Question: How to initialize a model in 4D search space, but run trials in only 2D for the first 100 trials through ask-tell interface. Then expand to 4D.

[GregorySchwing]

Use case: quickly search the 2D space and evaluate ability to minimize a loss function while holding 2 of the variables constant. After finding the Pareto front of 2D space, allow the model to search full space to see if a better point exists.

How would I do this? I am running batched ask-tell with checkpointing of the model after every batch of ask-tell. Is it possible to load the checkpointed model, change the domain of the last two variables from constant to a range, and then continue with ask-tell evaluation?

Thanks