Sample stats for blackjax nuts (#6264)

* tests for added sample statistics
* sample stats test with more draws
* redesigned test for older blackjax version
* record blackjax sample stats
* moved sample stats argument to partial call

Author: TimOliverMaier

pymc/sampling/jax.py

@@ -142,6 +142,40 @@ def _sample_stats_to_xarray(posterior):
     return data
 
 
+def _blackjax_stats_to_dict(sample_stats, potential_energy) -> Dict:
+    """Extract compatible stats from blackjax NUTS sampler
+    with PyMC/Arviz naming conventions.
+
+    Parameters
+    ----------
+    sample_stats: NUTSInfo
+        Blackjax NUTSInfo object containing sampler statistics
+    potential_energy: ArrayLike
+        Potential energy values of sampled positions.
+
+    Returns
+    -------
+    Dict[str, ArrayLike]
+        Dictionary of sampler statistics.
+    """
+    rename_key = {
+        "is_divergent": "diverging",
+        "energy": "energy",
+        "num_trajectory_expansions": "tree_depth",
+        "num_integration_steps": "n_steps",
+        "acceptance_rate": "acceptance_rate",  # naming here is
+        "acceptance_probability": "acceptance_rate",  # depending on blackjax version
+    }
+    converted_stats = {}
+    converted_stats["lp"] = potential_energy
+    for old_name, new_name in rename_key.items():
+        value = getattr(sample_stats, old_name, None)
+        if value is None:
+            continue
+        converted_stats[new_name] = value
+    return converted_stats
+
+
 def _get_log_likelihood(model: Model, samples, backend=None) -> Dict:
     """Compute log-likelihood for all observations"""
     elemwise_logp = model.logp(model.observed_RVs, sum=False)
@@ -360,9 +394,9 @@ def sample_blackjax_nuts(
             "Only supporting the following methods to draw chains:" ' "parallel" or "vectorized"'
         )
 
-    states, _ = map_fn(get_posterior_samples)(keys, init_params)
+    states, stats = map_fn(get_posterior_samples)(keys, init_params)
     raw_mcmc_samples = states.position
-
+    potential_energy = states.potential_energy
     tic3 = datetime.now()
     print("Sampling time = ", tic3 - tic2, file=sys.stdout)
 
@@ -372,7 +406,7 @@ def sample_blackjax_nuts(
         *jax.device_put(raw_mcmc_samples, jax.devices(postprocessing_backend)[0])
     )
     mcmc_samples = {v.name: r for v, r in zip(vars_to_sample, result)}
-
+    mcmc_stats = _blackjax_stats_to_dict(stats, potential_energy)
     tic4 = datetime.now()
     print("Transformation time = ", tic4 - tic3, file=sys.stdout)
 
@@ -406,6 +440,7 @@ def sample_blackjax_nuts(
         log_likelihood=log_likelihood,
         observed_data=find_observations(model),
         constant_data=find_constants(model),
+        sample_stats=mcmc_stats,
         coords=coords,
         dims=dims,
         attrs=make_attrs(attrs, library=blackjax),

pymc/tests/sampling/test_jax.py

@@ -365,3 +365,52 @@ def test_numpyro_nuts_kwargs_are_used(mocked: mock.MagicMock):
     assert nuts_sampler._adapt_step_size == adapt_step_size
     assert nuts_sampler._adapt_mass_matrix
     assert nuts_sampler._target_accept_prob == target_accept
+
+
+@pytest.mark.parametrize(
+    "sampler_name",
+    [
+        "sample_blackjax_nuts",
+        "sample_numpyro_nuts",
+    ],
+)
+def test_idata_contains_stats(sampler_name: str):
+    """Tests whether sampler statistics were written to sample_stats
+    group of InferenceData"""
+    if sampler_name == "sample_blackjax_nuts":
+        sampler = sample_blackjax_nuts
+    elif sampler_name == "sample_numpyro_nuts":
+        sampler = sample_numpyro_nuts
+
+    with pm.Model():
+        pm.Normal("a")
+        idata = sampler(tune=50, draws=50)
+
+    stats = idata.get("sample_stats")
+    assert stats is not None
+    n_chains = stats.dims["chain"]
+    n_draws = stats.dims["draw"]
+
+    # Stats vars expected for both samplers
+    expected_stat_vars = {
+        "acceptance_rate": (n_chains, n_draws),
+        "diverging": (n_chains, n_draws),
+        "energy": (n_chains, n_draws),
+        "tree_depth": (n_chains, n_draws),
+        "lp": (n_chains, n_draws),
+    }
+    # Stats only expected for blackjax nuts
+    if sampler_name == "sample_blackjax_nuts":
+        blackjax_special_vars = {}
+        stat_vars = expected_stat_vars | blackjax_special_vars
+    # Stats only expected for numpyro nuts
+    elif sampler_name == "sample_numpyro_nuts":
+        numpyro_special_vars = {
+            "step_size": (n_chains, n_draws),
+            "n_steps": (n_chains, n_draws),
+        }
+        stat_vars = expected_stat_vars | numpyro_special_vars
+    # test existence and dimensionality
+    for stat_var, stat_var_dims in stat_vars.items():
+        assert stat_var in stats.variables
+        assert stats.get(stat_var).values.shape == stat_var_dims