Hi all,
I have some pretty heavily skewed data in some A/B testing that I’m running. I think it would probably be best approximated by a skewed normal distribution (pm.SkewedNormal). What would it’s conjugate prior is, would it be a Skewed normal too?
You don’t usually have to worry about conjugacy when sampling in PyMC (we don’t take advantage of it either).
Usually the only concern is to pick priors that respect the constraints of the parameters (e.g, must be positive or not) and other prior knowledge or regularization requirements.