Feature ray

[bradday4]

This PR address issue #26 .

Things to note.

• I've updated the tests.yml to install ray in a local environment for testing. Also created a new pytest.ini file for ignoring deprecation warnings that ray throws. This is an issue that the maintainers of joblib will deal with so I don't see any reason to clutter pytest responses with a 100+ deprecation warnings.

• Also made some minor modifications to the Runner docstring to reflect support for ray.
  I've performed the unit tests locally any everything checks out.

• Code coverage has one caveat being the try - except block in Runner._run which catches the import error for ray. If we really wanted to run this bit of code we would need to alter tests.yml to run a specific test without ray installed, then install ray and run the normal test suite. I figured this was too much work for 5 lines of coverage.

• other odds and ends.

Let me know if you need me to do anything else.

[koaning]

Is there a reason why we're not able to use ray.init() here?

[bradday4]

Yah we could do that. It would just require passing kwargs along to ray.init() through Runner.run then to Runner._run I thinks it's cleaner to let the user do ray.init() from their script / code. But, that brings ups a question though. Is the try except block even necessary if the user performs ray.init() in their code because they'll receive an import error well beforehand. The argument I then see for putting ray.init() here is to assist users with a customized error message that will give them a little bit more help.

[koaning]

My initial guess would be that the average user will have access to a laptop and not a compute cluster. Given that the main use-case will be merely a laptop it occurs to me as more plausible that the user doesn't want to be bothered with setting up a ray-cluster locally. Especially when using this package.

We could support both, something like ray and ray-cluster as backend strings but I'm wondering if that over-complicates thing. If we had to assume one I'd favor the ray.init() route. Mainly because I imagine that's how many folks will want to use it.

@bradday4 what are your thoughts?

[koaning]

Mainly have a question on weather or not ray needs to be running as a service beforehand or if we can start ray just for our own run.

[bradday4]

Mainly have a question on weather or not ray needs to be running as a service beforehand or if we can start ray just for our own run.

Ray does not need to be running as a service. I.e called explicitly from the CLI before hand. We can use ray.init() in the code itself to start the server locally. I guess it would make things easier from a CI perspective since its one less build step during testing.