Merge branch 'main' into ashay/cmake-export-filename

.azure-pipelines/Linux-CI.yml

@@ -81,7 +81,7 @@ jobs:
   - script: |
       source venv/bin/activate
 
-      pytest -sv --cov=onnx --cov-report=xml --cov-append --cov-branch --junit-xml pytest.xml
+      pytest -sv --cov=onnx --cov-report=xml --cov-append --cov-branch --junit-xml pytest.xml -n auto --dist loadscope
       if [ $? -ne 0 ]; then
         echo "pytest failed"
         exit 1

.azure-pipelines/MacOS-CI.yml

@@ -73,7 +73,7 @@ jobs:
     displayName: 'Install dependencies and ONNX'
 
   - script: |
-      pytest
+      pytest -n auto --dist loadscope
       if [ $? -ne 0 ]; then
         echo "pytest failed"
         exit 1
@@ -83,7 +83,7 @@ jobs:
       export ORT_MAX_IR_SUPPORTED_VERSION=8
       export ORT_MAX_ML_OPSET_SUPPORTED_VERSION=3
       export ORT_MAX_ONNX_OPSET_SUPPORTED_VERSION=19
-      pytest
+      pytest -n auto --dist loadscope
       if [ $? -ne 0 ]; then
         echo "pytest failed when testing onnx with onnxruntime"
         exit 1

.azure-pipelines/Windows-CI.yml

@@ -72,7 +72,7 @@ jobs:
       )
 
       pip install -e ".[reference]" -v
-      pytest
+      pytest -n auto --dist loadscope
       IF NOT %ERRORLEVEL% EQU 0 (
         @echo "pytest failed"
         EXIT 1
@@ -118,7 +118,7 @@ jobs:
         pip uninstall -y onnx
         pip install .
 
-        pytest
+        pytest -n auto --dist loadscope
         IF NOT %ERRORLEVEL% EQU 0 (
           @echo "pytest failed when testing onnx with libprotobuf=3.20"
           EXIT 1

cmake/summary.cmake

@@ -27,6 +27,8 @@ function (onnx_print_configuration_summary)
   message(STATUS "  ONNX_WERROR                       : ${ONNX_WERROR}")
   message(STATUS "  ONNX_BUILD_TESTS                  : ${ONNX_BUILD_TESTS}")
   message(STATUS "  ONNX_BUILD_BENCHMARKS             : ${ONNX_BUILD_BENCHMARKS}")
+  message(STATUS "  ONNX_BUILD_SHARED_LIBS            : ${ONNX_BUILD_SHARED_LIBS}")
+  message(STATUS "  BUILD_SHARED_LIBS                 : ${BUILD_SHARED_LIBS}")
   message(STATUS "")
   message(STATUS "  Protobuf compiler                 : ${PROTOBUF_PROTOC_EXECUTABLE}")
   message(STATUS "  Protobuf includes                 : ${PROTOBUF_INCLUDE_DIRS}")

docs/Changelog-ml.md

@@ -1085,9 +1085,9 @@ This version of the operator has been available since version 3 of the 'ai.onnx.
 <dt><tt>aggregate_function</tt> : string (default is SUM)</dt>
 <dd>Defines how to aggregate leaf values within a target. <br>One of 'AVERAGE,' 'SUM,' 'MIN,' 'MAX.'</dd>
 <dt><tt>base_values</tt> : list of floats</dt>
-<dd>Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
+<dd>Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
 <dt><tt>base_values_as_tensor</tt> : tensor</dt>
-<dd>Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
+<dd>Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
 <dt><tt>n_targets</tt> : int</dt>
 <dd>The total number of targets.</dd>
 <dt><tt>nodes_falsenodeids</tt> : list of ints</dt>

docs/Changelog.md

@@ -21043,7 +21043,9 @@ This version of the operator has been available since version 17 of the default
         Let `d[i]` indicate the i-th dimension of `X`.
         If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`,
         the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`.
-        `Y` and `X` have the same shape.
+        `Y` and `X` have the same shape. This operator supports unidirectional broadcasting
+        (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`);
+        for more details please check [the doc](Broadcasting.md).
 
 #### Version
 

docs/Operators-ml.md

@@ -1038,9 +1038,9 @@ Other versions of this operator: <a href="Changelog-ml.md#ai.onnx.ml.TreeEnsembl
 <dt><tt>aggregate_function</tt> : string (default is SUM)</dt>
 <dd>Defines how to aggregate leaf values within a target. <br>One of 'AVERAGE,' 'SUM,' 'MIN,' 'MAX.'</dd>
 <dt><tt>base_values</tt> : list of floats</dt>
-<dd>Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
+<dd>Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
 <dt><tt>base_values_as_tensor</tt> : tensor</dt>
-<dd>Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
+<dd>Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)</dd>
 <dt><tt>n_targets</tt> : int</dt>
 <dd>The total number of targets.</dd>
 <dt><tt>nodes_falsenodeids</tt> : list of ints</dt>

docs/Operators.md

@@ -13172,7 +13172,9 @@ expect(
         Let `d[i]` indicate the i-th dimension of `X`.
         If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`,
         the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`.
-        `Y` and `X` have the same shape.
+        `Y` and `X` have the same shape. This operator supports unidirectional broadcasting
+        (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`);
+        for more details please check [the doc](Broadcasting.md).
 
 #### Version
 

onnx/defs/nn/defs.cc

@@ -2516,7 +2516,9 @@ static const char* LayerNormalization_ver17_doc = R"DOC(
       Let `d[i]` indicate the i-th dimension of `X`.
       If `X`'s shape is `[d[0], ..., d[axis-1], d[axis], ..., d[rank-1]]`,
       the shape of `Mean` and `InvStdDev` is `[d[0], ..., d[axis-1], 1, ..., 1]`.
-      `Y` and `X` have the same shape.
+      `Y` and `X` have the same shape. This operator supports unidirectional broadcasting
+      (tensors `Scale` and `B` should be unidirectional broadcastable to tensor `X`);
+      for more details please check [the doc](Broadcasting.md).
 )DOC";
 
 bool BuildContextDependentFunctionBodyLayerNormalization(

onnx/defs/traditionalml/defs.cc

@@ -998,12 +998,12 @@ ONNX_ML_OPERATOR_SET_SCHEMA(
             std::string("SUM"))
         .Attr(
             "base_values",
-            "Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)",
+            "Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)",
             AttributeProto::FLOATS,
             OPTIONAL_VALUE)
         .Attr(
             "base_values_as_tensor",
-            "Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)",
+            "Base values for regression, added to final prediction after applying aggregate_function; the size must be the same as the classes or can be left unassigned (assumed 0)",
             AttributeProto::TENSOR,
             OPTIONAL_VALUE)
         .TypeAndShapeInferenceFunction([](InferenceContext& ctx) {

onnx/reference/ops/aionnxml/op_tree_ensemble_regressor.py

@@ -97,8 +97,10 @@ def _run(  # type: ignore
                 )
         if aggregate_function == "AVERAGE":
             res /= n_trees
+
+        # Convention is to add base_values after aggregate function
         if base_values is not None:
-            res[:, :] = np.array(base_values).reshape((1, -1))
+            res[:, :] += np.array(base_values).reshape((1, -1))
 
         if post_transform in (None, "NONE"):
             return (res,)

onnx/reference/ops/op_tfidf_vectorizer.py

@@ -17,8 +17,8 @@ def __init__(self):
         self.added_keys = []
 
     def emplace(self, key, value):
-        if not isinstance(key, int):
-            raise TypeError(f"key must be a NGramPart not {type(key)}.")
+        if not isinstance(key, (int, str)):
+            raise TypeError(f"key must be a int or str not {type(key)}.")
         if not isinstance(value, NgramPart):
             raise TypeError(f"value must be a NGramPart not {type(value)}.")
         if key not in self:
@@ -147,11 +147,12 @@ def __init__(self, onnx_node, run_params):  # type: ignore
         self.output_size_ = max(self.ngram_indexes_) + 1
         self.weights_ = self.weights  # type: ignore
         self.pool_int64s_ = self.pool_int64s  # type: ignore
+        self.pool_strings_ = self.pool_strings  # type: ignore
 
         self.int64_map_ = NgramPart(-10)
         self.int64_map_.init()
 
-        total_items = len(self.pool_int64s_)
+        total_items = len(self.pool_int64s_ or self.pool_strings_)
         ngram_id = 1  # start with 1, 0 - means no n-gram
         # Load into dictionary only required gram sizes
         ngram_size = 1
@@ -170,7 +171,7 @@ def __init__(self, onnx_node, run_params):  # type: ignore
                     and ngram_size <= self.max_gram_length_
                 ):
                     ngram_id = populate_grams(
-                        self.pool_int64s_,
+                        self.pool_int64s_ or self.pool_strings_,
                         start_idx,
                         ngrams,
                         ngram_size,
@@ -359,7 +360,7 @@ def _run(  # type: ignore
             # TfidfVectorizer returns a zero tensor of shape
             # {b_dim, output_size} when b_dim is the number of received observations
             # and output_size the is the maximum value in ngram_indexes attribute plus 1.
-            return self.output_result(B, frequencies)  # type: ignore[arg-type]
+            return (self.output_result(B, frequencies),)  # type: ignore[arg-type]
 
         def fn(row_num):
             self.compute_impl(

onnx/test/reference_evaluator_ml_test.py

@@ -10,6 +10,7 @@
 
 import numpy as np  # type: ignore
 from numpy.testing import assert_allclose  # type: ignore
+from parameterized import parameterized
 
 from onnx import ONNX_ML, TensorProto, TypeProto, ValueInfoProto
 from onnx.checker import check_model
@@ -757,7 +758,7 @@ def test_linear_classifier_unary(self):
 
     @staticmethod
     def _get_test_tree_ensemble_regressor(
-        aggregate_function, rule="BRANCH_LEQ", unique_targets=False
+        aggregate_function, rule="BRANCH_LEQ", unique_targets=False, base_values=None
     ):
         X = make_tensor_value_info("X", TensorProto.FLOAT, [None, None])
         Y = make_tensor_value_info("Y", TensorProto.FLOAT, [None, None])
@@ -786,6 +787,7 @@ def _get_test_tree_ensemble_regressor(
             domain="ai.onnx.ml",
             n_targets=1,
             aggregate_function=aggregate_function,
+            base_values=base_values,
             nodes_falsenodeids=[4, 3, 0, 0, 0, 2, 0, 4, 0, 0],
             nodes_featureids=[0, 2, 0, 0, 0, 0, 0, 2, 0, 0],
             nodes_hitrates=[1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
@@ -828,23 +830,34 @@ def _get_test_tree_ensemble_regressor(
         check_model(onx)
         return onx
 
+    @parameterized.expand(
+        [
+            (f"{agg}_{base_values}", base_values, agg)
+            for base_values in (None, [1.0])
+            for agg in ("SUM", "AVERAGE", "MIN", "MAX")
+        ]
+    )
     @unittest.skipIf(not ONNX_ML, reason="onnx not compiled with ai.onnx.ml")
-    def test_tree_ensemble_regressor(self):
+    def test_tree_ensemble_regressor(self, name, base_values, agg):
+        self.assertTrue(ONNX_ML)
+        del name  # variable only used to print test name
         x = np.arange(9).reshape((-1, 3)).astype(np.float32) / 10 - 0.5
         expected_agg = {
             "SUM": np.array([[0.576923], [0.576923], [0.576923]], dtype=np.float32),
             "AVERAGE": np.array([[0.288462], [0.288462], [0.288462]], dtype=np.float32),
             "MIN": np.array([[0.076923], [0.076923], [0.076923]], dtype=np.float32),
             "MAX": np.array([[0.5], [0.5], [0.5]], dtype=np.float32),
         }
-        for agg in ("SUM", "AVERAGE", "MIN", "MAX"):
-            expected = expected_agg[agg]
-            with self.subTest(aggregate_function=agg):
-                onx = self._get_test_tree_ensemble_regressor(agg)
-                self._check_ort(onx, {"X": x}, equal=True)
-                sess = ReferenceEvaluator(onx)
-                got = sess.run(None, {"X": x})
-                assert_allclose(expected, got[0], atol=1e-6)
+
+        expected = expected_agg[agg]
+        if base_values is not None:
+            expected += base_values[0]
+        with self.subTest(aggregate_function=agg):
+            onx = self._get_test_tree_ensemble_regressor(agg, base_values=base_values)
+            self._check_ort(onx, {"X": x}, equal=True)
+            sess = ReferenceEvaluator(onx)
+            got = sess.run(None, {"X": x})
+            assert_allclose(expected, got[0], atol=1e-6)
 
     @unittest.skipIf(not ONNX_ML, reason="onnx not compiled with ai.onnx.ml")
     def test_tree_ensemble_regressor_rule(self):
@@ -1789,6 +1802,86 @@ def test_svm_regressor_linear_one_class(self):
                 got = sess.run(None, {"X": x})
                 assert_allclose(expected[0], got[0], atol=1e-6)
 
