Open raff7 opened 4 years ago
That's a great question, we've thought about various ways of doing this over the years. I don't think there is one obvious route. We'd be very open to suggestons.
One effort we have made in that direction is Emukit, which is, to a certain extent, agnostic to a modelling framework, so it is possible to use with models defined in Tensorflow or PyTorch. It is close to a feature parity with GPyOpt, but user is supposed to define a model themselves. Have a look, perhaps it works out for your case?
As Andrei has mentioned Emukit seems a better place to do this since you can plug-in your model built in your favorite framework. Emukit uses numpy to communicate between modules though, so if you wanted to link the model and the acquisition to use autodiff end-to-end you would need to implement both in the same framework. If I am not wrong this would give you GPU support.
El lun., 3 feb. 2020 a las 11:28, Andrei Paleyes (notifications@github.com) escribió:
One effort we have made in that direction is Emukit https://github.com/amzn/emukit, which is, to a certain extent, agnostic to a modelling framework, so it is possible to use with models defined in Tensorflow or PyTorch. It is close to a feature parity with GPyOpt, but user is supposed to define a model themselves. Have a look, perhaps it works out for your case?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/SheffieldML/GPyOpt/issues/293?email_source=notifications&email_token=ABUJF456UBYEP54CWO4K7T3RA75XLA5CNFSM4KOHIL5KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEKTPU2A#issuecomment-581368424, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABUJF46OGDLDUC7LLDKTQGTRA75XLANCNFSM4KOHIL5A .
Are you considering implementing GPU support for this package? either using GPyTorch instead of GPy or by implementing GPU support for all GPy kernels?