Issue with Keras and Jupyter Notebook

I am importing Keras into a new notebook. I am having an issue with MXNET. It looks like Keras is configured by default with MXNET. I am actually looking to use Tensorflow, unfortunately I am not able to even change programmatically the backend as the import fails.

Could you please let us know the configuration? I installed Keras, Python, Tensorflow and MXNet packages to be run on the instance.

I found a video that shows the usage of the notebook, I am not having the same experience at this point.

All I have in my code is the following:
import keras

Please le me know if I am missing something.
Thank you!

My goal is to be able to do some development with Keras with TensorFlow backend with and without the support of a GPU.

Which version of DLS you are using (Cloud, Desktop/Linux, Desktop/Windows)?

Hi Rajendra,

Thank you for following up with me. I am using the cloud version.

Hi @mtoumi,

Thanks for reporting, We have identified this issue with Jupyter Notebook and CPU instance, which is on the Cloud version only, and we will release the fix shortly.
Meanwhile you could use GPU instance to proceed with Notebook.

Hi Architha,

Thank you for your help, I will try again today and report the result.


I tried again today and the issue with the initial crash is solved, but I am still not able to use tensorflow as a backend. The default backend is still mxnet. I looked at the .keras.json configuration file and it is set with tensorflow as backend interestingly.

Please confirm if you can load tensorflow.


We do not support tensorflow backend inside Jupyter notebook. Jupyter runs inside the same environment as DLS and both uses MxNet backend.

Environments (pytorch, tensorflow, keras etc) are run inside a separate container and Jupyter runs in another container (on Linux). Which is why the keras.json inside environments container is showing tensorflow backend while Jupyter is still loading MxNet.

Hi Rajendra,

Thank you very much for clarification. I am currently looking at a GPU environment with Jupyter+Tensorflow to do the bulk of the documentation and prototyping development combined with DCS.