Changes:
- Pure GPU implementation of NDCG metric
- Enabled LQ loss function
- Fixed NDCG metric on CPU
- Added
model_sum
mode to command line interface - Added SHAP values benchmark (#566)
- fixed
random_strength
forPlain
boosting (#448) - Enabled passing a test pool to caret training (#544)
- Fixed a bug in exporting the model as python code (#556)
- Fixed label mapper for multiclassification custom labels (#523)
- Fixed hash type of categorical features (#558)
- Fixed handling of cross-validation fold count options in python package (#568)