Transform ..._interval__ to truncated normal ...?

My model has an RV skill that has a beta distribution as its prior:

skill = pm.Beta('skill', 2.0, 5.0)

The trace collects values for both skill and for its transformation skill_logodds__.

Unfortunately my model exhibits some divergences. In investigating the divergences, I examine the chain warnings, which collects only the logodds skill_logodds__. Fortunately I can transform values of logodds back to the domain of skill with scipy’s expit.

So far so good.

My model also has an RV winnable that has a truncated normal as its prior:

winnable = pm.TruncatedNormal('winnable', mu=winnable_mu, sigma=winnable_sigma, lower=0, upper=opportunities)

The trace collects values for both winnable and for its transformation winnable_interval__. But as with skill, chain warnings only collects winnable_interval__.

How can I transform a value from the interval (as collected) back to the domain of the original RV?

You can do: pm.transforms.interval(0., opportunities).backward(trace['winnable']).eval()