Hi,
A customer was asking about the pros and cons of the two different correlation functions in Guassian Process Models. The help documentation tells us that the cubic function is a generalisation of the cubic spline and that it allows correlation to be zero at certain distance, whereas guassian is always non-zero.
Any thoughts on the practical implications of this? When would you use gaussian versus cubic?
I had a quick look at a simple example and found the variance between points was greater for cubic. Should we expect that?
Thanks,
Phil
