Disabling missing data imputation

[ricardoV94]

For completeness, I think this is what @ckrapu was suggesting above:

with pm.Model() as m:
    mu = pm.Normal("mu")
    sigma = pm.HalfNormal("sigma")
    pm.Potential(
        'likelihood',
        pm.logp(pm.Normal.dist(mu, sigma), obs_dataset)[mask]
    )

m.point_logps()  # {'mu': -0.92, 'sigma': -0.73, 'likelihood': -61.98}