Scalable hyperparameter selection with Ray Tune

[alanakbik]

Ray Tune is a framework for hyperparameter selection that supports running processes in parallel across multiple CPUs/GPUs and has connectors to a number of hyperparameter selection frameworks such as Hyperopt and Bayesopt.

As discussed in #563 with @myaldiz and @richardliaw it would be interesting to add support for this framework and potentially replace the current hyperparameter selection classes based on Hyperopt with this more general framework. For this, we need to slightly refactor the model trainer and the checkpointing mechanism.

[richardliaw]

I can help review if you want! Just let me know when you open the PR.

[alanakbik]

@richardliaw thanks a lot - I'll let you know when it's ready!

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.