dataloader.py

source data pytorch

Manages the creation of PyTorch DataLoaders for training, validation, and testing.

Functions

prepare_dataloader(...)

Creates a single DataLoader for a specific split.

• Args: dataset_type, split, signers, signs, batch_size, train_transforms.
• Logic:
  • Selects LazyKArSLDataset or MmapKArSLDataset.
  • For train split, it further performs an 80/20 random split for training/validation if not explicitly handling separate files (though typically train and test are distinct).
  • Returns (train_dl, val_dl) or test_dl.

prepare_dataloaders(...)

Wrapper function to generate all three dataloaders (Train, Val, Test) in one go.

Related Code

• mmap_dataset.py
• data_preparation.py