Starting values are NaN when using Interpolated to build customized prior

When I’m using the

Interpolated

method in pymc.distributions to build a customized prior for a simple Bayesian logistic regression problem (I’m pretty much following the tutorial here).

I obtained the following error

SamplingError: Initial evaluation of model at starting point failed!
Starting values:
{‘intercept_interval__’: array(nan), ‘coeffs_interval__’: array(nan)}

Initial evaluation results:
{‘intercept’: nan, ‘coeffs’: nan, ‘Y_obs’: -inf}

Any idea what causes the model to starting at nan? I’m using the default NUTS sampler btw.

Same as Interpolated distribution in V4 · Issue #5959 · pymc-devs/pymc · GitHub