Submodels and metadata #301
Description
It would be great to have a convenient way of packaging the different components of a workflow into their own repositories with metadata. This could be combined with a community curated super repository of components and datasets. This would enable things like searching for the highest rated feature extractor for the MNIST dataset, for example. DataLad implements something like this for datasets.
The components should be self-contained with all the dependencies required to run the code. I recommend nix-shells for this purpose, but docker containers could also be used.