Closed seabbs closed 5 years ago
@seabbs Thanks, this is great.
@seabbs whoops! got my threads crossed, the previous comment (now deleted to avoid confusion) was meant for Noam's PR. I need a bigger screen so I can see where I'm typing!
@seabbs Thanks for the PR! Just wanted to get your take on preferred tags that you think would make it easiest to for users to find this image. In the spirit of streamlining our stack & minimizing confusion, I think it would make sense to have this as the primary GPU / ML image. Do you prefer as a Docker tag:
rocker/gpu
rocker/ml
rocker/tidy-gpu
@eddelbuettel thoughts? (Sam's version is basically an update to the recipe that currently sits in the gpu
dir here and is the only one built on hub, as rocker/ml:gpu
, but I think it's confusing to use tags for both versions and flavors).
Something that included gpu
would probably be best to make it clear what it enables.
Just a note: The Dockerfile as it currently stands does not install keras
or tensorflow
- unlike the current one in the gpu
repo.
I have no strong feelings on naming. If it combines the *verse and gpu then maybe both should in the name too?
It may make sense to also layer keras and tf either in the same container or an additional one.
@seabbs Thanks, I think this provides a really nice building block, particularly with the README you've included. I'll merge it here and explore a bit on some ideas for stacking on the ML and naming images in a way that seems sufficiently intuitive...
Hi,
Added my attempt at GPU support for a rocker/tidyverse container as per seabbs/tidyverse-gpu#1
Everything seems to be in order based on my current - not exhaustive - testing. Happy to make any changes suggested.
Thanks
Sam