Skip to content

Latest commit

 

History

History
54 lines (41 loc) · 2.15 KB

File metadata and controls

54 lines (41 loc) · 2.15 KB

Data

This directory holds the dataset used by the experiments. No data files are redistributed in this repository. The German Credit (Statlog) dataset is third-party data and is fetched at runtime from OpenML by the data-preparation script.

How to generate the data

From the repository root:

python -m src.data_prep

This downloads the German Credit dataset via sklearn.datasets.fetch_openml("credit-g", version=1) and writes three CSV files into this directory:

File Rows Description
german_credit_chiappa.csv 1000 Full dataset mapped to Chiappa's variables
german_credit_train.csv 700 Stratified 70% train split (random_state=42)
german_credit_test.csv 300 Stratified 30% test split (random_state=42)

The experiment scripts and src/data_prep.load_data() regenerate these files automatically if they are missing, so you normally do not need to run the command above by hand.

Variable mapping (Chiappa, 2019)

The raw OpenML features are mapped to the causal DAG variables as follows:

Var Meaning Source feature Encoding
A Sex personal_status 1 = male, 0 = female
C Age age continuous
S1 Checking account status checking_status ordinal 0–3
S2 Savings account savings_status ordinal 0–4
S3 Housing housing ordinal 0–2
R1 Credit amount credit_amount continuous
R2 Duration (months) duration continuous
Y Credit risk class 1 = good, 0 = bad

Provenance and citation

  • Hofmann, H. (1994). Statlog (German Credit Data). UCI Machine Learning Repository. https://doi.org/10.24432/C5NC77. Distributed via OpenML as dataset credit-g (https://www.openml.org/d/31).
  • Chiappa, S. (2019). Path-Specific Counterfactual Fairness. Proceedings of the AAAI Conference on Artificial Intelligence, 33(01). Source of the causal DAG over the German Credit variables.

The dataset is provided by its original distributors under their own terms; please consult OpenML / the UCI repository before any redistribution.