@kizill kizill released this Oct 11, 2018 · 1411 commits to master since this release

Assets 8

Changes:

  • Added EvalResult output after GPU catboost training

  • Supported prediction type option on GPU

  • Added get_evals_result() method and evals_result_ property to model in python wrapper to allow user access metric values

  • Supported string labels for GPU training in cmdline mode

  • Many improvements in JNI wrapper

  • Updated NDCG metric: speeded up and added NDCG with exponentiation in numerator as a new NDCG mode

  • CatBoost doesn't drop unused features from model after training

  • Write training finish time and catboost build info to model metadata

  • Fix automatic pairs generation for GPU PairLogitPairwise target