Hyperparameter Optimization without Auto class.

[philip-ndikum]

Given the diversity of hardware and software it may be prudent to have an option for users to select the Pytorch Lightning an Ray settings, so instead of working with wrapper classes, the end-user can define the low level hyperparameter optimization & define GPU & CPU classes to work with in Pytorch trainer instead of having this done using the auto class.

Can this example be added to the core code base?