keirkwame
commented
5 years ago Re large training sets and Gaussian processes, I am thinking at the moment about clustered GPs that could be a good solution, which essentially train multiple GPs on various subsets of the training data making each individual training process tractable.
From: Andreu Font-Ribera notifications@github.com Sent: Wednesday, May 29, 2019 2:26:16 PM To: andreufont/LyaCosmoParams Cc: Subscribed Subject: [andreufont/LyaCosmoParams] Try GP from celerite (#19)
Boris suggested to use Celerite instead of GPy. He claims that it would be able to handle much larger training sets, but I wasn't able to figure out whether it would be useful in our case.
https://celerite.readthedocs.io/en/stable
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/andreufont/LyaCosmoParams/issues/19?email_source=notifications&email_token=ACJBFUPPU3SCHVJO3EAUJ7DPXZY6RA5CNFSM4HQMHT72YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4GWPB5PA, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ACJBFUKJ4PTROQHMNK2FPJTPXZY6RANCNFSM4HQMHT7Q.
Boris suggested to use Celerite instead of GPy. He claims that it would be able to handle much larger training sets, but I wasn't able to figure out whether it would be useful in our case.
https://celerite.readthedocs.io/en/stable