Add smoke tests for divergences

src/jnotype/energy/_dfd.py

@@ -54,7 +54,7 @@ def discrete_fisher_divergence(log_q: LogProbFn, ys: DataSet) -> Float[Array, "
 def _besag_pseudolikelihood_onepoint(
     log_q: LogProbFn, y: DataPoint
 ) -> Float[Array, " "]:
-    """Calculates Julian Besag's pseudolikelihood on a single data point.
+    """Calculates Julian Besag's pseudo-log-likelihood on a single data point.
 
     Namely,
     $$
@@ -76,9 +76,9 @@ def besag_pseudolikelihood_sum(
     log_q: LogProbFn,
     ys: DataSet,
 ) -> Float[Array, " "]:
-    """Besag pseudolikelihood calculated over the whole data set.
+    """Besag pseudo-log-likelihood calculated over the whole data set.
 
-    Note that pseudolikelihood is additive.
+    Note that pseudo-log-likelihood is additive.
     """
     n_points = ys.shape[0]
     return n_points * besag_pseudolikelihood_mean(log_q, ys)
@@ -88,10 +88,10 @@ def besag_pseudolikelihood_mean(
     log_q: LogProbFn,
     ys: DataSet,
 ) -> Float[Array, " "]:
-    """Average Besag pseudolikelihood.
+    """Average Besag pseudo-log-likelihood.
 
     Note:
-        As the pseudolikelihood is additive, for generalised
+        As the pseudo-log-likelihood is additive, for generalised
         Bayesian inference one should multiply by the data set size.
 
     See Also:

tests/energy/test_dfd.py

@@ -0,0 +1,56 @@
+"""Tests for discrete Fisher divergence ans pseudolikelihood methods."""
+
+import jax.numpy as jnp
+
+import pytest
+import jnotype.energy._dfd as dfd
+
+
+def linear_model(params, y):
+    """Linear (independent) model.
+
+    Args:
+        params: vector of shape (G,)
+        y: data point of shape (G,)
+    """
+    return jnp.sum(params * y)
+
+
+def quadratic_model(params, y):
+    """Quadratic (Ising) model.
+
+    Args:
+        params: matrix of shape (G, G)
+        y: data point of shape (G,)
+    """
+    return jnp.einsum("ij,i,j->", params, y, y)
+
+
+SETTINGS = [
+    (jnp.zeros(3), linear_model),
+    (jnp.zeros(5), linear_model),
+    (jnp.zeros((3, 3)), quadratic_model),
+    (jnp.zeros((5, 5)), quadratic_model),
+]
+
+
+@pytest.mark.parametrize("setting", SETTINGS)
+@pytest.mark.parametrize(
+    "divergence",
+    [
+        dfd.discrete_fisher_divergence,
+        dfd.besag_pseudolikelihood_mean,
+        dfd.besag_pseudolikelihood_sum,
+    ],
+)
+def test_quasidivergence_smoke_test(setting, divergence, n_points: int = 2):
+    theta, model = setting
+    G = theta.shape[0]
+
+    def logq(y):
+        return model(theta, y)
+
+    ys = jnp.zeros((n_points, G), dtype=int)
+
+    value = divergence(logq, ys)
+    assert value.shape == ()