v.19.1New Features
Added Support for Catboost Multiclass Models Evaluation
Added support forCatBoostmulticlass models evaluation. FunctionmodelEvaluatereturns tuple with per-class raw predictions for multiclass models.libcatboostmodel.soshould be built with #607. #3959 (KochetovNicolai)
Why it matters
This feature enables users to perform multiclass model evaluation directly within ClickHouse. It addresses the need to handle multiclass classification outputs by providing detailed per-class raw prediction scores, enhancing model analysis and inference capabilities.How to use it
Ensure that the CatBoost librarylibcatboostmodel.so is built with the changes from CatBoost PR #607. Then use the modelEvaluate function in ClickHouse to evaluate multiclass CatBoost models and obtain per-class raw prediction tuples.