search

SheffieldML / GPy

Gaussian processes framework in python
BSD 3-Clause "New" or "Revised" License
2.01k stars 557 forks source link

cross validation with GPy #797

Open chengjun opened 4 years ago

chengjun commented 4 years ago

how to do cross validation with GPy? is it possible to do it within the framework of sklearn? your reply would be highly appreciated. thanks!

MikeHeiber commented 4 years ago

It would be great if there was example documentation of how to wrap a GPy model with an sklearn estimator so that you can use all of sklearn's cross validation tooling to do model scoring/comparison and then tune the GPy model hyperparameters.

lionfish0 commented 2 years ago

Off topic: No idea if this is useful, but there's a trick that lets you do LOO x-validation without having to recompute/invert the whole covariance matrix each time (as long as you don't mind that the hyperpamaters won't change). When I first joined Sheffield I had a vague look at this (but I was too new to GPs, etc, to really help with it). There's a branch floating about in GPy where this was implemented... https://github.com/SheffieldML/GPy/tree/LOO - not sure that's very useful. But anyway, I thought it might be relevant. Mike.

PS It would be nice to see an example using the sklearn framework. If you've got it working, maybe stick it in a colab or gist or something?

On Mon, 21 Feb 2022 at 10:28, Abdulmomen Ghalkha @.***> wrote:

Have you guys solved this issue?

— Reply to this email directly, view it on GitHub https://github.com/SheffieldML/GPy/issues/797#issuecomment-1046718903, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB4MGQFMAEMW6BQ2D7XPU5TU4IHT7ANCNFSM4JPNIADQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You are receiving this because you are subscribed to this thread.Message ID: @.***>