Hello zindians. I would love to know what cross validation methods you guys used and if they were reliable. Especially from the first team. They really did an amazing job. They had few submissions meaning their local was good enough to do their testing and most amazingly low score.
We tried different cross validation strategies with your usual gradient boosted decision trees (XGBoost, LightGBM and CatBoost), and it didn't work quite well (at least for us). We had to rethink the entire approach and instead focused on multi-variate time series forecasting. I'm confident we'll be able to share our approach before the end of the week.