Open azinoma opened 2 years ago
Purely based on observation: Try reducing the number of training samples N
and see if it makes any difference.
Can confirm. For me it seemed to work if i rerun optimize!(gp)
a few times. Not exactly sure what is happening behind the curtains.
Dear all,
I am trying to use exact GPs and the hyperparameter optimization keeps producing errors, such as
AssertionError: isfinite(phi_c) && isfinite(dphi_c)
. Consider this minimal example:N = 100
X = rand(3, N)
y = sin.(X[1,:]) .* cos.(X[2,:]) .+ 2. .* X[3,:]
kern = Matern(5/2,[0.0, 0.0, 0.0],0.0) + SE(0.0, 0.0)
m = MeanZero()
gp = GP(X,y,m,kern)
optimize!(gp)
Am I doing something wrong, or is there an error?
Many greetings, Marlon