No Weighting ? about gpyrn HOT 3 CLOSED

As it stand a GPRN cannot use a constant weighting (or constant kernel) because the matrix used in the weight need to be inverted in the middle of the calculations, this is due to the Mean field variational inference used.
As far as not using weight at all, well you would end up with just a GP in that case and again you would ran into problems with the variational inference method we use. In our case, the variational inference used was specifically developed for the GPRN not a GP, e.g. unlike a GP we dont use nor have a way to calculate the log likelihood. Removing or not using a weight would break all the calculations used.

from gpyrn.

Ok. Thanks :)

from gpyrn.

In any case if you are looking for something more similar to how a gp works check this package https://github.com/jdavidrcamacho/artgpn
It might be what you are looking for

from gpyrn.