Why can deeper oblivious trees make CatBoost SHAP computation expensive even if inference remains acceptable?
Classic MLLevel: IntermediateCatBoost · SHAP values
Why can deeper oblivious trees make CatBoost SHAP computation expensive even if inference remains acceptable?
Hints
- Prediction and explanation execute different algorithms.
- Inspect the depth term in the complexity expression.
Answer
The documented SHAP complexity includes a factor exponential in tree depth for the stated oblivious-tree regime, in addition to samples and trees. Benchmark explanation latency separately and limit the explanation batch or model complexity if necessary.
Source and licence: CatBoost · SHAP values · Apache-2.0 · answer: Quantoros editorial