Max Balandat

@hvarfner you have done some evaluations here - any thoughts? My understanding was that we didn't really see any benefit from using a scale kernel across a variety of functions...

Thanks for raising this! Seems like an interesting advanced use case of BoTorch, would be great if we could properly support it. > one can either (a) enumerate all n_combos^n...

> For me, this would be an interesting feature, that opens up new opportunities. @Balandat what do you think? @jduerholt yes, I think this could be useful. I'm not sure...

> What do you think of the bold adjustments below to optimize_acqf_mixed? @MoritzEckhoff So this is basically a greedy solution with respect to the constraints. How would we ensure that...

@Mustafaessou what exactly is the error you're running into? If this still persists could you please open a new issue that includes the error + stack trace and a full...

Thanks @seashoo for the PR - this is a big one! It'll take me a bit of time to review this in detail, I plan to do a first higher-level...

Thanks for all the updates, @seashoo. Sorry for the delay on the review; I've been traveling for a conference and taking some time off. I should be able to provide...

Thanks! > merging with BoTorch's upstream is causing dependency issues with GPyTorch imports on my local. I will investigate this issue further and commit the working upstream in the coming...

Computing the product of the individually modeled outcomes makes sense, this is a case of BO for composite functions (https://proceedings.mlr.press/v97/astudillo19a.html). I guess one could come up with some set of...

@hvarfner curious if you have any thoughts on this PR re the information gain aspects