Open sleepingPhD opened 1 year ago
Hello, We haven't got to parameter optimization yet, partly because EFGP doesn't provide a fast algorithm for likelihood evaluation. However, as we mention in the preprint, cross-validation on held-out data is still possible. Since there are only 2-4 parameters (\sigma, \ell, maybe \nu, maybe k(0)), simple finite-differencing could be used to estimate parameter-derivatives of the held-out RMSE. Another idea is use GPyTorch to estimate parameters on a subset of data, then use EFGP for the large-scale problem. That's all I recommend for now. Maybe you can invent a faster way to optimize parameters? Best, Alex
On Fri, Jul 7, 2023 at 8:17 AM sleepingPhD @.***> wrote:
Is it possible to optimize the hyperparameter in Matlab? Or hou can I just calculate them before regression?
— Reply to this email directly, view it on GitHub https://github.com/flatironinstitute/gp-shootout/issues/4, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACNZRSXSNKFB5SB7G4A3IOTXO745NANCNFSM6AAAAAA2BXTPA4 . You are receiving this because you are subscribed to this thread.Message ID: @.***>
-- *-------------------------------------------------------------------~^`^~._.~' |\ Alex Barnett Center for Computational Mathematics, Flatiron Institute | \ http://users.flatironinstitute.org/~ahb 646-876-5942
Is it possible to optimize the hyperparameter in Matlab? Or hou can I just calculate them before regression?