I’m interested in setting up a search space/alg where we perform grid search on a conditional search space. The tricky thing is, my conditional search space includes certain parameters which are only defined if a given other parameter has a certain option. For example, one parameter is model_type, and depending which value we’ve sampled for it, the set of conditional parameters we must sample will change.
I’ve looked at this guide: Working with Tune Search Spaces — Ray 2.4.0, but it only shows an example for conditional parameters which are always sampled, just on different distributions.
I’ve also personally set up hyperopt search spaces which have worked with Tune and which do the type of sampling that I want: Hyperopt Conditional Search Space Example — Ray 2.4.0. However, I don’t think hyperopt is compatible with “grid_search” parameters.
Has anyone else run into this or found a solution?