Kaggle Playground Series 3, Episode 4 competition

NOTES:

• During first training, catboost performed the best, follwed by logistic regression and lasso regression. XGBoost and LGBM performed worse. Public leaderboard agrees.

Tasks to obtain the best model:

• Basic eda
• Cross-validation
• First training
• Detailed eda
• Feature engineering
• Use ten folds
• Use AutoGluon framework
• Merge original dataset
• Hyperparameter tunning
• Ensemble of best tunned models
• Ensemble of multiple algorithms
• Ensemble of multiple seeds, ¿or without setting it?
• Predict with fastai tabular

cd src
conda activate ml
python create_folds.py
python -W ignore train.py [--model=lgbm]  # [rf|svd|xgb|lgbm|cb]

Submission is stored in outputs folder (see config.py for complete path)