FeatureHasher, raise an error when the delimiter length is > 1 (#1036)

* FeatureHasher, raise an error when the delimiter length is > 1 Signed-off-by: Xavier Dupre <[email protected]> * fix error message Signed-off-by: Xavier Dupre <[email protected]> --------- Signed-off-by: Xavier Dupre <[email protected]>
onnx · Oct 20, 2023 · 8e612bb · 8e612bb
1 parent 84046e4
commit 8e612bb
Showing 1 changed file with 12 additions and 7 deletions.
diff --git a/skl2onnx/operator_converters/feature_hasher.py b/skl2onnx/operator_converters/feature_hasher.py
@@ -54,13 +54,18 @@ def convert_sklearn_feature_hasher(
             out_indices = scope.get_unique_variable_name(f"out_indices{i}")
             out_text = scope.get_unique_variable_name(f"out_text{i}")
             out_shape = scope.get_unique_variable_name(f"out_shape{i}")
-            container.add_node(
-                "StringSplit",
-                [reshaped, delimiter, skip_empty],
-                [out_indices, out_text, out_shape],
-                op_domain="ai.onnx.contrib",
-                op_version=1,
-            )
+            if len(separator) <= 1:
+                container.add_node(
+                    "StringSplit",
+                    [reshaped, delimiter, skip_empty],
+                    [out_indices, out_text, out_shape],
+                    op_domain="ai.onnx.contrib",
+                    op_version=1,
+                )
+            else:
+                raise RuntimeError(
+                    f"Only one character separator are supported but delimiter is {separator!r}."
+                )
             shape = scope.get_unique_variable_name(f"shape{i}")
             container.add_node("Shape", [col_to_split.full_name], [shape])