Open alicanb opened 6 years ago
Although slow, I think there can be a naive implementation for the multivariate polygamma functions correct? If the naive implementation is okay, I would be happy to take this up.
The naive implementation would be based on a summation of univariate polygamma functions.
I'm not sure it would be necessary. If you implement multivariate lgamma naively, autograd should take care of the rest? (Do take the multivariate lgamma up though...)
Ah, yes. Autograd should take up multivariate polygamma. In fact, we can use this as a testing mechanism. Shall I implement multivariate lgamma, if that's fine?
I think this will need multivariate polygamma
Task list: