Using tensorflow 2.X fails with "module 'tensorflow' has no attribute 'Session'"

[xivarri]

Thanks for making this! Unfortunately I can't seem to get the install down - I have to use tensorflow2.X because python3.8 doesn't support 1.X, and for some reason I cannot get a 3.6 or 3.7 virtualenv to work - and while the install seems to work, when trying to instantiate a StressRNN, I get:

<ipython-input-4-f44bb10095c3> in <module>
----> 1 stress_rnn = StressRNN()

~/gits/StressRNN/stressrnn/stressrnn.py in __init__(self, f_name_add_exception_dict)
     64         self.graph = tf.Graph()
     65         with self.graph.as_default():
---> 66             self.session = tf.Session(graph=self.graph)
     67             with self.session.as_default():
     68                 self.model = tf.keras.models.model_from_json(model_json)

AttributeError: module 'tensorflow' has no attribute 'Session'

The internet seems to suggest running

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()

But doing that in an interpreter at least before importing StressRNN doesn't seem to work. Is there any workaround? Thanks

[xivarri]

An update - I got it to work by replacing the line

import tensorflow as tf

by

import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()

in both stressrnn.py and model.py before installing. So if there's no better workaround, would be good to add that to the install instructions.

[dbklim]

When using TensorFlow v2.X, the speed of the project is greatly reduced, several times.

But now I have made a small update and added support for TensorFlow v2.X "out of the box". Although, at the moment it is better to use the project with ONNX Runtime, it works faster.