Add EMA callback

[jemrobinson]

@marianovitasari20 @erinuclkwon here's a small wrapper class to implement EMA using the callback that will be included in the next version of pytorch-lightning. You can set decay_rate plus every_n_epochs and every_n_steps which control how often the EMA averaging will be applied.

I've added this as one of our default callbacks for training, so it will be applied to every model by default. It should remove the need for you to call an EMA instance inside your training loop. Does that do everything you need?

Closes #75

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
ice_station_zebra/callbacks
__init__.py
ema_weight_averaging_callback.py					24-28, 34-40
weight_averaging.py					96-112, 133, 149-161, 191-197, 212-217, 232-233, 248-249, 264-265, 277, 289, 310-326, 349-376, 387-398, 407-416
Project Total

This report was generated by python-coverage-comment-action

[erinuclkwon]

Yes, it does. LGTM. Thanks!