Colabtools: AttributeError: 'Model' object has no attribute '_get_distribution_strategy'

[milansoliya4210]

I am using Tensorflow 2.x bundle package in Google Colab But it gives me error when I am trying to train model:
AttributeError: 'Model' object has no attribute '_get_distribution_strategy'``

Please Give any suggestion what is the issue and how can solve it??

[georgesymeonidis]

Hello,

I had the same issue and what I did is simple change my imports:

From
from keras.models import Model, load_model,Sequential
from keras.layers import Input, Dense
from keras.callbacks import ModelCheckpoint, TensorBoard
from keras import regularizers

To
from tensorflow.keras.models import Model, load_model,Sequential
from tensorflow.keras.layers import Input, Dense
from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard
from tensorflow.keras import regularizers

and it worked.

[johncleveland]

Seems the error is in keras/callbacks.py with self.model._get_distribution_strategy()

In what file do I make the above changes?

[Unimax]

same here. @georgesymeonidis does all imports from keras can be moved to tensorflow.keras?

i also think its from keras.callbacks import TensorBoard but changing to from tensorflow.keras.callbacks import TensorBoard not working

[lrscse]

hi @georgesymeonidis i made changes but still not working

[amandinecaut]

If you use Tensorflow 2.1.0, try to uninstall keras, then change the import from "keras.(...)" to "tensorflow.keras.(...)", it works for me!
Another solution is to downgrade tenforflow and keras: https://stackoverflow.com/questions/59090404/keras-that-does-not-support-tensorflow-2-0-we-recommend-using-tf-keras-or-al/59090507#59090507

[ElliotWood]

same here. @georgesymeonidis does all imports from keras can be moved to tensorflow.keras?

i also think its from keras.callbacks import TensorBoard but changing to from tensorflow.keras.callbacks import TensorBoard not working

This is still an issue

[subhadeepkaran]

had to create an virtual env with py 3.5.2, keras 2.1 and tensorflow 1.6 to get around this issue