Max Balandat

Results 476 comments of Max Balandat

Hmm interesting. Yeah not sure why the docs contain this, seems like this was never implemented. Support is there for many unary or binary operators, but `torch.cat` operates on a...

Yes, that makes a lot of sense and would be great to have. I'm not sure if that is easy to do with the `__torch_function__` setup that we leverage for...

Thanks for the suggestion, @hughsalimbeni! @gpleiss, @jacobrgardner and I have talked in the past about expanding `linear_operator` beyond the current focus on square (really, symmetric PSD) matrices. The `BlockDiagLinearOperatorNonSquare` extending...

> it would actually be symmetric tensors with shape `channel*width*height*channel*width*height` Do you mean `channel x (width*height x width*height)` or `(channel*width*height) x (channel*width*height)`? I.e. is this a covariance per channel or...

Looks like there may be a number of places where negative indexing isn't properly supported. I'll put up a fix for the ` CatLinearOperator` case, but this should probably be...

As to the changes here, they look good in isolation.

Similar considerations arise when passing arguments to the acquisition function constructors, see https://github.com/facebook/Ax/issues/2401?fbclid=IwZXh0bgNhZW0CMTAAAR3SSZxhZ_hdjkn5IdT2A5O5b8aP_isVXoxRsF_dEbeOHc5V2QyQTL4fd2M_aem_AZkcTeoS_8xcLndEm7ZcXeyhYwA9qC3o6FLOGVyUNuqch7oOKUchFM4VKxO_xXg6ZHgr2ONViqthDU6O8h-jAKoG#issuecomment-2079566395. It would be great if we could automatically apply the transforms to the arguments for which they...

This also broke some linux unit tests for us: https://github.com/pytorch/botorch/actions/runs/9253409995/job/25453018453 Note that `distutils` has been deprecated since py3.10 and was removed in py3.12: https://peps.python.org/pep-0632/

@tanguyen77 how would you envision this to work? Should this produce a 3D plot (I assume this will often be very hard to read / understand without making it interactive...