Trainer Class

The trainer class performs high-level orchestration of the training process. Specifically, it takes care of the following:

• The main training loop, keeping count of training steps and epochs.

• Checkpoint saving and loading existing checkpoints at the start of training.

• Calling validation.

• Logging to Weights & Biases (W&B).

The trainer class is only the orchestrator, it relies on the problem class to implement these operations. Each problem class is a subclass of ProblemBase.