Code for the private Google competition: https://www.kaggle.com/competitions/ml-competition-2024-for-ukrainians/leaderboard
lgmb-0-70.ipynb - get the best private score of 0.70199. In general, for LGBN param tuning, I suggest using LGBM Tuner described here: https://medium.com/optuna/lightgbm-tuner-new-optuna-integration-for-hyperparameter-optimization-8b7095e99258
Generating additional samples with VAE hasn't improved the score, but it was generally interesting.