Ember

Basic utils and base classes for training and evaluating models in PyTorch. Includes:

• BaseTrainer
• BaseDataset
• EarlyStopping with a PyTorch Scheduler interface plus other methods.

The base classes are designed with the argparse_args functionalities of LeGM in mind. BaseTrainer is also using ExperimentManager from LeGM directly.