These are the codes I written for reaching 27th place (top 7%) in week 55 of Numerai hedge fund competition.

This is a three-level stacking model. The first two levels are xgboost and lightgbm, while the last level is simple linear regression. All training and stacking are done with K-fold cross-validation to minimize information leakage. Feature engineering and parameter tuning are performed probabilistically through Bayesian optimization.