Team Eeveelution: Pokémon Battle Prediction

This repository contains the code for a machine learning pipeline designed to predict the outcome of Pokémon battles.

The framework utilizes CatBoost, LightGBM, and XGBoost as its core models, and is explicitly designed to produce submissions and files suitable for ensembling (stacking/blending).

Repository Guide

Below is a brief description of the main files and folders and their purpose.

Main Folders

• /CatBoost_Data_Pipeline: This is the output folder for all intermediate .csv files. It contains the processed data, engineered features, and split datasets (train/validation/holdout) ready to be used by the CatBoost model.

• /*_Model_Outputs: Saves all analysis results for each model. For CatBoost, it includes diagnostic plots (e.g., Feature Importance, ROC/AUC, Confusion Matrix), classification reports, and the .json files with optimal parameters and iterations. For all others, it includes the preprocessing_medians and final_features.json.

• /Meta_Model: Saves the models that generated the stacking, including both LogReg and the BEST_model (chosen from LGBM, LogReg, Ridge, VotingSoft, VotingHard).

• /OOF_Predictions: A crucial folder for ensembling. It contains the final predictions for both the training set (Out-of-Fold) and the test set, saved as .npy files.

• /Submissions: Contains the submission.csv files, both those useful for blending (based on probabilities) and the final blended and stacked ones.

Main Files

• utils_*.py: The "brain" of the entire pipeline. These files contain all the logic, utility functions, paths, and definitions for each phase of data loading, feature engineering, validation, training, and blending/stacking.

• Report_Team_Eeveelution.pdf: The PDF report file related to the challenge.