Fix SemanticSimilarity for sentence-transformers 5.4+

src/evidently/legacy/features/semantic_similarity_feature.py

@@ -26,8 +26,8 @@ def normalized_cosine_distance(left, right):
 
         model = SentenceTransformer(self.model)
 
-        first = model.encode(data[self.columns[0]].fillna(""))
-        second = model.encode(data[self.columns[1]].fillna(""))
+        first = model.encode(data[self.columns[0]].fillna("").tolist())
+        second = model.encode(data[self.columns[1]].fillna("").tolist())
 
         return pd.DataFrame(
             {

tests/features/test_semantic_similarity_feature.py

@@ -0,0 +1,47 @@
+import sys
+import types
+
+import numpy as np
+import pandas as pd
+
+from evidently.legacy.features.semantic_similarity_feature import SemanticSimilarityFeature
+from evidently.legacy.pipeline.column_mapping import ColumnMapping
+from evidently.legacy.utils.data_preprocessing import create_data_definition
+
+
+def test_semantic_similarity_encodes_lists(monkeypatch):
+    class FakeSentenceTransformer:
+        instances = []
+
+        def __init__(self, model):
+            self.model = model
+            self.calls = []
+            self.instances.append(self)
+
+        def encode(self, sentences):
+            if not isinstance(sentences, list):
+                raise ValueError(f"Unsupported input type: {type(sentences).__name__}")
+            self.calls.append(sentences)
+            return np.array([[float(index + 1), 1.0] for index, _ in enumerate(sentences)])
+
+    sentence_transformers = types.ModuleType("sentence_transformers")
+    sentence_transformers.SentenceTransformer = FakeSentenceTransformer
+    monkeypatch.setitem(sys.modules, "sentence_transformers", sentence_transformers)
+
+    feature_generator = SemanticSimilarityFeature(columns=["answer", "reference_answer"])
+    data = pd.DataFrame(
+        {
+            "answer": ["same", None],
+            "reference_answer": ["same", "different"],
+        },
+        index=[10, 11],
+    )
+
+    result = feature_generator.generate_feature(
+        data=data,
+        data_definition=create_data_definition(None, data, ColumnMapping()),
+    )
+
+    assert FakeSentenceTransformer.instances[0].calls == [["same", ""], ["same", "different"]]
+    expected = pd.DataFrame({"answer|reference_answer": [1.0, 1.0]}, index=[10, 11])
+    pd.testing.assert_frame_equal(result, expected)