Training_testing_pipeline.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ab6ec3aa",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "import re\n",
    "import xgboost as xgb\n",
    "import optuna\n",
    "\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.model_selection import train_test_split, KFold\n",
    "from sklearn.metrics import accuracy_score, classification_report, confusion_matrix, roc_auc_score\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.base import BaseEstimator, TransformerMixin\n",
    "from sklearn.preprocessing import FunctionTransformer, OneHotEncoder\n",
    "from sklearn.pipeline import Pipeline, FeatureUnion\n",
    "from sklearn.compose import ColumnTransformer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "2fd3943f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# importing transformations for feature pipeline from .py script\n",
    "from pipeline_helpers.feature_transformations import *\n",
    "from pipeline_helpers.target_transformation import *"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "422ebdb8",
   "metadata": {},
   "source": [
    "### Configuration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "0f489439",
   "metadata": {},
   "outputs": [],
   "source": [
    "# set visibility for all the columns\n",
    "pd.set_option('display.max_columns', None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "4e9888c6",
   "metadata": {},
   "outputs": [],
   "source": [
    "def missing_data_info(df):\n",
    "    total_missing = df.isnull().sum()\n",
    "    percent_missing = (df.isnull().sum() / df.shape[0]) * 100\n",
    "    \n",
    "    # Combine the results into a DataFrame and sort by the 'Percent' column\n",
    "    missing_data = pd.concat([total_missing, percent_missing], axis=1, keys=['Total', 'Percent'])\n",
    "    missing_data = missing_data[missing_data['Percent'] > 0].sort_values(by='Percent', ascending=False)\n",
    "    \n",
    "    return missing_data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9dd197e7",
   "metadata": {},
   "source": [
    "### Data Loading"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "1d137d6a",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load the datasets\n",
    "train_df = pd.read_csv('data/Training_dataset.csv')\n",
    "test_df = pd.read_csv('data/Test_dataset.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "8a5161bb",
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead 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>Total</th>\n",
       "      <th>Percent</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>VehColorInt</th>\n",
       "      <td>728</td>\n",
       "      <td>11.559225</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Vehicle_Trim</th>\n",
       "      <td>405</td>\n",
       "      <td>6.430613</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehDriveTrain</th>\n",
       "      <td>401</td>\n",
       "      <td>6.367101</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehEngine</th>\n",
       "      <td>361</td>\n",
       "      <td>5.731978</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehPriceLabel</th>\n",
       "      <td>285</td>\n",
       "      <td>4.525246</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehFeats</th>\n",
       "      <td>275</td>\n",
       "      <td>4.366466</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehSellerNotes</th>\n",
       "      <td>243</td>\n",
       "      <td>3.858368</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehHistory</th>\n",
       "      <td>201</td>\n",
       "      <td>3.191489</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehTransmission</th>\n",
       "      <td>197</td>\n",
       "      <td>3.127977</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehColorExt</th>\n",
       "      <td>73</td>\n",
       "      <td>1.159098</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Dealer_Listing_Price</th>\n",
       "      <td>52</td>\n",
       "      <td>0.825659</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SellerListSrc</th>\n",
       "      <td>2</td>\n",
       "      <td>0.031756</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SellerZip</th>\n",
       "      <td>2</td>\n",
