neurodocker
neurodocker copied to clipboard
minimize space by choosing optimal libraries, removing source files, etc.,.
as an example, the tensorflow docker image is tiny relative to any tensorflow image we may produce with miniconda or some such.
- dynamically linked binaries are going to reduce space for FSL/ANTs/AFNI etc.,. so we need to figure out recipes to build those from source across at least a few OSes
- the additions for the build process itself should not be in a container post build. so i think every build recipe should have a
-pre
,-build
, and-post
section to help with cleanup.
just some thoughts here as we try to optimize container sizes, and build steps.