Under CatBoost Bayesian bootstrap, what happens to positive random weights a^t when baggingtemperature t is zero?
Classic MLLevel: IntermediateCatBoost · Bootstrap
Under CatBoost Bayesian bootstrap, what happens to positive random weights a^t when bagging_temperature t is zero?
Hints
- Evaluate the exponent before interpreting the training behavior.
- One randomness control is not a global determinism switch.
Answer
They become one, so the Bayesian reweighting disappears. This does not turn off every other source of model randomness: split-score noise and permutations are separate mechanisms.
Source and licence: CatBoost · Bootstrap · Apache-2.0 · answer: Quantoros editorial