I want to create two separate environments with TensorFlow in anaconda, one with CPU only support, which is compiled from sources, and one with GPU support using official Tensorflow binaries.
I have set up two environments with conda, using
conda create -n gpu --clone root. I activate my new environment using
source activate gpu. However, whenever I install a TensorFlow version using pip, the old environment is affected as well.
What am I doing wrong?
Conda uses hard links extensively. It's likely that pip is overwriting files that conda has already created, without removing the file first, and thereby affecting files in your other conda environments.
Try adding the
--copy flag to your
create --clone command.