How to verify that uncertainty (estimated from pymc3) is accurate?

[junpenglao]

@hd2005

My parameters are typically positive and within a range [A, B] (govern by physics). Do you think that “B = Uniform(‘B’, lower=A, upper=B)” is an appropriate prior?

Well, it’s prior, so if you have information more than the range you should try to code that in!

@hd2005

Since it is so cheap to perform curve_fit in scipy to estimate B-hat. Do you think it is appropriate to perform curve_fit to estimate B-hat and then choose the prior as “B = Normal(‘B’, mu=B-hat, sd = some #)” for Bayesian fitting?

This is the idea of empirical Bayes - I think it works for some use case but you might want to read a bit more to decide whether you should use it.