Deterministic variable in posterior predictive samples

When generating posterior predictive samples using pm.sample_posterior_predictive the result only shows the observed variable. How can I access deterministic variables after sampling?

Here is an example. After using pm.sample_posterior_predictive I would like to access mu, which is a pm.Deterministic variable but the result only includes y.

import pymc3 as pm
from numpy.random import default_rng
rng = default_rng(seed=0)
x1 = rng.standard_normal((1000, 1)) + 3
y = 10 + x1 * 2
with pm.Model() as model: # Define priors sigma = pm.HalfCauchy("sigma", beta=10, testval=1.0) intercept = pm.Normal("Intercept", 0, sigma=20) x_coeff = pm.Normal("x", 0, sigma=20) # I would like this variable in the posterior predictive samples mu = pm.Deterministic("mu", intercept + x_coeff * x1) # Define likelihood likelihood = pm.Normal("y", mu=mu, sigma=sigma, observed=y) # Sample trace = pm.sample(1000, return_inferencedata=True, cores=1)
ppc = pm.sample_posterior_predictive(trace, model=model)
print(ppc.keys()) # Only shows y

1 Answer

The deterministic variable can be accessed in the posterior predictive after explicitly naming it in var_names.

ppc = pm.sample_posterior_predictive(trace, model=model, var_names=['y', 'mu'])

This shows both y and mu.