Lack of convergence in a super simple toy model

Questions
2

You are not learning sigma, the likelihood should probably be:

Otherwise you’re saying that sigma should be 1 (the default when not specified in pm.Normal), and that’s why you have such a large posterior spread.

That’s also why your posterior is so irregular, the sigma and the intercept are contributing equally to the likelihood and are jointly undertermined (as if you had two intercepts)

1 Like