Max Balandat

Thanks for flagging this. Yes, there are currently some unfortunate inconsistencies in how transforms are being (are are not being) applied. We generally use `condition_on_observations()` under the hood as part...

Fundamentally, you need to implement this in a way that's compliant with the basic [`Model`](https://github.com/pytorch/botorch/tree/main/botorch/models) API. You need to implement a ``posterior()`](https://github.com/pytorch/botorch/blob/main/botorch/models/model.py#L81) method that returns a [`Posterior`](https://github.com/pytorch/botorch/blob/main/botorch/posteriors/posterior.py#L19) API. Since you're...

It appears your questions are mostly about the MLP with dropout and not so much about botorch, so I'm not sure how much I can help you with this. >...

Yes, this is a pretty common occurrence in practice. As @jakobzeitler indicates above, we do consider this a problem of transfer learning with heterogeneous search spaces. We do have models...

Hmm, interesting, thanks for flagging this. I'm not entirely sure what causes this, but it's got to be something in the implementation-weeds. One thing to note is that MVNXPB is...

Interesting. Impressive improvement from logEI, but I agree with @saitcakmak that these performance comparisons aren't the best fit for the tutorials. I think if we do have them then we...

Regarding 1, it's not hard in principle to sample uniformly from the simplex (we implement this in botorch here: https://github.com/pytorch/botorch/blob/main/botorch/utils/sampling.py#L187-L225). Things get a bit more complicated when there are additional...

> I found that once you increase the search space to ~8 parameters you get SearchSpaceExhausted. The example in your post only has 5 parameters - I also saw that...

Yeah definitely rejection sampling to blame. Using `fallback_to_sample_polytope` should be the quickest way around, see my comment here: https://github.com/facebook/Ax/issues/2510#issuecomment-2160828417

Hmm interesting. At least in the past there must have been some reason to not just use `MultitaskMultivariateNormal.from_batch_mvn()` here. I fail to recall why exactly without digging into this more...