soft.crisp.offline package

Submodules

soft.crisp.offline.supervised module

Implements supervised training and can be used for auto-encoder training as well.

soft.crisp.offline.supervised.setup_from_config(config: ~YACS.yacs.Config) -> (<class 'torch.nn.modules.module.Module'>, <class 'torch.optim.optimizer.Optimizer'>)

Given a configuration, load the necessary loss functions and optimizers accordingly. :param config: The configuration to use.

Returns:

loss_function, optimizer

soft.crisp.offline.supervised.train(train_dataset: SupervisedDataset, val_dataset: SupervisedDataset, model: Module, config: Config | None = None) Tuple[Module, History]

Train the model given the training dataset and validation dataset, using a configuration (config).

Parameters:
  • train_dataset – Dataset for training.

  • val_dataset – Dataset for validation.

  • model – The PyTorch model to be trained.

  • config – The configuration to use.

Returns:

model, losses [dictionary where keys are ‘train’ and ‘val’ for training losses (list), validation losses (list), respectively]

Module contents