Computes the log-predictive probability for Gaussian likelihood
[allLog] = logp_g(likpar,y,mX,sigX2);
[allLog] = logp_g(likpar,y,mX,sigX2)
- returns the average
log-predictive probabilities of the inputs -- assuming additive Gaussian
noise.
It assumes that the test samples are independent, thus ignoring any possible non-diagonal part in the joint posterior distribution.
Parameters:
likpar
- the current likelihood parameters.
y
- desired output.
mx
- mean of the GP marginal at the new input.
allLog
- the average error.
ogp
, ogptrain
, demogp_reg
, err_mse
Copyright (c) Lehel Csató (2001-2004)