Are custom likelihoods implemented in PyMC4? I didn’t see a DensityDist method. I was able to define and compute the likelihood as a tensor function but it is unattached and doesn’t affect sampling.
Not yet. Currently, the way to do this is to write down your log prob function, call it within a
pm.model and wrap the resulting tensor with a potential like this:
@pm.model() def model(): # something yield pm.Potential(your_log_prob_function(*args, **kwargs))
Thanks alot. That worked.