Keras: Why explicitly depend on tensorflow on Keras 2.0?

[adamcavendish]

It looks like we have to install tensorflow (CPU) 1.0.1 when upgrading to keras 2.0. However I installed the GPU version, is there any way to fix this?

[adamcavendish]

Found one work-around:

pip install --upgrade --no-deps keras installs keras without dependencies.

Hope it helps.

[leriomaggio]

@adamcavendish What do you mean?
AFAIK, If you have tensorflow-gpu installed, the dependency should already be satisfied...

[adamcavendish]

The problem is that it was not satisfied, and downloads tensorflow (aka. the cpu version) for me.

[fchollet]

If you already have TF installed, use the pip flag --no-deps.

This seems like more of a TF issue than a Keras issue. On the Keras side,
it is reasonable to list tensorflow as a dependency.

On 14 March 2017 at 11:12, Adam Basfop Cavendish [e-mail]
wrote:

The problem is that it was not satisfied, and downloads tensorflow (aka.
the cpu version) for me.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/fchollet/keras/issues/5766#issuecomment-286511700,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AArWbylzSfGG02wi6H51LsKzVDOVe3Ycks5rltijgaJpZM4MczL8
.

[VaibhavSaini1996]

Sir, does by installing keras without its dependencies work?
i mean that will it use the conda installed tensorflow(1.2.0) or will it require tensorflow(1.0.1). i am trying to set an environment in anaconda for deep learning.
i performed following step:

1. installed anaconda
2. installed all nvidia reuirements.
3. installed tensorlow from github repo
4. installed bazel to build tensorflow.
5. At this step when i tried to use tensorflow it worked fine without keras.
6. conda install keras with all dependecied( i.e. tensorflow(1.0.1)etc).
7. now when i import keras it just says using tensorflow as backend.
   What am i doing wrong? please help

[Joian12]

I only want keras but when i try to run keras its always tels me that using tensorflow backend and i didnt even installed tensorflow.