN  \n",
       "6833    8127.612300       NaN  \n",
       "6834   10542.572300       NaN  \n",
       "6835   21982.957000       NaN  \n",
       "\n",
       "[6836 rows x 11 columns]"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# aggregate trees in a pd.DataFrame\n",
    "xgb_model.get_booster().trees_to_dataframe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "f490b87c-1fc4-4749-a060-b1a923edb2ae",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "booster[0]:\n",
      "0:[bm<7] yes=1,no=2,missing=1,gain=255.927734,cover=146231.156\n",
      "\t1:[bm<2] yes=3,no=4,missing=3,gain=40.2109375,cover=116025.688\n",
      "\t\t3:[ageph<56] yes=7,no=8,missing=7,gain=13.6708984,cover=81035.4375\n",
      "\t\t\t7:leaf=-0.0381389894,cover=49579.9375\n",
      "\t\t\t8:leaf=-0.0408193506,cover=31455.498\n",
      "\t\t4:[fuel<1] yes=9,no=10,missing=9,gain=6.78710938,cover=34990.25\n",
      "\t\t\t9:leaf=-0.0331041776,cover=11479.3447\n",
      "\t\t\t10:leaf=-0.0360973291,cover=23510.9043\n",
      "\t2:[bm<11] yes=5,no=6,missing=5,gain=32.3564453,cover=30205.4707\n",
      "\t\t5:[ageph<27] yes=11,no=12,missing=11,gain=9.27807617,cover=19862.8926\n",
      "\t\t\t11:leaf=-0.0265295003,cover=4825.41748\n",
      "\t\t\t12:leaf=-0.0312158596,cover=15037.4746\n",
      "\t\t6:[agec<7] yes=13,no=14,missing=13,gain=6.33276367,cover=10342.5791\n",
      "\t\t\t13:leaf=-0.0263584312,cover=3765.06372\n",
      "\t\t\t14:leaf=-0.0211947113,cover=6577.51514\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# dump model (direct access to each tree) and load it\n",
    "tree_selected = 0\n",
    "xgb_model.get_booster()[tree_selected].dump_model(\"./data/out.txt\", with_stats=True)\n",
    "\n",
    "# read the contents of the file\n",
    "with open('./data/out.txt', 'r') as f:\n",
    "    txt_model = f.read()\n",
    "print(txt_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b12aa2df-51eb-4377-b7cb-26cd12141b72",
   "metadata": {},
   "source": [
    "We can see that the predictions for #ofclaims/exposure are negative. This happens because the predictions for target are **on log scale**. Let's show it fitting a smaller xgb model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "8a221640-b6a0-4fb1-acf8-2c73bcfed25d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=1,\n",
       "             colsample_bynode=1, colsample_bytree=1, enable_categorical=False,\n",
       "             gamma=0, gpu_id=-1, importance_type=None,\n",
       "             interaction_constraints='', learning_rate=0.1,\n",
       "             max_delta_step=0.699999988, max_depth=3, min_child_weight=1,\n",
       "             missing=nan, monotone_constraints='()', n_estimators=2, n_jobs=16,\n",
       "             num_parallel_tree=1, objective='count:poisson', predictor='auto',\n",
       "             random_state=0, reg_alpha=0, reg_lambda=1, scale_pos_weight=None,\n",
       "             subsample=1, tree_method='exact', validate_parameters=1,\n",
       "             verbosity=None)"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "xgb_model_2 = xgb.XGBRegressor(n_estimators = 2, \n",
    "                              objective='count:poisson',\n",
    "                              # monotone_constraints = (1,0,0,0,0,0,0,0,0), # the way this is implemented has changed, how? good question ahaha\n",
    "                              max_depth = 3,\n",
    "                              learning_rate = 0.1)\n",
    "\n",
    "xgb_model_2.fit(x,y,sample_weight = w)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "eeff822e-b49b-4af7-a1c0-fc31705aa9da",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "booster[0]:\n",
      "0:[bm<7] yes=1,no=2,missing=1,gain=255.927734,cover=146231.156\n",
      "\t1:[bm<2] yes=3,no=4,missing=3,gain=40.2109375,cover=116025.688\n",
      "\t\t3:[ageph<56] yes=7,no=8,missing=7,gain=13.6708984,cover=81035.4375\n",
      "\t\t\t7:leaf=-0.0381389894,cover=49579.9375\n",
      "\t\t\t8:leaf=-0.0408193506,cover=31455.498\n",
