AttributeError: 'google.protobuf.pyext._message.RepeatedCompositeCo' object has no attribute 'append'

Question

I am building a transfer learning model on the MobileNetv2 pretrained model on Google Collab. Till yesterday, everything was fine. But, today, on executing

#Create the base model(feature_extractor) from the pre-trained model MobileNet V2
_URL = "https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/2"
feature_extractor = hub.KerasLayer(_URL, input_shape=(_TARGET_SIZE, _TARGET_SIZE,3))

I get the error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-29-663d4cbb70df> in <module>()
      2 _TARGET_SIZE = 224
      3 _URL = "https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/2"
----> 4 feature_extractor = hub.KerasLayer(_URL, input_shape=(_TARGET_SIZE, _TARGET_SIZE,3))
      5 #print(feature_extractor._layers)


/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/resource_variable_ops.py in _variable_handle_from_shape_and_dtype(shape, dtype, shared_name, name, graph_mode, initial_value)
    165     handle_data = cpp_shape_inference_pb2.CppShapeInferenceResult.HandleData()
    166     handle_data.is_set = True
--> 167     handle_data.shape_and_type.append(
    168         cpp_shape_inference_pb2.CppShapeInferenceResult.HandleShapeAndType(
    169             shape=shape.as_proto(), dtype=dtype.as_datatype_enum))

AttributeError: 'google.protobuf.pyext._message.RepeatedCompositeCo' object has no attribute 'append'

Any idea why this happens and do I need to get into the /usr/local/lib/python3.6/dist-packages/tensorflow_core/python/ops/resource_variable_ops.py file and make some changes ? I think its related to some update issue. Any help on how to tackle it?

Answer 1

I had same error with tensorflow (version 2.2.0-dev20200128) and fixed it by upgrading protobuf (As explained in this issue):

pip install -U protobuf==3.8.0

Or if your using Notebook (like Google Colab notebook) try this:

!pip install -U protobuf==3.8.0

Answer 2

If it was running fine until yesterday and you did not change anything, then check for two things- Th TF version, if it has changed bc recently they set 'default' TF version to 1.15.rc from 1.14.

If that is same then save this file and close all the colab windows even your chrome or whatever browser you are using then open again and try running the file.

EDIT: As I said above it must be because of the TF version. So revert back to the one that you used when the model was working. As you mentioned in your comments below it was working on version ‘dev20191010’, so rolling back to it will fix your issue.