peal.data.datasets¶
Concrete PealDataset implementations for tabular and image data.
SymbolicDataset reads a data.csv of tabular features; ImageDataset
is the shared base of the image datasets and implements the pieces
explainers and adaptors call back into: contrastive collages of
factual/counterfactual pairs, serialization of generated counterfactual
datasets to disk and DINO-based generator metrics (FID, LPIPS, Mahalanobis
outlier score). Image2MixedDataset loads images with multi-attribute CSV
labels (CelebA style, optional masks/hints, groups, text descriptions) and
Image2ClassDataset loads ImageFolder-style class directories. Datasets
toggle what __getitem__ returns via enable_*/disable_* methods.
Classes
ImageFolder-style dataset: one sub-directory per class. |
|
Image dataset with multi-attribute CSV labels (CelebA-style). |
|
Base class of the image datasets with the explainer-facing callbacks. |
|
Tabular (symbolic) dataset read from |