cidl-uhh 0.3.0

Read-only Python access library for remotely hosted causal inference benchmark datasets.

CIDL - Causal Inference Data Library

CIDL is a read-only Python library for accessing remotely hosted benchmark datasets for causal inference research. It provides a structured and reproducible interface for loading simulation datasets, aligned ground-truth artifacts, and dataset metadata.

The current release focuses on the Atlantic Causal Inference Conference 2022 Data Challenge (ACIC22). CIDL is designed as a broader framework and may support additional dataset contexts in future versions.

Important: CIDL does not include the benchmark datasets themselves. The package provides access logic only. Dataset artifacts are hosted remotely and require separate read-only access credentials.

What CIDL provides

CIDL supports:

• index-based dataset selection
• loading simulation datasets
• loading aligned ground-truth artifacts
• sequential iteration over many dataset instances
• local download of selected datasets
• access to dataset metadata and context information

The package is intended for researchers who want reproducible access to benchmark data for developing and evaluating causal inference methods.

Installation

Install CIDL via PyPI using pip install cidl-uhh. Import the package with import cidl.

Access requirements

CIDL uses an S3-compatible object store as backend, to host the datasets. Access is credential-based. To request credentials for read-only access, follow: ACIC22 — Access Configuration (S3)

Documentation

For an overview of the ACIC22 datasets (structure, variables, DGPs, target estimands), see:
ACIC22 — Data Info

Hands-on examples

• Quickstart.py — minimal “works end-to-end” example

Scope

CIDL is a data access library. It is not an estimator library, an evaluation framework, a data-hosting platform, or a credential management system.

Naming convention

ACIC22 refers specifically to the ACIC 2022 Data Challenge dataset collection. CIDL refers to the broader library/framework.

License

This project is licensed under the MIT License.

Authors

Julian Denzel Master’s Thesis Project, University of Hamburg [e-mail]

Martin Spindler Head of the Data Science Chair, University of Hamburg Business School [e-mail]

First issued: April 2026