baybe.surrogates.custom

Functionality for building custom surrogates.

Note that ONNX surrogate models cannot be retrained. However, having the surrogates raise a NotImplementedError would currently break the code since baybe.recommenders.pure.bayesian.base.BayesianRecommender assumes that surrogates can be trained and attempts to do so for each new DOE iteration.

It is planned to solve this issue in the future.

Functions

register_custom_architecture([...])

Wrap a given custom model architecture class into a `Surrogate`.

Classes

CustomONNXSurrogate

A wrapper class for custom pretrained surrogate models.