Known limitations

  • Windows is not supported. Work in progress.
  • GPU is not supported. Work in progress.
  • Text and embeddings features are not supported. Work in progress.
  • Feature distribution statistics (like calc_feature_statisticson CatBoost python package) with datasets on Spark is not supported. But it is possible to run such analysis with models exported to local files in usual CatBoost format.
  • Generic string class labels are not supported. String class labels can be used only if these strings represent integer indices.
  • boosting_type=Ordered is not supported.
  • Training of models with non-symmetric trees is not supported. But such models can be loaded and applied on datasets in Spark.
  • Monotone constraints are not supported.
  • Multitarget training is not supported.
  • Stochastic Gradient Langevin Boosting mode is not supported.