+    def test_onnxrt_tfidf_vectorizer_ints(self):
+        inputi = np.array([[1, 1, 3, 3, 3, 7], [8, 6, 7, 5, 6, 8]]).astype(np.int64)
+        output = np.array(
+            [[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0]]
+        ).astype(np.float32)
+
+        ngram_counts = np.array([0, 4]).astype(np.int64)
+        ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64)
+        pool_int64s = np.array([2, 3, 5, 4, 5, 6, 7, 8, 6, 7]).astype(  # unigrams
+            np.int64
+        )  # bigrams
+
+        model = make_model_gen_version(
+            make_graph(
+                [
+                    make_node(
+                        "TfIdfVectorizer",
+                        ["tokens"],
+                        ["out"],
+                        mode="TF",
+                        min_gram_length=2,
+                        max_gram_length=2,
+                        max_skip_count=0,
+                        ngram_counts=ngram_counts,
+                        ngram_indexes=ngram_indexes,
+                        pool_int64s=pool_int64s,
+                    )
+                ],
+                "tfidf",
+                [make_tensor_value_info("tokens", TensorProto.INT64, [None, None])],
+                [make_tensor_value_info("out", TensorProto.FLOAT, [None, None])],
+            ),
+            opset_imports=OPSETS,
+        )
+
+        oinf = ReferenceEvaluator(model)
+        res = oinf.run(None, {"tokens": inputi})
+        self.assertEqual(output.tolist(), res[0].tolist())
+
+    def test_onnxrt_tfidf_vectorizer_strings(self):
+        inputi = np.array(
+            [["i1", "i1", "i3", "i3", "i3", "i7"], ["i8", "i6", "i7", "i5", "i6", "i8"]]
+        )
+        output = np.array(
+            [[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 1.0]]
+        ).astype(np.float32)
+
+        ngram_counts = np.array([0, 4]).astype(np.int64)
+        ngram_indexes = np.array([0, 1, 2, 3, 4, 5, 6]).astype(np.int64)
+        pool_strings = np.array(
+            ["i2", "i3", "i5", "i4", "i5", "i6", "i7", "i8", "i6", "i7"]
+        )
+
+        model = make_model_gen_version(
+            make_graph(
+                [
+                    make_node(
+                        "TfIdfVectorizer",
+                        ["tokens"],
+                        ["out"],
+                        mode="TF",
+                        min_gram_length=2,
+                        max_gram_length=2,
+                        max_skip_count=0,
+                        ngram_counts=ngram_counts,
+                        ngram_indexes=ngram_indexes,
+                        pool_strings=pool_strings,
+                    )
+                ],
+                "tfidf",
+                [make_tensor_value_info("tokens", TensorProto.INT64, [None, None])],
+                [make_tensor_value_info("out", TensorProto.FLOAT, [None, None])],
+            ),
+            opset_imports=OPSETS,
+        )
+
+        oinf = ReferenceEvaluator(model)
+        res = oinf.run(None, {"tokens": inputi})
+        self.assertEqual(output.tolist(), res[0].tolist())
+
 
 if __name__ == "__main__":
     unittest.main(verbosity=2)

onnx/test/version_converter/automatic_upgrade_test.py

@@ -14,12 +14,14 @@
 # to the most recent version and runs checker and shape inference on the final upgraded model.
 ####################################################################################
 
-tested_ops = []
-
 
 class TestAutomaticUpgrade(automatic_conversion_test_base.TestAutomaticConversion):
+    @classmethod
+    def setUpClass(cls):
+        cls.tested_ops = []
+
     def _test_op_upgrade(self, op, *args, **kwargs):
-        tested_ops.append(op)
+        self.tested_ops.append(op)
         self._test_op_conversion(op, *args, **kwargs, is_upgrade=True)
 
     def test_Abs(self) -> None:
@@ -1739,9 +1741,10 @@ def test_RegexFullMatch(self) -> None:
         )
 
     def test_ops_tested(self) -> None:
+        # NOTE: This test is order dependent and needs to run last in this class
         all_schemas = onnx.defs.get_all_schemas()
-        all_op_names = [schema.name for schema in all_schemas if schema.domain == ""]
-        excluded_ops = [
+        all_op_names = {schema.name for schema in all_schemas if schema.domain == ""}
+        excluded_ops = {
             # Sequence-based and Optional-based ops disabled because
             # the version converter doesn't play nicely with sequences
             "ConcatFromSequence",
@@ -1757,12 +1760,11 @@ def test_ops_tested(self) -> None:
             "OptionalGetElement",
             "OptionalHasElement",
             "StringSplit",
-        ]
-        all_op_names = [op for op in all_op_names if op not in excluded_ops]
+        }
+        expected_tested_ops = all_op_names - excluded_ops
 
-        untested_ops = set(all_op_names) - set(tested_ops)
-        print(untested_ops)
-        assert len(untested_ops) == 0
+        untested_ops = expected_tested_ops - set(self.tested_ops)
+        self.assertEqual(untested_ops, set())
 
 
 if __name__ == "__main__":

requirements-dev.txt

@@ -6,6 +6,7 @@ parameterized
 protobuf
 pytest
 pytest-cov
+pytest-xdist
 setuptools
 twine
 wheel