Investigate issue 1102 (#1103)

* Investigate issue 1102

Signed-off-by: Xavier Dupre <[email protected]>

* move test

Signed-off-by: Xavier Dupre <[email protected]>

* another try

Signed-off-by: Xavier Dupre <[email protected]>

* update

Signed-off-by: Xavier Dupre <[email protected]>

---------

Signed-off-by: Xavier Dupre <[email protected]>

tests_onnxmltools/test_issues_onnxmltools_2024.py

@@ -0,0 +1,94 @@
+# SPDX-License-Identifier: Apache-2.0
+import unittest
+import packaging.version as pv
+import onnx
+from sklearn.utils._testing import ignore_warnings
+from sklearn.exceptions import ConvergenceWarning
+from onnxruntime import __version__ as ort_version
+
+
+class TestInvestigateOnnxmltools(unittest.TestCase):
+
+    @unittest.skipIf(
+        pv.Version(ort_version) < pv.Version("1.17.3"),
+        reason="opset 19 not implemented",
+    )
+    @ignore_warnings(category=(ConvergenceWarning, FutureWarning))
+    def test_issue_1102(self):
+        from typing import Any
+        from sklearn.datasets import make_regression
+        import lightgbm
+        import numpy
+        import onnxruntime
+        import skl2onnx
+        from onnx.reference import ReferenceEvaluator
+        from onnxmltools.convert.lightgbm.operator_converters.LightGbm import (
+            convert_lightgbm,
+        )
+        from skl2onnx import update_registered_converter, to_onnx
+        from skl2onnx.common import shape_calculator
+        from sklearn import base, multioutput, pipeline, preprocessing
+
+        def Normalizer() -> list[tuple[str, Any]]:
+            return [
+                ("cast64", skl2onnx.sklapi.CastTransformer(dtype=numpy.float64)),
+                ("scaler", preprocessing.StandardScaler()),
+                ("cast32", skl2onnx.sklapi.CastTransformer(dtype=numpy.float32)),
+            ]
+
+        def Embedder(**kwargs: dict[str, Any]) -> list[tuple[str, Any]]:
+            return [("basemodel", lightgbm.LGBMRegressor(**kwargs))]
+
+        def BoL2EmotionV2(
+            backbone_kwargs: dict[str, Any] | None = None,
+        ) -> base.BaseEstimator:
+            backbone = Embedder(**(backbone_kwargs or {}))
+            normalizer = Normalizer()
+            model = pipeline.Pipeline([*normalizer, *backbone])
+            return multioutput.MultiOutputRegressor(model)
+
+        model = BoL2EmotionV2()
+        X, y = make_regression(100, n_features=4, n_targets=2)
+        model.fit(X, y)
+
+        update_registered_converter(
+            lightgbm.LGBMRegressor,
+            "LightGbmLGBMRegressor",
+            shape_calculator.calculate_linear_regressor_output_shapes,
+            convert_lightgbm,
+            options={"split": None},
+        )
+
+        sample = X.astype(numpy.float32)
+
+        exported = to_onnx(
+            model,
+            X=sample,
+            name="BoL2emotion",
+            target_opset={"": 19, "ai.onnx.ml": 2},
+        )
+        expected = model.predict(sample)
+        onnx.shape_inference.infer_shapes(exported)
+
+        ref = ReferenceEvaluator(exported)
+        got = ref.run(None, dict(X=sample))[0]
+        numpy.testing.assert_allclose(expected, got, 1e-4)
+
+        sess = onnxruntime.InferenceSession(
+            exported.SerializeToString(), providers=["CPUExecutionProvider"]
+        )
+        got = sess.run(None, dict(X=sample))[0]
+        numpy.testing.assert_allclose(expected, got, 1e-4)
+
+        with open("dump_model.onnx", "wb") as f:
+            f.write(exported.SerializeToString())
+
+        sess = onnxruntime.InferenceSession(
+            "dump_model.onnx", providers=["CPUExecutionProvider"]
+        )
+        got = sess.run(None, dict(X=sample))[0]
+        numpy.testing.assert_allclose(expected, got, 1e-4)
+
+
+if __name__ == "__main__":
+    unittest.main(verbosity=2)