      "\t\t4:[ageph<58] yes=9,no=10,missing=9,gain=6.96289063,cover=34990.25\n",
      "\t\t\t9:leaf=-0.0343534909,cover=27195.8984\n",
      "\t\t\t10:leaf=-0.0377732851,cover=7794.35107\n",
      "\t2:[bm<11] yes=5,no=6,missing=5,gain=32.3564453,cover=30205.4707\n",
      "\t\t5:[ageph<27] yes=11,no=12,missing=11,gain=9.35437012,cover=19862.8926\n",
      "\t\t\t11:leaf=-0.0261318367,cover=4825.41748\n",
      "\t\t\t12:leaf=-0.0312158596,cover=15037.4746\n",
      "\t\t6:[agec<7] yes=13,no=14,missing=13,gain=6.33276367,cover=10342.5791\n",
      "\t\t\t13:leaf=-0.0263584312,cover=3765.06372\n",
      "\t\t\t14:leaf=-0.0211947113,cover=6577.51514\n",
      "booster[1]:\n",
      "0:[bm<7] yes=1,no=2,missing=1,gain=256.384766,cover=141087.844\n",
      "\t1:[bm<3] yes=3,no=4,missing=3,gain=40.7910156,cover=111704.883\n",
      "\t\t3:[ageph<56] yes=7,no=8,missing=7,gain=16.2763672,cover=86347.5469\n",
      "\t\t\t7:leaf=-0.0374729559,cover=54066.9961\n",
      "\t\t\t8:leaf=-0.0403234884,cover=32280.5527\n",
      "\t\t4:[ageph<56] yes=9,no=10,missing=9,gain=5.88916016,cover=25357.334\n",
      "\t\t\t9:leaf=-0.0330735669,cover=18952.8906\n",
      "\t\t\t10:leaf=-0.0366149954,cover=6404.44336\n",
      "\t2:[bm<11] yes=5,no=6,missing=5,gain=31.7434082,cover=29382.9648\n",
      "\t\t5:[ageph<27] yes=11,no=12,missing=11,gain=9.24084473,cover=19276.2715\n",
      "\t\t\t11:leaf=-0.0255090725,cover=4700.9541\n",
      "\t\t\t12:leaf=-0.0306311157,cover=14575.3184\n",
      "\t\t6:[power<40] yes=13,no=14,missing=13,gain=6.17269897,cover=10106.6924\n",
      "\t\t\t13:leaf=-0.0276952516,cover=1849.80078\n",
      "\t\t\t14:leaf=-0.0212796591,cover=8256.8916\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# dump model (direct access to each tree) and load it\n",
    "xgb_model_2.get_booster().dump_model(\"./data/out_2.txt\", with_stats=True)\n",
    "\n",
    "# read the contents of the file\n",
    "with open('./data/out_2.txt', 'r') as f:\n",
    "    txt_model = f.read()\n",
    "print(txt_model)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "1ebd854f-bcec-436f-a347-e10369ddd510",
   "metadata": {},
   "source": [
    "What records go into the predict?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "fc8d002a-4487-4d8a-8a89-97f5dd6f141b",
   "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>bm</th>\n",
       "      <th>ageph</th>\n",
       "      <th>agec</th>\n",
       "      <th>power</th>\n",
       "      <th>coverage</th>\n",
       "      <th>fuel</th>\n",
       "      <th>sex</th>\n",
       "      <th>fleet</th>\n",
       "      <th>use</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>5</td>\n",
       "      <td>50</td>\n",
       "      <td>12</td>\n",
       "      <td>77</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>5</td>\n",
       "      <td>64</td>\n",
       "      <td>3</td>\n",
       "      <td>66</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   bm  ageph  agec  power  coverage  fuel  sex  fleet  use\n",
       "0   5     50    12     77         2     1    1      0    0\n",
       "1   5     64     3     66         1     1    0      0    0"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x[:2]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8aefa179-1ad3-428a-a238-a59de5874e42",
   "metadata": {},
   "source": [
    "We can visually assess which is \"the final leaf\" of the prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "75c69562-e516-48b4-9360-a3acd160c9ab",
   "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>0</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-0.034353</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-0.037773</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          0\n",
       "0 -0.034353\n",
       "1 -0.037773"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_2 = pd.DataFrame(xgb_model_2.predict(x[:2], iteration_range=(0, 1), output_margin=True) - np.log(0.5)); pred_2"
   ]
  }
 ],
 "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": "3.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}