Common Sense reduced to Priors
When working with PyMC3 you often find yourself looking at a trace that you know isn't going to converge anyway. Wouldn't it be great if there was an automatic way to detect that and stop the sampling process?
As a maintainer of scikit-lego, I often see PR's that want to move attribute validation from the `fit` method into the `__init__`. In this post we'll explore why, in most cases, that won't work as you expect.