Towards composable machine learning

[junjihashimoto]

It's just an issue to consider what to do in the future.
The underling idea is to design parts of machine learning flow as nix-functions.
I think the implementation of the idea is a little done.
However, it is difficult to use as a library or datahub, because there is no way to find out the function user want to use.
For now, nix does not have documentation-engine like haddock.

[junjihashimoto]

dhall has generator of documentation.
https://hackage.haskell.org/package/dhall-docs
https://store.dhall-lang.org/Prelude-v20.2.0/
https://discourse.dhall-lang.org/t/dhall-docs-cli-utility/267

[junjihashimoto]

https://discourse.nixos.org/t/nickel-the-nix-language-spin-off/9592/5
I'm more attracted to nickel than dhall, but nickel does not support exporting to nix yet.