@kizill kizill released this Dec 10, 2018 · 802 commits to master since this release

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 for Plain 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)
Assets 8