nntoolbox.learner module

class nntoolbox.learner.DistillationLearner(train_data: torch.utils.data.dataloader.DataLoader, val_data: torch.utils.data.dataloader.DataLoader, model: torch.nn.modules.module.Module, teacher: torch.nn.modules.module.Module, criterion: torch.nn.modules.module.Module, optimizer: torch.optim.optimizer.Optimizer, temperature: float, teacher_weight: float, hard_label_weight: float, device=device(type='cpu'))

Bases: nntoolbox.learner.SupervisedLearner

Distilling Knowledge from a big teacher network to a smaller model (UNTESTED)

References:

Geoffrey Hinton, Oriol Vinyals, Jeff Dean. “Distilling the Knowledge in a Neural Network.” https://arxiv.org/abs/1503.02531

TTIC Distinguished Lecture Series - Geoffrey Hinton. https://www.youtube.com/watch?v=EK61htlw8hY

compute_loss(inputs: torch.Tensor, labels: torch.Tensor) → torch.Tensor

class nntoolbox.learner.Learner(train_data: torch.utils.data.dataloader.DataLoader, val_data: torch.utils.data.dataloader.DataLoader, model: torch.nn.modules.module.Module, criterion: torch.nn.modules.module.Module, optimizer: torch.optim.optimizer.Optimizer)

Bases: object

class nntoolbox.learner.SupervisedLearner(train_data: torch.utils.data.dataloader.DataLoader, val_data: torch.utils.data.dataloader.DataLoader, model: torch.nn.modules.module.Module, criterion: torch.nn.modules.module.Module, optimizer: torch.optim.optimizer.Optimizer, device=device(type='cpu'), mixup: bool = False, mixup_alpha: float = 0.4)

Bases: nntoolbox.learner.Learner

compute_loss(inputs: torch.Tensor, labels: torch.Tensor, train: bool) → torch.Tensor

compute_outputs(inputs: torch.Tensor, train: bool) → torch.Tensor

evaluate() → float

learn(n_epoch: int, callbacks: Optional[Iterable[nntoolbox.callbacks.callbacks.Callback]] = None, metrics: Optional[Dict[str, nntoolbox.metrics.metrics.Metric]] = None, final_metric: str = 'accuracy', load_path=None) → float

learn_one_iter(inputs: torch.Tensor, labels: torch.Tensor)