CatBoost is a depth-wise gradient boosting library developed by Yandex. The algorithm grows a balanced tree using oblivious decision trees.
It uses the same features to make the right and left split at each level of the tree.
For example in the image below, you can see that
297,value>0.5 is used through that level.
Other notable features of CatBoost include:
- native support for categorical features,
- supports training on multiple GPUs,
- results in good performance with the default parameters,
- fast prediction via CatBoost’s model applier,
- handles missing values natively,
- support for regression and classification problems.