Gaussian Process -Statistical rethinking

[roesta07]

Hi @fonnesbeck

What does the input data look like here? Is it standardized?

Yes the distance matrix Dmat is standardized where maximum value is 1 for this particular example.

Also, we have a built-in Linear mean function already, so no need to code your own.

Great thanks; this works fine.

Just for fun, try using half-normals instead of taking the absolute values of a normal (then add your offsets). Probably not the issue, but can’t hurt.

Does it mean like
1 + HalfNormal("etasq",0.25) instead of pm.abs(pm.Normal("etasq,1,0.25))?

Also went into another issue, while exploring (link below), As I believe that using PYMC’s expEquad kernel function (below equation) is providing different estimates than when calculated manually! resulting in some loss in precession but I might have mis-specify something here…

Comparing Manual and PYMC ExpQuad Kernel Function : DIFFERENT RESULTS

Hi all; I am trying to explore Pymc’s Gp class. and I am having a hard time replicating Mclearth’s estimates from Gaussian Process (island example). As I believe that using PYMC’s expEquad kernel function (below equation) is providing different estimates than when calculated manually, which led to loss in precision. This is ExpEquad kernel function which I am trying to replicate… [latex_kernel] USING PYMC’S ExpQuad # Dmat is some Distance Matrix ; source link below ls=1 ## lets assume ls …