Mean and variance at the training points

[rcnlee]

Is there an easy way to get the mean and variance at the training points? Is it already available in gp after calling fit! (thus saving compute)? or do I need to use predict_y to compute it?

Thanks!

[maximerischard]

No, this isn't implemented, and I agree that it should be, as recomputing the covariance matrix is wasteful. If you've already implemented this, I would be happy to integrate it into the package.

[patrickocal]

I posted this question on Julia discourse:

As part of a broader Value Function Iteration exercise, I wish to feed the output of predictMVN or predict_f back into an instantiation of GP .
Eg. Suppose we have created an instance gp by feeding GP the usual ingredients: training data x1 , target data t1 , along with eg MeanPoly(B) and a suitable kernel kern .
After optimising or fitting, how would I extract the posterior mean and covariance functions from gp and, moreover, extract them as type Mean and Kernel respectively?

@maximerischard, from what you're saying, I should currently be able to recompute. Can you explain how one does that? I might then attempt to fix this issue.