Max Balandat
Max Balandat
I have been messing with this recently to enable this for a certain kind of budget constraints where we can generate the one-shot initial conditions by doing some sampling from...
@Muhammad-Rebaal sure thing, definitely a good idea to raise an informative error for now. Thanks for working on this!
Actually, @Muhammad-Rebaal, looks like this was already done recently in #2775. Let me close out this issue.
We don't have a dedicated community channel at this point. In the past we have coordinated with contributors on the pytorch slack for specific (and more complex) issues, but we...
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...
@sdaulton did you ever look into implementing MOO PoI? For non-batch (`q=1`) versions this should not be too bad (potentially using some of @j-wilson's MVN CDF work). MC sampling will...
@Ruan-Yixiang were you able to try this? Would be great if you could share your findings here.
FYI here is a (non-MOO) implementation of qPI: #1684
Thanks for raising this issue. > I am getting solutions [...] with many values close to 2pi and 0.0. This is caused by the model having larger uncertainty at the...