Recomputes the likelihood parameter for pos.-exp. noise.
[newLik] = em_exp(oldLikPar,y,cavM,cavV,postM,postV);
[newLik] = em_exp(oldLikPar,y,cavM,cavV,postM,postV)
-the M-step in lik.par.
estimation assuming one-sided exponential noise. The data is assumed to be
independent, and the EM approximation to the new value is
newlik = sum(avg(Theta(y-f),postM,postV))/N;
where Theta
is a function that sets the negative values of its
argument to zero and avg
is the average operator with respect to a
Gaussian.
Parameters:
oldLikPar
- the preceeding value of the lik.-par..
y
- desired output.
cavM,cavV
- (cavity) mean and variance of the GP.
postM,postV
- mean and variance of the posterior process.
newLik
- the new parameter.
demogp_reg
, ogp
, ogptrain
, c_reg_exp
Copyright (c) Lehel Csató (2001-2004)