peal.data.tabular_datasets¶

Tabular datasets for the symbolic (non-image) CFKD/DiDAE experiments.

Each class wraps SymbolicDataset around one data.csv with a Target column and numeric-coded features. When the csv is missing the constructor tries to build it: the synthetic Circle data is generated locally, Adult is downloaded via the Kaggle API, COMPAS from the FairML GitHub mirror and German Credit from OpenML. The constructors also set config.input_size from the csv header so the tabular MLP predictors and the tabular DDPM know their width.

Module attributes

`ADULT_FEATURES`	The published tabular Adult cell was produced by a one-off curation script (`tools/curate_data_adult.py`) whose one-hot column names are what `configs/tabular_experiments/adaptors/adult_cfkd_dice.yaml` selects, e.g. `occupation_Professional` and `gender_Male`.
`ADULT_RENAME`	the raw UCI table spells two of those differently from the Kaggle mirror
`ADULT_CATEGORIES`	it is the output of the OLD AdultDataset path, which category-coded every string column, so the category names the curation needs are gone.

Functions

`curate_adult_frame`	Turn a raw Adult table into the one-hot encoding the configs select.
`decode_adult_codes`	Map an integer-coded Adult table back to its category strings.
`setup_kaggle_api`	Helper to ensure Kaggle API is authenticated, prompting the user if necessary.

Classes

`AdultDataset`	UCI Adult income dataset (`wenruliu/adult-income-dataset` on Kaggle).
`CircleDataset`	Synthetic 2-D "circle" dataset with a planted confounder.
`CompassDataset`	ProPublica COMPAS recidivism dataset (FairML preprocessed csv).
`GermanDataset`	German Credit dataset (OpenML `credit-g`).