| # DataLoader builders with train/val splits and augmentation. | |
| from typing import Tuple | |
| from torch.utils.data import DataLoader, random_split, Dataset | |
| def build_dataloaders( | |
| dataset: Dataset, | |
| batch_size: int = 32, | |
| val_split: float = 0.2, | |
| num_workers: int = 0, | |
| seed: int = 42, | |
| ) -> Tuple[DataLoader, DataLoader]: | |
| """Create reproducible training and validation data loaders.""" | |
| val_size = int(len(dataset) * val_split) | |
| train_size = len(dataset) - val_size | |
| train_set, val_set = random_split( | |
| dataset, | |
| [train_size, val_size], | |
| generator=None, | |
| ) | |
| train_loader = DataLoader( | |
| train_set, | |
| batch_size=batch_size, | |
| shuffle=True, | |
| num_workers=num_workers, | |
| ) | |
| val_loader = DataLoader( | |
| val_set, | |
| batch_size=batch_size, | |
| shuffle=False, | |
| num_workers=num_workers, | |
| ) | |
| return train_loader, val_loader | |