Kaggle competition https://www.kaggle.com/competitions/cmi-detect-behavior-with-sensor-data/ 1st place solution by Devin | Ogurtsov | zyz team

Training notebooks from public_solution_ogurtsov could be mapped to inference notebook by experiment number in file name, e.g. exp_339. _comb in file name means "for this model best result per fold was picked among 3 runs" (it's not crucial part of the solution, results from single run play almost equally well in the final ensemble).

Devin's training code source: github repo

All parts of the solution are distributed under LGPL 2.1 License according to competition rules.