peal.data

Datasets, dataloaders and the DataConfig schema they are built from.

interfaces.PealDataset / DataConfig define what every dataset offers the rest of PEAL (normalization round trips, contrastive collages, task configs); datasets and custom_datasets implement the image datasets (CelebA, Waterbirds, Square, ImageNet pairs, MNIST variants, …), tabular_datasets the tabular ones and dataset_generators synthesise confounded datasets. dataset_factory.get_datasets returns the train / validation / test splits for a config and dataloaders adds the DataloaderMixer / WeightedDataloaderList used to mix counterfactual data into predictor training.

Submodules

custom_datasets	Dataset classes for the concrete benchmarks PEAL experiments run on.
dataloaders	Dataloader construction and mixing for PEAL training loops.
dataset_factory	Build the train / validation / test datasets described by a DataConfig.
dataset_generators	Generators for PEAL's synthetic and confounded benchmark datasets.
dataset_utils	Label-file parsers that turn csv/json annotations into tensors and split keys.
datasets	Concrete PealDataset implementations for tabular and image data.
interfaces	Base dataset class and the pydantic DataConfig shared by all PEAL data.
tabular_datasets	Tabular datasets for the symbolic (non-image) CFKD/DiDAE experiments.
transformations	Callable image transformations used by PEAL datasets and generators.