Recomputes the (Gaussian) noise variance.
[newLikPar] = em_gauss(oldLikPar,y,cavM,cavV,postM,postV);
[newLikPar] = em_gauss(oldLikPar,y,cavM,cavV,postM,postV)
- implements the
M-step in the GP inference assuming Gaussian additive noise. The data is
assumed to be factorising, thus we have a direct result as
newLikPar = sum(postV + (y-postM).^2) / N
where N
is the number of training samples.
Parameters:
oldLikPar
- preceeding value of the likelihood parameter.
y
- training output.
cavM,cavV
- (cavity) mean and variance of the GP.
postM,postV
- mean and variance of the posterior process.
newLikPar
- the new noise variance.
ogptrain
, demogp_reg
, c_reg_gauss
, g_l_gauss
Copyright (c) Lehel Csató (2001-2004)