I would like to have various different conda environments, coexisting with the main DL frameworks installation provided by lambda stack.
The goal is to experiment with framework versions different from the current release (e.g., TF master branch).
I was about to install conda but then I stumbled over this:
which says installing conda “ruined” the stack install. Is this true?
It would be great if someone could indicate whether Anaconda is officially supported and / or if it empirically works for you
It’s not officially supported yet, it eventually will be. That said, I’ve heard rumors of people getting both to work side-by-side but haven’t tried myself.
Thank you! If I get to try it out I’ll update the thread
any timeline to when it might be supported?
Unfortunately we aren’t planning on adding official support any time soon.
How about miniconda? I think that would give all the virtualenv stuff without installing versions of packages.
Lambda stack and miniconda exist harmoniously. Miniconda gives one the ability to create virtual envs and utilize environment.yml files, but it doesn’t install all the base packages. So one can create an env, if one only has instructions for conda, then install packages using pip or conda. If you break a conda env, it’s no problem. Just delete it. I hope this helps. You’ve certainly helped me. Cheers. And thank you again. Miniconda — conda documentation
Is there an update on this topic? I would like to use Anaconda but, barring that, Miniconda may suffice.
Yes, Anaconda is isolated for python/pip/tensorflow/pytorch/cudnn/cuda.
So with Anaconda or python venv you would need to properly install the cuda with the PyTorch (see www.pytorch.org getting started matrix). or Tensorflow. Also if you need cuDNN make sure you set the path for the version you install. Another common issue is tensorrt.
I have examples at Anaconda/Miniconda work the same way.