You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Go through all the distributions and make sure that they throw exceptions at boundaries. Also change all the doc so it's clear about positive vs. non-negative. For example, scale parameter sigma must be positive for normal distribution and prior count parameters alpha and beta must be positive for beta distribution.
Right now, I think some of the distributions try to be clever, like defining normal(mu,0) to be a discrete distribution with 100% of its mass at mu. Don't!