Project to predict the outcome of Pokémon battles from the early phases of a match. Raw data (jsonl files) are loaded into a SQLite DB, preprocessed, and used to train/evaluate various ML models.

1. Creates a DB connection and registers a new Dataset (Train or Test).
2. For each match in the JSONL:
   • Inserts match metadata via insert_battle.
   • Inserts team Pokémon with load_pokemon and load_team.
   • Inserts turns with insert_turn, which in turn calls insert_state_move to save states and moves.
3. Maintains consistency with INSERT OR IGNORE for reference tables (type, status, moves).
4. Final commit of the data into the DB.

The data_analyzer/ folder contains code to extract, preprocess, select models and run experiments:

• data_analyzer/init.py — exports main functions from lib.py.
• data_analyzer/lib.py — utilities and preprocessing / I/O pipeline:
  • get_datapoints — builds the dataset with all normalized features (as described in the report).
  • load_datapoints — reads preprocessed tables (Input, Output, TestInput, TestOutput).
  • create_submission — generates submission CSVs from predictions.
  • load_best_model — reconstructs a model from the information in models.json.
• data_analyzer/model_selection.py — classes and helpers for hyperparameter search and validation:
  • ModelTrainer and various implementations (LogisticRegressionTrainer, RandomForestClassifierTrainer, XGBClassifierTrainer, etc.) for cross‑validation and hyperparameter search.
  • plot_history — saves validation plots.
• data_analyzer/main.py — CLI for analysis operations (save dataset, PCA, training, ensemble, etc.). Runs routines that use functions from lib.py and model_selection.py.

• main.py contains the entire pipeline to reproduce the results; it was later converted into a notebook for the Kaggle challenge.

• Main script: main.py
• DB creation: analisi/create_db.sql
• LaTeX report template: Latex/
• Saved models info: models.json
• PCA importance/features: pca.json