       "      <td>0.031756</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehFuel</th>\n",
       "      <td>2</td>\n",
       "      <td>0.031756</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehListdays</th>\n",
       "      <td>2</td>\n",
       "      <td>0.031756</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VehMileage</th>\n",
       "      <td>2</td>\n",
       "      <td>0.031756</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                      Total    Percent\n",
       "VehColorInt             728  11.559225\n",
       "Vehicle_Trim            405   6.430613\n",
       "VehDriveTrain           401   6.367101\n",
       "VehEngine               361   5.731978\n",
       "VehPriceLabel           285   4.525246\n",
       "VehFeats                275   4.366466\n",
       "VehSellerNotes          243   3.858368\n",
       "VehHistory              201   3.191489\n",
       "VehTransmission         197   3.127977\n",
       "VehColorExt              73   1.159098\n",
       "Dealer_Listing_Price     52   0.825659\n",
       "SellerListSrc             2   0.031756\n",
       "SellerZip                 2   0.031756\n",
       "VehFuel                   2   0.031756\n",
       "VehListdays               2   0.031756\n",
       "VehMileage                2   0.031756"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "missing_data_info(train_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6c86da32",
   "metadata": {},
   "source": [
    "### Preparation of target variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "5ef89905",
   "metadata": {},
   "outputs": [],
   "source": [
    "train_df_final = preprocessing_target_variables(train_df)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3baece33",
   "metadata": {},
   "source": [
    "## 1. Classifier - Vehicle_Trim"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a5dec12e",
   "metadata": {},
   "source": [
    "### Training and preprocessing using pipeline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "7bcf5b27",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Label encode Vehicle_Trim\n",
    "le = LabelEncoder()\n",
    "y = le.fit_transform(train_df_final['Vehicle_Trim'])\n",
    "\n",
    "# Split data into training and test sets\n",
    "X = train_df_final.drop(['Vehicle_Trim', 'Dealer_Listing_Price'], axis=1)\n",
    "X_train, X_valid, y_train, y_valid = train_test_split(X, y, test_size=0.2, random_state=42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "81230dfa",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0 1 2 3 4 5 6 7 8 9]\n",
      "[0 1 2 3 4 5 6 7 8 9]\n",
      "[0 1 2 3 4 5 6 7 8 9]\n"
     ]
    }
   ],
   "source": [
    "print(np.unique(y))\n",
    "print(np.unique(y_train))\n",
    "print(np.unique(y_valid))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "0b431bdf",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define and use your transformation pipeline on X\n",
    "CURRENT_YEAR = 2023\n",
    "index_feature = ['ListingID']\n",
    "unused_features = ['VehTransmission','VehBodystyle','SellerZip','VehType','VehSellerNotes',\n",
    "                       'VehPriceLabel','VehModel','SellerIsPriv','SellerName','SellerCity']\n",
    "\n",
    "log_transform_features = ['SellerRevCnt', 'VehListdays']\n",
    "boolean_feature = ['VehCertified']\n",
    "one_hot_features = ['SellerListSrc', 'SellerState', 'VehFuel', 'VehDrive','BasicExtColor','Engine_Category']\n",
    "\n",
    "preprocessing_pipeline = Pipeline([\n",
    "    (\"set_index\", SetIndex(index_feature=index_feature)),\n",
    "    (\"drop_unused\", DropUnused(features=unused_features)),\n",
    "    (\"log_transform\", LogTransform(features=log_transform_features)),\n",
    "    (\"age_feature\", AgeFeatureTransform(current_year=CURRENT_YEAR)),\n",
    "    (\"boolean_encode\", BooleanEncoding(features=boolean_feature)),\n",
    "    ('veh_drive_train_transform', VehDriveTrainTransform()),\n",
    "    ('veh_feats_transform', VehFeatsTransform()),\n",
    "    ('veh_color_ext_transform', VehColorExtTransform()),\n",
    "    ('veh_color_int_transform', VehColorInternalTransform()),\n",
    "    ('veh_engine_transform', VehEngineTransform()),\n",
    "    ('veh_make_transform', VehMakeTransform()),\n",
    "    ('veh_mileage_transform', VehMileageTransform()),\n",
