log integrand in BayesModel

[johannesgiersdorf]

Using the logarithm of the integrand in the BayesModel provides a more stable way of computing functions regarding the integrand (e.g. logjoint)

[johannesgiersdorf]

Using the logarithm of the integrand only works with positive integrands. My suggestion would be to use a separate model Type for positive Integrand (which works directly with the logarithm of the integrand) for example BayesLogModel and one model that can take any integrand.

[theogf]

But by definition for a BayesModel the integrand needs to be positive!

[johannesgiersdorf]

Right, then we could add a Type Model or something similar for the general case.

[theogf]

Yep! That sounds good! Maybe something like GeneralModel, Model is a bit too "generic"