Sait Cakmak
Sait Cakmak
Ah, you're right. I always find this parameter / raw parameter setup confusing. Going back through code history, looks like the noise constraint on the likelihood was added by @Balandat...
https://github.com/pytorch/botorch/pull/2544 would remove `transform=None` across the board. I'll check with others to see if there are any concerns with removing this.
I came across this again yesterday and looked into the GPyTorch constraints a bit more. Seems like we could disable the application of constraints by mocking a simple property: ```...
cc @jelena-markovic
Gaussian noise is a common assumption we make throughout the code base, so I think it is ok to continue with that.
Yeah, I am guessing any fix for this will involve some changes on PyTorch side.
Hi @jduerholt. That seems reasonable overall but I am curious about one aspect of the approach. Why model using one-hot encoded features, rather than using something like `CategoricalKernel`? If you're...
I guess one could also use the `IndexKernel` here, instead of `CategoricalKernel`. That one uses an embedding under the hood to compute distances, which seems like it'd work just as...
> it would be fine for you if I would file a PR with some kind of generic NumericToEncoding input transform? I'm reluctant to add code to BoTorch unless we...
This was fixed in https://github.com/pytorch/botorch/pull/2269