Software

Released, documented, runnable

Reproducibility is a commitment of the research program, not a courtesy at the end of it. Everything below is public, documented, and built to be used by other researchers, students, and practitioners rather than only by its author.

CDANs

pip install cdans

Constraint-based causal discovery with optimized conditioning sets, built for autocorrelation, non-stationarity, and time-varying causal structure in multivariate time series. The package implements the MLHC 2023 method with a refined API and an emphasis on resource efficiency, which makes it usable for bedside monitoring and edge deployment.

Ferdous, Hasan, and Gani. CDANs: Temporal Causal Discovery from Autocorrelated and Non-Stationary Time Series Data. MLHC 2023.

Repository Paper
Stable release on PyPI

TimeGraph

A synthetic benchmark suite for time series causal discovery. It generates datasets with controlled causal structure, calibrated autocorrelation, non-stationarity, several noise families, and seasonal patterns, so that methods can be compared honestly and the conditions under which each one fails become visible rather than incidental.

Ferdous, Hossain, and Gani. TimeGraph: Synthetic Benchmark Datasets for Robust Time-Series Causal Discovery. KDD 2025.

Paper (DOI)
Published at KDD 2025

Composable Causality

A benchmarking toolkit that connects time series causal discovery to treatment-effect estimation in one pipeline. Data generators, discovery baselines, intervention models, and evaluators are separate components that recombine freely, so a discovered graph can be scored not only on structural recovery but on the quality of the intervention estimates it supports.

Ferdous and Gani. Composable Causality: A Toolkit for Systematic Time-Series Causal Discovery and Treatment-Effect Benchmarking. Submitted to NeurIPS 2026.

Under review

Citing and contributing

If you use any of these tools, please cite the corresponding paper. BibTeX for each is on the publications page.

Contributions, bug reports, and feature requests are welcome. If you are a student or researcher who wants to extend one of these tools, whether that means additional baselines, support for a new data modality, or application to a new domain, open an issue on the relevant repository or write to me at h.ferdous@umbc.edu.