This repo contains different CatBoost benchmarks.

Go to subdirectory quality benchmarks to see quality benchmarks. These are benchmarks in binary classification mode. They compare CatBoost vs XGBoost vs LightGBM vs H20.

You can find scripts to run LightGBM/XGBoost/CatBoost CPU and GPU versions and compare its runtime in training speed subdirectory

This benchmark shows speedup of GPU over CPU on different dataset sizes and on different devices.

Benchmarks with comparison of applier speed with other libraries are in folder model evaluation speed

This benchmark shows how different libraries and modes perform on existing open source ranking datasets.

Shap values calculation benchmarks are in shap speed subdirectory. This benchmark will show the complexity of SHAP calculation for each library. And will show a speed comparison on a fixed dataset.

This is the folder where we are adding quality comparisons on some kaggle datasets. Currently, it only contains comparison of different libraries on Rossman store sales competition.