peal.training.trainers.mixup¶
- peal.training.trainers.mixup(data, targets, alpha, n_classes)[source]¶
Mix each sample with a random partner (Zhang et al. mixup).
- Parameters:
data (
torch.Tensor) – Input batch of shape(B, ...).targets (
torch.Tensor) – Integer class labels of shape(B,).alpha (
float) – Beta distribution parameter; onelam ~ Beta(alpha, alpha)is drawn for the whole batch.n_classes (
int) – Width of the one-hot targets.
- Returns:
(mixed_data, mixed_onehot_targets)with targets of shape(B, n_classes).- Return type: