[WIP] Fix conversion of XGBRFClassifier when n_estimators>2

.gitignore

@@ -25,6 +25,9 @@ build/
 # test generated files
 .pytest_cache
 .cache
+debug*
+dump*
+model*.json
 tests/temp
 tests/utils/models/coreml_OneHotEncoder_BikeSharing_new.json
 tests/utils/models/coreml_OneHotEncoder_BikeSharing2.onnx

tests/xgboost/test_xgboost_issue.py

@@ -0,0 +1,65 @@
+# SPDX-License-Identifier: Apache-2.0
+import unittest
+import numpy as np
+from numpy.testing import assert_almost_equal
+from sklearn.datasets import load_iris
+from xgboost import XGBClassifier, XGBRegressor, XGBRFClassifier, XGBRFRegressor
+from onnx.defs import onnx_opset_version
+from onnxconverter_common.onnx_ex import DEFAULT_OPSET_NUMBER
+from onnxmltools.convert import convert_xgboost
+from onnxmltools.convert.common.data_types import FloatTensorType
+from onnxruntime import InferenceSession
+
+
+TARGET_OPSET = min(DEFAULT_OPSET_NUMBER, onnx_opset_version())
+
+
+class TestXGBoostIssue(unittest.TestCase):
+ def common_test(self, cls, n_estimators):
+ dataset = load_iris()
+ X, y = dataset.data, dataset.target
+ model = cls(
+ n_estimators=n_estimators,
+ learning_rate=1.0,
+ subsample=0.8,
+ colsample_bynode=0.8,
+ reg_lambda=1e-5,
+ )
+ model.fit(X, y)
+ data = np.random.rand(5, 4).astype(np.float32)
+ expected_labels = model.predict(data)
+ expected_probabilities = (
+ model.predict_proba(data) if hasattr(model, "predict_proba") else None
+ )
+
+ onnx_model = convert_xgboost(
+ model, initial_types=[("input", FloatTensorType(shape=[None, None]))]
+ )
+
+ session = InferenceSession(
+ onnx_model.SerializeToString(), providers=["CPUExecutionProvider"]
+ )
+
+ if expected_probabilities is None:
+ (onnx_predictions,) = session.run(None, {"input": data})
+ assert_almost_equal(expected_labels, onnx_predictions.ravel())
+ else:
+ onnx_predictions, onnx_probabilities = session.run(None, {"input": data})
+ assert_almost_equal(expected_probabilities, onnx_probabilities)
+ assert_almost_equal(expected_labels, onnx_predictions.ravel())
+
+ def test_issue_663_classifier(self):
+ self.common_test(XGBClassifier, 1)
+ self.common_test(XGBRFClassifier, 1)
+ self.common_test(XGBClassifier, 2)
+ self.common_test(XGBRFClassifier, 2)
+
+ def test_issue_663_regressor(self):
+ self.common_test(XGBRegressor, 1)
+ self.common_test(XGBRFRegressor, 1)
+ self.common_test(XGBRegressor, 2)
+ self.common_test(XGBRFRegressor, 2)
+
+
+if __name__ == "__main__":
+ unittest.main(verbosity=2)