Causal Discovery Algorithms For Time Series Data

[Cho-Geonwoo]

Description:

I am planning to implement various causal discovery methods for time series data. The methods I am particularly interested in include CDAN, ACD, TiMINo, and NTS-NOTEARS. Each of these methods offers unique approaches and advantages for uncovering causal relationships in time series datasets.

Methods of Interest:

CDAN (Causal Discovery with Additive Noise Models) ACD (Auto Regressive Causal Discovery) TiMINo (Time Series Interventions with Models for Interventions) NTS-NOTEARS (Nonlinear Time Series with NOTEARS) Reference: For a detailed comparison and discussion of these methods, please refer to the paper available here.

Thank you in advance for your valuable input!

[kunwuz]

Thank you for your interest in contributing! For now, let's focus on CDAN, ACD, and TiMINo first, as these algorithms fit well into the existing categories in causal-learn (CDAN under constraint-based, ACD under Granger causality-based, and TiMINo under FCM-based). For NTS-NOTEARS, we can consider including it later alongside other continuous optimization methods we are currently working on. What do you think?

[Cho-Geonwoo]

Thank you for the guidance! I completely agree with your suggestion. I'll begin by implementing CDAN first and will ensure it aligns well with the existing categories in causal-learn. Looking forward to contributing to this project!