[Hyperopt] Experiment Management Integrations

[ShreyaR]

Add integrations for experiment management for hyperopt.

Notes & Requirements:

• Should work with a ray backend
• The current tensorboard tfevents file only stores the tune logs, and not the actual trainer logs.
  • E.g. if AUROC is the metric being optimized, then tfevents only contains max(AUROC_val) for each step, instead of containing the AUROC_val metric directly.
  • Additionally, all metrics and losses for all splits should be logged.
• Log additional parameters like learning rate, batch size, etc in the case of decay.
• Log total number of parameters of the model as metadata
• Should be able to do a parameter importance analysis.