    "    ('veh_history_transform', VehHistoryTransform()),\n",
    "    (\"column_transformer\", ColumnTransformer([\n",
    "        (\"one_hot_encode\", OneHotEncoder(handle_unknown='ignore', dtype=int), \n",
    "         one_hot_features)], remainder='passthrough'))])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f646b965",
   "metadata": {},
   "source": [
    "- `remainder=passthrough` argument ensures that columns not explicitly specified in the transformer list are left unchanged.\n",
    "- `handle_unknown=ignore` ensures that unseen categories in the test set won't cause errors and will be ignored. This is essential to handle potential differences in categories between the training and test datasets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "12606434",
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "# Transform the train and validation data\n",
    "X_train_transformed = preprocessing_pipeline.fit_transform(X_train)\n",
    "X_valid_transformed = preprocessing_pipeline.transform(X_valid)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "5d317d34",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original X_train shape: (4642, 27)\n",
      "Transformed X_train shape: (4642, 94)\n",
      "y_train shape: (4642,)\n"
     ]
    }
   ],
   "source": [
    "print('Original X_train shape:',X_train.shape)\n",
    "print('Transformed X_train shape:',X_train_transformed.shape)\n",
    "print('y_train shape:',y_train.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "20687aa4",
   "metadata": {},
   "source": [
    "#### Classifier Baseling Model Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "e13d9d82",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ROC Score:  0.9392575340413764\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.51      0.24      0.33        92\n",
      "           1       0.61      0.52      0.56        42\n",
      "           2       0.57      0.42      0.48       150\n",
      "           3       0.64      0.88      0.74       380\n",
      "           4       0.56      0.69      0.62       123\n",
      "           5       0.61      0.42      0.50       121\n",
      "           6       0.81      0.68      0.74        37\n",
      "           7       0.80      0.71      0.75       147\n",
      "           8       1.00      1.00      1.00        23\n",
      "           9       0.70      0.50      0.58        46\n",
      "\n",
      "    accuracy                           0.65      1161\n",
      "   macro avg       0.68      0.61      0.63      1161\n",
      "weighted avg       0.64      0.65      0.63      1161\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 800x600 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model = xgb.XGBClassifier(use_label_encoder=False, eval_metric='merror')\n",
    "model.fit(X_train_transformed, y_train) \n",
    "\n",
    "# Predictions\n",
    "y_valid_prob  = model.predict_proba(X_valid_transformed)\n",
    "y_valid_label = np.argmax(y_valid_prob, axis=1)\n",
    "\n",
    "roc_auc = roc_auc_score(y_valid, y_valid_prob, multi_class='ovr')\n",
    "\n",
    "print('ROC Score: ',roc_auc)\n",
    "\n",
    "report = classification_report(y_valid, y_valid_label)\n",
    "print(report)\n",
    "\n",
    "cm = confusion_matrix(y_valid, y_valid_label)\n",
    "plt.figure(figsize=(8,6))\n",
    "sns.heatmap(cm, annot=True, fmt=\"d\", cmap=\"Blues\")\n",
    "plt.ylabel('True Label')\n",
    "plt.xlabel('Predicted Label')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ea7d4d07",
   "metadata": {},
   "source": [
    "### Hyperparameter Tunning using Optuna"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "704108c8",
   "metadata": {},
   "outputs": [],
   "source": [
    "def objective(trial):\n",
    "    # Define hyperparameter search space\n",
    "    print('Trial: ', trial.number)\n",
    "\n",
    "        \n",
    "    param = {\n",
    "        'n_estimators': trial.suggest_int('n_estimators', 100, 1000),\n",
    "        'max_depth': trial.suggest_int('max_depth', 3, 9),\n",
    "        'learning_rate': trial.suggest_float('learning_rate', 0.005, 0.5, log=True),\n",
    "        'subsample': trial.suggest_float('subsample', 0.5, 1),\n",
    "        'colsample_bytree': trial.suggest_float('colsample_bytree', 0.5, 1),\n",
