Using a random variable as observed

Questions
19

This thread might help "log_likelihood" not found in InferenceData - #4 by karthur

What I’ve done is (probably too) simple:

After calling

idata = pm.sample()

I manually append the log_likelihood calculated using a modified copy of the usual function as I described above: compute_log_likelihood_for_potential

rvs_names = ['potential1']  # etc

idata.add_groups(
    dict(
        log_likelihood=compute_log_likelihood_for_potential(
            idata=idata,
            model=model,
            var_names=rvs_names,
            extend_inferencedata=False,
        )
    )
)

That’s pretty much it.

For later use in az.compare, you’ll also want to make sure the log_likelihood variables have the same string name as the observed_data variables

2 Likes