Prroy/converter feature union (#183)

Added feature union converter

Author: prabhat00155

onnxmltools/convert/sklearn/_parse.py

@@ -150,6 +150,29 @@ def _parse_sklearn_pipeline(scope, model, inputs):
     return inputs
 
 
+def _parse_sklearn_feature_union(scope, model, inputs):
+    '''
+    :param scope: Scope object
+    :param model: A scikit-learn FeatureUnion object
+    :param inputs: A list of Variable objects
+    :return: A list of output variables produced by feature union
+    '''
+    # Output variable name of each transform. It's a list of string.
+    transformed_result_names = []
+    # Encode each transform as our IR object
+    for name, transform in model.transformer_list:
+        transformed_result_names.append(_parse_sklearn_simple_model(scope, transform, inputs)[0])
+    # Create a Concat ONNX node
+    concat_operator = scope.declare_local_operator('SklearnConcat')
+    concat_operator.inputs = transformed_result_names
+
+    # Declare output name of scikit-learn FeatureUnion
+    union_name = scope.declare_local_variable('union', FloatTensorType())
+    concat_operator.outputs.append(union_name)
+
+    return concat_operator.outputs 
+
+
 def _parse_sklearn(scope, model, inputs):
     '''
     This is a delegate function. It doesn't nothing but invoke the correct parsing function according to the input
@@ -161,6 +184,8 @@ def _parse_sklearn(scope, model, inputs):
     '''
     if isinstance(model, pipeline.Pipeline):
         return _parse_sklearn_pipeline(scope, model, inputs)
+    elif isinstance(model, pipeline.FeatureUnion):
+        return _parse_sklearn_feature_union(scope, model, inputs)
     else:
         return _parse_sklearn_simple_model(scope, model, inputs)
 

onnxmltools/convert/sklearn/operator_converters/Concat.py

@@ -0,0 +1,15 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for
+# license information.
+# --------------------------------------------------------------------------
+
+from ...common._registration import register_converter
+
+
+def convert_sklearn_concat(scope, operator, container):
+    container.add_node('Concat', [s for s in operator.input_full_names],
+                       operator.outputs[0].full_name, name=scope.get_unique_operator_name('Concat'), axis=1)
+
+
+register_converter('SklearnConcat', convert_sklearn_concat)

onnxmltools/convert/sklearn/operator_converters/__init__.py

@@ -6,6 +6,7 @@
 
 # To register converter for scikit-learn operators, import associated modules here.
 from . import Binarizer
+from . import Concat 
 from . import DictVectorizer
 from . import DecisionTree
 from . import GradientBoosting

onnxmltools/convert/sklearn/shape_calculators/Concat.py

@@ -0,0 +1,15 @@
+# -------------------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License. See License.txt in the project root for
+# license information.
+# --------------------------------------------------------------------------
+
+from ...common._registration import register_shape_calculator
+from ...common.utils import check_input_and_output_numbers
+
+
+def calculate_sklearn_concat(operator):
+    check_input_and_output_numbers(operator, output_count_range=1)
+
+
+register_shape_calculator('SklearnConcat', calculate_sklearn_concat)

onnxmltools/convert/sklearn/shape_calculators/LabelEncoder.py

@@ -10,7 +10,7 @@
 from ...common.utils import check_input_and_output_numbers, check_input_and_output_types
 
 
-def calculate_sklearn_lebel_encoder_output_shapes(operator):
+def calculate_sklearn_label_encoder_output_shapes(operator):
     '''
     This function just copy the input shape to the output because label encoder only alters input features' values, not
     their shape.
@@ -22,4 +22,4 @@ def calculate_sklearn_lebel_encoder_output_shapes(operator):
     operator.outputs[0].type = Int64TensorType(copy.deepcopy(input_shape))
 
 
-register_shape_calculator('SklearnLabelEncoder', calculate_sklearn_lebel_encoder_output_shapes)
+register_shape_calculator('SklearnLabelEncoder', calculate_sklearn_label_encoder_output_shapes)

onnxmltools/convert/sklearn/shape_calculators/__init__.py

@@ -5,7 +5,7 @@
 # --------------------------------------------------------------------------
 
 # To register shape calculators for scikit-learn operators, import associated modules here.
-from . import TextVectorizer
+from . import Concat 
 from . import DictVectorizer
 from . import Imputer
 from . import LabelEncoder
@@ -15,3 +15,4 @@
 from . import Scaler
 from . import SVM
 from . import SVD
+from . import TextVectorizer