Create a semantic test of mvnormal.dlogpdf

[axch]

on the grounds that the derivation is long enough that it should be tested. For instance, one could test that gradient ascent recovers the correct parameter values on an example where the answer is obvious; or even that one gradient from a specific point matches the analytic result (in a case where the analytic result is obvious, if there is one).

• This should also cover the derivatives of the covariance functions.