Surrogate residuals from an arbitrary sample

[cbujard]

We should be able to get the residuals from the current surrogate evaluated on an aribtrary (test) sample (for which we know the true values). The following instruction evaluates a "metric" based on the provided sample file (which contains both the input parameters and their corresponding true values). Any of these metrics are in fact directly computed from the residuals, but there seem to be no way to actually obtain these residuals. In the special case of gaussian process surrogates, it would also be desirable to not only get the residual at a given input point, but also the surrogate uncertainty at this point.

https://snl-dakota.github.io/docs/6.18.0/users/usingdakota/reference/model-surrogate-global-import_challenge_points_file.html

While these metrics offer a basic mean to evaluate the accuracy of a surrogate, these are not generally sufficient and more work based on such testing residuals is often needed.