Defining a Model class for ModelListGP about ax

[bernardbeckerman]

Thanks for asking this! Let me follow up internally to see who might be best to answer.

[saitcakmak]

A small correction on the SurrogateSpec suggestion: You do not want to define multiple surrogate specs. Multiple surrogates is intended only for some use cases that we never got around to implementing. To make sure the surrogate utilizes ModelListGP, you can use the following:

                            model_kwargs={
                                "botorch_acqf_class": qMultiFidelityHypervolumeKnowledgeGradient,
                                "surrogate_specs": {"model_list": SurrogateSpec(allow_batched_models=False)},
                            },

[Balandat]

So the issue here is that by default Ax will select a (batched) non-model-list model that is incompatible with the decoupled acquisition function that qMultiFidelityHypervolumeKnowledgeGradient is a subclass of.

There is a way to avoid using batched models by passing in allow_batched_models=False as part of the surrogate specs of the model (a way of specifying model_kwargs that allow finer granular control). This would look something like passing the following (as part of model_kwargs:

"surrogate_specs": {"a": SurrogateSpec(allow_batched_models=False), "b": SurrogateSpec(allow_batched_models=False)},

However, I'm running into some weird issues with this, so I for now just hacked my way around it, see #2514 (comment) - that again surfaced a more serious limitation, see that comment.

[Abrikosoff]

Let me test this out @saitcakmak , thanks a lot!

[lena-kashtelyan]

@Abrikosoff, did @saitcakmak's solution work out? Please close the issue if you are all set : )