Added informative errors when passing non-text data without a custom encoder

Author: Pringled

semhash/utils.py

@@ -134,7 +134,15 @@ def featurize(
             col_texts = [r[col] for r in records]
         except KeyError as e:
             raise ValueError(f"Missing column '{col}' in one or more records") from e
-        col_emb = model.encode(col_texts)
+        try:
+            col_emb = model.encode(col_texts)
+        except TypeError as e:
+            sample_type = type(col_texts[0]).__name__ if col_texts else "unknown"
+            raise TypeError(
+                f"Failed to encode column '{col}' (data type: {sample_type}). "
+                f"If encoding non-text data, provide a compatible encoder via the `model` parameter. "
+                f"See our documentation for more info."
+            ) from e
         embeddings_per_col.append(np.asarray(col_emb))
 
     return np.concatenate(embeddings_per_col, axis=1)

tests/test_utils.py

@@ -94,6 +94,16 @@ def test_featurize(model: Encoder) -> None:
     with pytest.raises(ValueError, match="Missing column 'missing'"):
         featurize(records, ["missing"], model)
 
+    # Non-text data with text encoder raises helpful TypeError
+    class FakeImage:
+        pass
+
+    records_with_images = [{"img": FakeImage()}, {"img": FakeImage()}]
+    with pytest.raises(TypeError, match="Failed to encode column 'img'"):
+        featurize(records_with_images, ["img"], model)
+    with pytest.raises(TypeError, match="data type: FakeImage"):
+        featurize(records_with_images, ["img"], model)
+
 
 def test_remove_exact_duplicates() -> None:
     """Test exact duplicate removal, with and without reference records."""