change class lable string encoding as 'utf-8'. (#128)

* change class lable string encoding as 'utf-8'.

* Update all converter string encoding as 'utf-8'

Author: wenbingl

onnxmltools/convert/coreml/operator_converters/GLMClassifier.py

@@ -72,7 +72,7 @@ def convert_glm_classifier(scope, operator, container):
         matrix_w[i, :] = w.value
 
     if glm.WhichOneof('ClassLabels') == 'stringClassLabels':
-        class_labels = list(s.encode('ascii') for s in glm.stringClassLabels.vector)
+        class_labels = list(s.encode('utf-8') for s in glm.stringClassLabels.vector)
         attrs['classlabels_strings'] = class_labels
         zipmap_attrs['classlabels_strings'] = class_labels
     elif glm.WhichOneof('ClassLabels') == 'int64ClassLabels':

onnxmltools/convert/coreml/operator_converters/OneHotEncoder.py

@@ -15,7 +15,7 @@ def convert_one_hot_encoder(scope, operator, container):
     if raw_model.HasField('int64Categories'):
         attrs['cats_int64s'] = list(int(i) for i in raw_model.int64Categories.vector)
     if raw_model.HasField('stringCategories'):
-        attrs['cats_strings'] = list(str(s).encode('ascii') for s in raw_model.stringCategories.vector)
+        attrs['cats_strings'] = list(str(s).encode('utf-8') for s in raw_model.stringCategories.vector)
 
     container.add_node(op_type, [operator.inputs[0].full_name], [operator.outputs[0].full_name],
                        op_domain='ai.onnx.ml', **attrs)

onnxmltools/convert/coreml/operator_converters/SVC.py

@@ -83,7 +83,7 @@ def convert_svm_classifier(scope, operator, container):
         attrs['classlabels_ints'] = class_labels
         zipmap_attrs['classlabels_int64s'] = class_labels
     elif svc_classes == 'stringClassLabels':
-        class_labels = list(str(s).encode('ascii') for s in params.stringClassLabels.vector)
+        class_labels = list(str(s).encode('utf-8') for s in params.stringClassLabels.vector)
         attrs['classlabels_strings'] = class_labels
         zipmap_attrs['classlabels_strings'] = class_labels
     else:

onnxmltools/convert/coreml/operator_converters/TensorToLabel.py

@@ -41,7 +41,7 @@ def convert_tensor_to_label(scope, operator, container):
     if model_type == 'neuralNetworkClassifier':
         model = operator.raw_operator.neuralNetworkClassifier
         if model.WhichOneof('ClassLabels') == 'stringClassLabels':
-            labels = list(s.encode('ascii') for s in model.stringClassLabels.vector)
+            labels = list(s.encode('utf-8') for s in model.stringClassLabels.vector)
             label_type = onnx_proto.TensorProto.STRING
         elif model.WhichOneof('ClassLabels') == 'int64ClassLabels':
             labels = list(int(i) for i in model.int64ClassLabels.vector)
@@ -51,7 +51,7 @@ def convert_tensor_to_label(scope, operator, container):
     elif model_type == 'pipelineClassifier':
         model = operator.raw_operator.pipelineClassifier
         if model.WhichOneof('ClassLabels') == 'stringClassLabels':
-            labels = list(s.encode('ascii') for s in model.pipelineClassifier.stringClassLabels.vector)
+            labels = list(s.encode('utf-8') for s in model.pipelineClassifier.stringClassLabels.vector)
             label_type = onnx_proto.TensorProto.STRING
         elif model.WhichOneof('ClassLabels') == 'int64ClassLabels':
             labels = list(int(i) for i in model.int64ClassLabels.vector)

