I'm working on building and testing a model found here to update it to Tensorflow 2, but after attempting to save the model with tf.keras.Model.save(teacher,"saved_model/tea_model"); and received the following error on Google Colab Pro
<ipython-input-9-b823fb312e0f> in main()
--> 513 tf.keras.Model.save(teacher, "saved_model/tea_model");
/usr/local/lib/python3.7/dist-packages/keras/utils/traceback_utils.py in error_handler(*args, **kwargs)
65 except Exception as e: # pylint: disable=broad-except
66 filtered_tb = _process_traceback_frames(e.__traceback__)
---> 67 raise e.with_traceback(filtered_tb) from None
68 finally:
69 del filtered_tb
/usr/local/lib/python3.7/dist-packages/tensorflow/python/training/tracking/autotrackable.py in _list_functions_for_serialization(self, unused_serialization_cache)
99 functions[attribute_name] = attribute_value
100 finally:
--> 101 logging.set_verbosity(logging_verbosity)
102
103 return functions
UnboundLocalError: local variable 'logging_verbosity' referenced before assignment
I do have tf.compat.v1.disable_eager_execution() as it was initially written in v1.14.0 and need to maintain the compatibility, but after attempting the saving methods for both versions of Tensorflow I get the same error.
I can save the weights on the model, but would prefer the SavedModel format.
My solution was to simply define verbosity in the flags I was using as such:
from tensorflow.python.platform import flags
...
flags.DEFINE_integer('verbosity', 0, 'Verbosity of the error messages')