How to improve sampling or re-parametrize model

Instead of comparing the number of failures, the observed data can be “compared” to the Bernoulli-distributed failure:

failure = pm.Bernoulli("failure", p=pfailure, dims="parts", observed=y)
1 Like