Skip to content

fellowship/benchmarks

 
 

Repository files navigation

Benchmarks

This repo contains different CatBoost benchmarks.

Quality: comparison with other libraries

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.

Training speed: comparisons with other libraries

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

Training speed: CPU vs GPU

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

Applier speed: comparison with other libraries

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

Ranking: compare quality of different GBDT libraries and different modes

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

SHAP values calculation speed: comparison with others

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.

Kaggle

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

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 93.8%
  • Python 4.8%
  • Other 1.4%