Can featureweights in XGBoost force a feature to have a larger prediction coefficient?
Classic MLLevel: IntermediateXGBoost · Parameters
Can feature_weights in XGBoost force a feature to have a larger prediction coefficient?
Hints
- Where in the algorithm are features sampled?
- Selection probability is not model contribution.
Answer
No. With column sampling, these weights influence selection probabilities, not fitted leaf values or coefficient signs. Validate any prior preference because it changes which splits the model can consider.
Source and licence: XGBoost · Parameters · Apache-2.0 · answer: Quantoros editorial