Matthew Carbone

@saitcakmak thank you, I'll give your hack a try for now!

@saitcakmak so just to be clear as to what you mean here. The idea would be to create an entirely new model with all of the same required objects (e.g....

@saitcakmak the problem though is that when initializing using what you show above, the correct hyper parameters are set for the transforms, but not the length scales. When you do...

@saitcakmak Your suggestion does work, but I'd like to bring another set of weird behaviors to your attention: ```python print(train_x) """ tensor([[0.0000], [0.7143], [1.4286], [2.1429], [2.8571]]) """ print(train_y) """ tensor([-1.6720,...

Ahh so that's what `_original_train_inputs` is for. I never noticed that it doesn't change during transforms. Ok all of this sounds good. Looking forward to the botorch+gpytorch update!

No problem, glad I could help!

@Balandat I'm digging up a really old PR here but it seemed like the right place to ask after doing a bit of searching for the right topic. It appears...

@CompRhys ah of course. This is what happens when I comment on PRs after just waking up. Thanks for clarifying!

I'm trying to get approval to contribute upstream to BoTorch (re: CLA) so if I end up getting the green light this would definitely be something I'd be interested in...

@sgbaird If you'd like, I can ask Kris [Reyes] about this next time I meet with him, we chat ~weekly 😁 I'm slowly being convinced that kernel flexibility is important....