credal_bnn¶
- probly.method.credal_bnn(base: Predictor[In, Out], use_base_weights: bool = False, posterior_std: float | list[float] | None = 0.05, prior_mean: float | list[float] | None = 0.0, prior_std: float | list[float] | None = 1.0, num_members: int = 5) BayesianEnsemblePredictor[In, Out][source]¶
Create a Credal BNN predictor from a base predictor based on [CDJ+24].
- Parameters:
base – The base model to replicate into a Bayesian ensemble.
use_base_weights – If True, the weights of the base model are used as the prior mean. Default is False.
posterior_std – Initial posterior standard deviation(s). Default is 0.05.
prior_mean – Prior mean(s). Default is 0.0.
prior_std – Prior standard deviation(s). Default is 1.0.
num_members – Number of BNN ensemble members. Default is 5.
- Returns:
The credal BNN predictor outputting a convex credal set.