onnxmltools/convert/coreml/operator_converters/TensorToProbabilityMap.py

@@ -25,15 +25,15 @@ def convert_tensor_to_probability_map(scope, operator, container):
     if model_type == 'neuralNetworkClassifier':
         model = operator.raw_operator.neuralNetworkClassifier
         if model.WhichOneof('ClassLabels') == 'stringClassLabels':
-            attrs['classlabels_strings'] = list(s.encode('ascii') for s in model.stringClassLabels.vector)
+            attrs['classlabels_strings'] = list(s.encode('utf-8') for s in model.stringClassLabels.vector)
         elif model.WhichOneof('ClassLabels') == 'int64ClassLabels':
             attrs['classlabels_int64s'] = list(int(i) for i in model.int64ClassLabels.vector)
         else:
             raise ValueError('Unknown label type found')
     elif model_type == 'pipelineClassifier':
         model = operator.raw_operator.pipelineClassifier
         if model.WhichOneof('ClassLabels') == 'stringClassLabels':
-            attrs['classlabels_strings'] = list(s.encode('ascii') for s in model.stringClassLabels.vector)
+            attrs['classlabels_strings'] = list(s.encode('utf-8') for s in model.stringClassLabels.vector)
         elif model.WhichOneof('ClassLabels') == 'int64ClassLabels':
             attrs['classlabels_int64s'] = list(int(i) for i in model.int64ClassLabels.vector)
         else:

onnxmltools/convert/coreml/operator_converters/TreeEnsemble.py

@@ -51,7 +51,7 @@ def convert_tree_ensemble_model(scope, operator, container):
             attrs['classlabels_int64s'] = class_labels
             zipmap_attrs['classlabels_int64s'] = class_labels
         else:
-            class_labels = list(s.encode('ascii') for s in raw_model.treeEnsembleClassifier.stringClassLabels.vector)
+            class_labels = list(s.encode('utf-8') for s in raw_model.treeEnsembleClassifier.stringClassLabels.vector)
             attrs['classlabels_strings'] = class_labels
             zipmap_attrs['classlabels_strings'] = class_labels
     elif raw_model.WhichOneof('Type') == 'treeEnsembleRegressor':

onnxmltools/convert/coreml/operator_converters/neural_network/BidirectionalLSTM.py

@@ -303,14 +303,14 @@ def convert_bidirectional_lstm(scope, operator, container):
     betas = []
     for activation in params.activationsForwardLSTM:
         activation_type, alpha, beta = extract_rnn_activation_info(activation)
-        activation_types.append(activation_type.encode('ascii'))
+        activation_types.append(activation_type.encode('utf-8'))
         if alpha is not None:
             alphas.append(alpha)
         if beta is not None:
             betas.append(beta)
     for activation in params.activationsBackwardLSTM:
         activation_type, alpha, beta = extract_rnn_activation_info(activation)
-        activation_types.append(activation_type.encode('ascii'))
+        activation_types.append(activation_type.encode('utf-8'))
         if alpha is not None:
             alphas.append(alpha)
         if beta is not None:

onnxmltools/convert/coreml/operator_converters/neural_network/GRU.py

@@ -143,7 +143,7 @@ def convert_gru(scope, operator, container):
     betas = []
     for activation in params.activations:
         activation_type, alpha, beta = extract_rnn_activation_info(activation)
-        activation_types.append(activation_type.encode('ascii'))
+        activation_types.append(activation_type.encode('utf-8'))
         if alpha is not None:
             alphas.append(alpha)
         if beta is not None:

onnxmltools/convert/coreml/operator_converters/neural_network/LSTM.py

@@ -236,7 +236,7 @@ def convert_unidirectional_lstm(scope, operator, container):
     betas = []
     for activation in params.activations:
         activation_type, alpha, beta = extract_rnn_activation_info(activation)
-        activation_types.append(activation_type.encode('ascii'))
+        activation_types.append(activation_type.encode('utf-8'))
         if alpha is not None:
             alphas.append(alpha)
         if beta is not None:

onnxmltools/convert/coreml/operator_converters/neural_network/SimpleRNN.py

@@ -179,7 +179,7 @@ def convert_simple_rnn(scope, operator, container):
 
     # Add RNN's information of activation function
     activation, alpha, beta = extract_rnn_activation_info(params.activation)
-    rnn_attrs['activations'] = [activation.encode('ascii')]
+    rnn_attrs['activations'] = [activation.encode('utf-8')]
     if alpha is not None:
         rnn_attrs['activation_alpha'] = [alpha]
     if beta is not None: