When using method=gridsearch the user is supposed to specify a loss which will be used to evaluate the performance. Currently loss defaults to a classification loss, independent of the likelihood used for the Laplace approximation itself. I suggest to not set any default and raise an informative error if gridsearch is selected without specifying a loss. Alternatively, we could choose the default based on likelihood.
When using
method=gridsearchthe user is supposed to specify alosswhich will be used to evaluate the performance. Currentlylossdefaults to a classification loss, independent of the likelihood used for the Laplace approximation itself. I suggest to not set any default and raise an informative error ifgridsearchis selected without specifying aloss. Alternatively, we could choose the default based onlikelihood.