GPSubspacePredSVD: 44ms, w/o messaging (k=20,l=12). GP subspace regression...

In rudazhang/gpsr: Gaussian Process Subspace Regression

44ms, w/o messaging (k=20,l=12). GP subspace regression prediction, posterior from equal likelihood

Description

44ms, w/o messaging (k=20,l=12). GP subspace regression prediction, posterior from equal likelihood

Usage

GPSubspacePredSVD(thetaNew, thetaTrain, len, t = k, method = "rsvd")

Arguments

thetaNew	a target parameter point
t	truncation level, defaults to k; if NULL, use thin SVD.

Details

To get the explicit principal directions, do matrix multiplication V = stdX$u %*% Vcirc.

Value

a list: Vcirc, principal directions as coordinates in the global basis, a c-by-t matrix (more than the mean prediction); sigma2, principal variances, a vector of length t; eps2, noise variance, a scalar.

Note

Require (K, XtX, VbtX)

rudazhang/gpsr documentation built on March 11, 2023, 5:49 p.m.