Before, priors were set by either empirical bayes, with a customizable scaling factor, or an uninformed prior. This was done for simplicity for the user, but with a wider user base, some users may want or require a specified prior. This implements that ability, but leaves the default as empirical bayes.
This also does a big refactor, so that the _ab_test_distributions class/_set_distributions is not handling the prior parameters and calculations. the prior is now its own class, that handles its own parameterization, and is an object that lives within ab_test_model
Before, priors were set by either empirical bayes, with a customizable scaling factor, or an uninformed prior. This was done for simplicity for the user, but with a wider user base, some users may want or require a specified prior. This implements that ability, but leaves the default as empirical bayes.
This also does a big refactor, so that the
_ab_test_distributionsclass/_set_distributionsis not handling the prior parameters and calculations. the prior is now its own class, that handles its own parameterization, and is an object that lives withinab_test_model