Fixed errors with python3 (apple#723)

* Convert tf signature values to idexable list object * Fixed python3 comparison of bytes and strings for recurrent layers Co-authored-by: sacha <[email protected]>
RobertRiachi · Jun 4, 2020 · 1afb9cb · 1afb9cb
1 parent 4134744
commit 1afb9cb
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 1 deletion.
diff --git a/coremltools/converters/tensorflow/_tf_converter.py b/coremltools/converters/tensorflow/_tf_converter.py
@@ -223,7 +223,7 @@ def _graph_def_from_saved_model_or_keras_model(filename):
               raise ValueError('Unable to load a model with no signatures provided.')
             if len(signatures) >= 2:
               raise ValueError('Unable to load a model with multiple signatures')
-            concrete_func = signatures.values()[0]
+            concrete_func = list(signatures.values())[0]
         frozen_func = _convert_to_constants.convert_variables_to_constants_v2(concrete_func)
         graph_def = frozen_func.graph.as_graph_def(add_shapes=True)
     except ImportError as e:

diff --git a/coremltools/models/neural_network/builder.py b/coremltools/models/neural_network/builder.py
@@ -24,6 +24,10 @@
 
 
 def _set_recurrent_activation(param, activation):
+
+    if isinstance(activation, bytes):
+        activation = activation.decode("utf8")
+
     activation = activation.upper() if isinstance(activation, str) else activation
     if activation == 'SIGMOID':
         param.sigmoid.MergeFromString(b'')