correct bug: exception should not be raised for num_null_samples > num_reference -> adjust to num_observed > num_reference

Author: thegialeo

bayesflow/diagnostics/metrics/mmd_hypothesis_test.py

@@ -76,17 +76,16 @@ def compute_mmd_hypothesis_test_from_summaries(
     Raises
     ------
     ValueError
-        - If the number of null samples exceeds the number of reference samples or if the shapes of observed and
-        reference summaries do not match.
+        - If number of reference summaries is less than number of observed summaries.
         - If the shapes of observed and reference summaries do not match on dimensions besides the first one.
     """
     num_observed: int = observed_summaries.shape[0]
     num_reference: int = reference_summaries.shape[0]
 
-    if num_null_samples > num_reference:
+    if num_observed > num_reference:
         raise ValueError(
-            f"Number of null samples ({num_null_samples}) cannot exceed"
-            f"the number of reference samples ({num_reference})."
+            f"Number of reference summaries ({num_reference}) must be greater than"
+            f"number of observed summaries ({num_observed})."
         )
 
     if observed_summaries.shape[1:] != reference_summaries.shape[1:]:

tests/test_diagnostics/test_diagnostics_metrics.py

@@ -147,12 +147,10 @@ def test_compute_hypothesis_test_from_summaries_mismatched_shapes():
         )
 
 
-def test_compute_hypothesis_test_from_summaries_num_null_samples_exceeds_reference_samples():
-    """Test that compute_hypothesis_test_from_summaries raises ValueError when num_null_samples exceeds the number of
-    reference samples.
-    """
-    observed_summaries = np.random.rand(10, 5)
-    reference_summaries = np.random.rand(5, 5)
+def test_compute_hypothesis_test_from_summaries_observed_larger_than_reference_summaries():
+    """Test that compute_hypothesis_test_from_summaries raises ValueError if observed is larger than reference."""
+    observed_summaries = np.random.rand(20, 5)
+    reference_summaries = np.random.rand(10, 5)
     num_null_samples = 10
 
     with pytest.raises(ValueError):