    "        'gamma': trial.suggest_float('gamma', 0, 0.5),\n",
    "        'lambda': trial.suggest_float('lambda', 1e-8, 1.0, log=True),\n",
    "        'alpha': trial.suggest_float('alpha', 1e-8, 1.0, log=True),\n",
    "        'min_child_weight': trial.suggest_int('min_child_weight', 1, 10),\n",
    "        'use_label_encoder': False,\n",
    "        'eval_metric': 'merror'\n",
    "    }\n",
    "\n",
    "    # Initialize and train classifier\n",
    "    model = xgb.XGBClassifier(**param)\n",
    "    model.fit(X_train_transformed, y_train, eval_set=[(X_valid_transformed, y_valid)], verbose=False)\n",
    "\n",
    "    # Predictions\n",
    "    y_valid_prob  = model.predict_proba(X_valid_transformed)\n",
    "    y_valid_label = np.argmax(y_valid_prob, axis=1)\n",
    "\n",
    "    roc_auc = roc_auc_score(y_valid, y_valid_prob, multi_class='ovr')\n",
    "\n",
    "    # Return AUC\n",
    "    return roc_auc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "f28f9b4a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:11,696] A new study created in memory with name: no-name-88d3a4ba-4fad-49b7-b42d-0916056c1e2c\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:13,389] Trial 0 finished with value: 0.9343284603295745 and parameters: {'n_estimators': 184, 'max_depth': 7, 'learning_rate': 0.20801540858510095, 'subsample': 0.7790570514833028, 'colsample_bytree': 0.8666208615355606, 'gamma': 0.1652390606905048, 'lambda': 0.08699160504393891, 'alpha': 0.00012048685617050913, 'min_child_weight': 7}. Best is trial 0 with value: 0.9343284603295745.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  1\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:20,107] Trial 1 finished with value: 0.9362333893642584 and parameters: {'n_estimators': 801, 'max_depth': 9, 'learning_rate': 0.39601121987977494, 'subsample': 0.7650159265566437, 'colsample_bytree': 0.6273346226955602, 'gamma': 0.10202640804337887, 'lambda': 0.01997463980098406, 'alpha': 1.2683410353550242e-08, 'min_child_weight': 1}. Best is trial 1 with value: 0.9362333893642584.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  2\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:26,124] Trial 2 finished with value: 0.9360699869761987 and parameters: {'n_estimators': 325, 'max_depth': 9, 'learning_rate': 0.10975716677532411, 'subsample': 0.5104383346628856, 'colsample_bytree': 0.9297716683863197, 'gamma': 0.16840630457856298, 'lambda': 9.591265648797883e-05, 'alpha': 1.6353143570297765e-06, 'min_child_weight': 1}. Best is trial 1 with value: 0.9362333893642584.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:35,347] Trial 3 finished with value: 0.9353057097405321 and parameters: {'n_estimators': 570, 'max_depth': 7, 'learning_rate': 0.0052797577098333374, 'subsample': 0.7746808767128921, 'colsample_bytree': 0.5976373765384495, 'gamma': 0.13062136736410251, 'lambda': 3.728684556334089e-07, 'alpha': 0.0047486993125976235, 'min_child_weight': 7}. Best is trial 1 with value: 0.9362333893642584.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  4\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:38,471] Trial 4 finished with value: 0.9385495054063785 and parameters: {'n_estimators': 202, 'max_depth': 8, 'learning_rate': 0.03772422362383541, 'subsample': 0.7877151952304214, 'colsample_bytree': 0.735922547889562, 'gamma': 0.07579306701012806, 'lambda': 1.610815709269123e-05, 'alpha': 8.19233708623463e-06, 'min_child_weight': 10}. Best is trial 4 with value: 0.9385495054063785.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  5\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:50,471] Trial 5 finished with value: 0.9358854363896464 and parameters: {'n_estimators': 891, 'max_depth': 8, 'learning_rate': 0.11007202291646828, 'subsample': 0.8286542801819421, 'colsample_bytree': 0.8851902127087519, 'gamma': 0.09853087697641105, 'lambda': 2.38519338277534e-05, 'alpha': 4.6743428538427764e-08, 'min_child_weight': 5}. Best is trial 4 with value: 0.9385495054063785.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  6\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:51,705] Trial 6 finished with value: 0.9391311339253695 and parameters: {'n_estimators': 110, 'max_depth': 6, 'learning_rate': 0.369124703659327, 'subsample': 0.9401069732556883, 'colsample_bytree': 0.7109187895427591, 'gamma': 0.3806574389570756, 'lambda': 3.167780550949483e-05, 'alpha': 8.976130225489175e-05, 'min_child_weight': 2}. Best is trial 6 with value: 0.9391311339253695.