causalbench-asu 0.2.4

Spatio Temporal Causal Benchmarking Platform

CausalBench

The up-to-date documentation regarding usage and features of CausalBench can be found at https://docs.causalbench.org.

Registration at CausalBench website is required in order to utilize the CausalBench package.

Install CausalBench:

pip install causalbench-asu

Overview

CausalBench is a flexible, fair, and easy-to-use evaluation platform designed to advance research in causal learning. It facilitates scientific collaboration by providing a suite of tools for novel algorithms, datasets, and metrics. Our mission is to promote scientific objectivity, reproducibility, fairness, and awareness of bias in causal learning research. CausalBench serves as a comprehensive benchmarking resource, impacting a broad range of scientific and engineering disciplines.

Features

• Transparent and Fair Evaluation: Ensures unbiased and transparent benchmarking processes.
• Facilitation of Scientific Collaboration: Encourages the sharing and development of novel algorithms, datasets, and metrics.
• Scientific Objectivity: Promotes objective assessment and comparison of causal learning methods.
• Reproducibility: Supports the reproducibility of research results, a cornerstone of scientific integrity.
• Awareness of Bias: Highlights and addresses biases in causal learning research.

Services Provided

1. Benchmarking Data: A repository of diverse datasets specifically curated for evaluating causal learning algorithms.
2. Algorithm Evaluation: Tools and frameworks for testing and comparing the performance of causal learning algorithms.
3. Model Benchmarking: Standards and protocols for assessing the efficacy of different causal models.
4. Metric Evaluation: A collection of metrics tailored for comprehensive evaluation of causal learning techniques.

Impact

CausalBench meets the needs of various scientific and engineering disciplines by providing essential resources and standards for evaluating causal learning methods. This platform helps researchers to:

• Collaborate and share advancements in causal learning.
• Ensure their work meets high standards of scientific rigor and fairness.
• Access a centralized repository of resources for benchmarking and evaluation.

Getting Started

To start using CausalBench, follow these steps:

1. Installation: Instructions for installing CausalBench can be found here.
2. Documentation: Comprehensive documentation for CausalBench, including CausalBench terms and tutorials, is available here.

Contributing

CausalBench is an open-source project and welcomes contributions from the community. We plan to announce the contribution guideline soon.

License

CausalBench is licensed under the Apache License.

Contact

For questions, feedback, or further information, please contact us at [e-mail].

Acknowledgments

This work is supported by NSF grant 2311716, "CausalBench: A Cyberinfrastructure for Causal-Learning Benchmarking for Efficacy, Reproducibility, and Scientific Collaboration".

Support Benchmark Context

CausalBench is structured to support different machine learning tasks and dataset types. With user contribution, the supported context will be expanded, currently (as of 8/12/25), these models and tasks are provided.

Dataset	File	Description
Abalone	data, static graph
Adult	data, static graph
Sachs	data, static graph
California Housing	data	Regression dataset from sklearn with 20,640 samples predicting median house values in California districts
Diabetes	data	Regression dataset from sklearn with 442 samples predicting disease progression from physiological variables
NetSim	data, static graph	Brain FMRI scan - 28 simulations - Each has different DGPs, num of nodes (5, 50), num of observations (50 to 5000), 1400 datasets in total
Time series simulated	data, temporal graph
Telecom	data, temporal graph	Summary graph*
ESS**	data, temporal graph	Summary graph*
Air Quality-[0-399]	data, temporal graph	Summary graph
Traffic-[0-399]	data, temporal graph	Summary graph*
Medical-[0-399]	data, temporal graph	Summary graph*
PCIC**	data, temporal graph	Summary graph*
* Lag set to 0 to be compatible with temporal causal DAGs with lag.

** ESS and PCIC datasets are not implemented yet.

Model	Task
PC	Static
GES	Static
VAR-LiNGAM	Temporal
PCMCIplus	Temporal

Metric	Task
Accuracy	Static
F1	Static
Precision	Static
Recall	Static
SHD	Static
Accuracy	Temporal
F1	Temporal
Precision	Temporal
Recall	Temporal
SHD	Temporal