SingleTaskMultiFidelityGP model

[S3anaaaaz]

Hi,
Why in SingleTaskMultiFidelityGP you are using 4 different Materns with different priors (Gamma (3,6) and Gamma (6,2))?
What is the point of summing up all those kernels?
I also figured out that I cannot fit a good model with his surrogate on even for example Rosenbrock function (Multi-fidelity version) compared to other GPs. attached a figure of that. Why does it train very bad?

[Balandat]

Why in SingleTaskMultiFidelityGP you are using 4 different Materns

See here: https://github.com/pytorch/botorch/blob/main/botorch/models/gp_regression_fidelity.py#L41-L45

Why in SingleTaskMultiFidelityGP you are using 4 different Materns

You'll need to provide a bit more information here - what are you fitting the model to and what are you predicting? Can you please attach the code to repro?

[S3anaaaaz]

Thank you for your answer. I'm using the Wing example as an input and I separated a part of the data for prediction. The attached code is the comparison for 4 models : ExactGP, LMGP, SingletaskGP and Single-task-multifidelity-GP. As you see, ExactGP and LMGP work well, but the other ones cannot fit a good model with that data, I should increase data to improve their performance.

…

On Mon, Jul 4, 2022 at 1:24 PM Max Balandat ***@***.***> wrote: Why in SingleTaskMultiFidelityGP you are using 4 different Materns See here: https://github.com/pytorch/botorch/blob/main/botorch/models/gp_regression_fidelity.py#L41-L45 <https://urldefense.com/v3/__https://github.com/pytorch/botorch/blob/main/botorch/models/gp_regression_fidelity.py*L41-L45__;Iw!!CzAuKJ42GuquVTTmVmPViYEvSg!K45gMglwgh0aq82dJ1TO73vH04QfPcmTwwI3AbTcJPpmBuTq8macHJbapF13tCSW7EywN36AiJtjviuJH4uzPPuG$> Why in SingleTaskMultiFidelityGP you are using 4 different Materns You'll need to provide a bit more information here - what are you fitting the model to and what are you predicting? Can you please attach the code to repro? — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/pytorch/botorch/discussions/1284*discussioncomment-3080248__;Iw!!CzAuKJ42GuquVTTmVmPViYEvSg!K45gMglwgh0aq82dJ1TO73vH04QfPcmTwwI3AbTcJPpmBuTq8macHJbapF13tCSW7EywN36AiJtjviuJH4ojSTyP$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AY2WGMNPSVR7SPF7PBRUEADVSNCAVANCNFSM52OGDQNQ__;!!CzAuKJ42GuquVTTmVmPViYEvSg!K45gMglwgh0aq82dJ1TO73vH04QfPcmTwwI3AbTcJPpmBuTq8macHJbapF13tCSW7EywN36AiJtjviuJHxJmQe9a$> . You are receiving this because you authored the thread.Message ID: ***@***.***>