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  7\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 11:59:57,489] Trial 7 finished with value: 0.9418972044013983 and parameters: {'n_estimators': 368, 'max_depth': 8, 'learning_rate': 0.03357107206875864, 'subsample': 0.934323784634609, 'colsample_bytree': 0.8021900331755882, 'gamma': 0.43104549562855793, 'lambda': 5.897496138692197e-05, 'alpha': 0.41653420328823987, 'min_child_weight': 6}. Best is trial 7 with value: 0.9418972044013983.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  8\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 12:00:08,904] Trial 8 finished with value: 0.9375194364571341 and parameters: {'n_estimators': 732, 'max_depth': 5, 'learning_rate': 0.05587686576374841, 'subsample': 0.5763810578117787, 'colsample_bytree': 0.9166762657948937, 'gamma': 0.15946639083351144, 'lambda': 0.035556342018008516, 'alpha': 0.13313179151433943, 'min_child_weight': 1}. Best is trial 7 with value: 0.9418972044013983.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial:  9\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2023-10-26 12:00:17,523] Trial 9 finished with value: 0.9349622875082396 and parameters: {'n_estimators': 574, 'max_depth': 5, 'learning_rate': 0.012983699299847851, 'subsample': 0.5999618596024573, 'colsample_bytree': 0.9946464146898829, 'gamma': 0.4560172640553016, 'lambda': 2.7526006349105397e-07, 'alpha': 2.704962568461496e-08, 'min_child_weight': 8}. Best is trial 7 with value: 0.9418972044013983.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of finished trials: 10\n",
      "\n",
      "Best trial parameters: {'n_estimators': 368, 'max_depth': 8, 'learning_rate': 0.03357107206875864, 'subsample': 0.934323784634609, 'colsample_bytree': 0.8021900331755882, 'gamma': 0.43104549562855793, 'lambda': 5.897496138692197e-05, 'alpha': 0.41653420328823987, 'min_child_weight': 6}\n",
      "\n",
      "Best AUC: 0.9418972044013983\n"
     ]
    }
   ],
   "source": [
    "# Initiate the optimizer\n",
    "study = optuna.create_study(direction='maximize')\n",
    "study.optimize(objective, n_trials=10)\n",
    "\n",
    "# Results\n",
    "print(f\"Number of finished trials: {len(study.trials)}\\n\")\n",
    "print(f\"Best trial parameters: {study.best_trial.params}\\n\")\n",
    "print(f\"Best AUC: {study.best_value}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "67a2cb1d",
   "metadata": {},
   "source": [
    "#### Classification on unseen test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "1b3cbec3",
   "metadata": {},
   "outputs": [],
   "source": [
    "X_test_transformed = preprocessing_pipeline.transform(test_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "a7b79d29",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original X_train shape: (4642, 27)\n",
      "Transformed X_train shape: (4642, 94)\n",
      "y_train shape: (4642,)\n",
      "\n",
      "Original X_test shape: (1000, 27)\n",
      "Transformed X_test shape: (1000, 94)\n"
     ]
    }
   ],
   "source": [
    "print('Original X_train shape:',X_train.shape)\n",
    "print('Transformed X_train shape:',X_train_transformed.shape)\n",
    "print('y_train shape:',y_train.shape)\n",
    "print()\n",
    "print('Original X_test shape:',test_df.shape)\n",
    "print('Transformed X_test shape:',X_test_transformed.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "bcc01f79",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Predictions\n",
    "y_test_prob  = model.predict_proba(X_test_transformed)\n",
    "y_test_label = np.argmax(y_test_prob, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "8f83856a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([7, 3, 0, 3, 4], dtype=int64)"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_test_label[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "18971377",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "73e5b72e",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "120e7add",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.10.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}