whitebob/transferentropy

Code for the paper "Estimating Transfer Entropy via Copula Entropy"

Estimating Transfer Entropy via Copula Entropy

This is the code for the preprint paper "Estimating Transfer Entropy via Copula Entropy" which available at here. A non-parametric method for estimating Transfer Entropy via estimating three Copula Entropy terms is proposed in this paper.

The proposed method is implemented in the R and Python package 'copent', available at

• R : here and CRAN,
• Python : here and PyPI.

The method is demonstrated in the experiment with the UCI Beijing PM2.5 data. The following conditional independence measures are compared in the experiment:

• Transfer Entropy via Copula Entropy (TE) [1];
• Conditional Distance Correlation (CDC) [2];
• Kernel-based Conditional Independence (KCI) [3];
• COnditional DEpendence Coefficient (CODEC) [4];
• Generalised Covariance Measure (GCM) [5];
• Kernel Partial Correlation (KPC) [6];
• Partial Correlation (pcor);
• Randomized conditional Correlation Test (RCoT) [7].

References

1. Ma, J. Estimating Transfer Entropy via Copula Entropy. arXiv preprint arXiv:1910.04375, 2019.
2. Wang, X.; Pan, W.; Hu, W.; Tian, Y. & Zhang, H. Conditional distance correlation. Journal of the American Statistical Association, 2015, 110, 1726-1734.
3. Zhang, K.; Peters, J.; Janzing, D. & Schölkopf, B. Kernel-based conditional independence test and application in causal discovery. Uncertainty in Artificial Intelligence, 2011, 804-813.
4. Azadkia, M. & Chatterjee, S. A simple measure of conditional dependence. arXiv preprint arXiv:1910.12327, 2019.
5. Shah, R. D. & Peters, J. The hardness of conditional independence testing and the generalised covariance measure. Annals of Statistics, 2020, 48, 1514-1538.
6. Huang, Z.; Deb, N. & Sen, B. Kernel Partial Correlation Coefficient -- a Measure of Conditional Dependence. arXiv preprint arXiv:2012.14804, 2020.
7. Strobl, E. V.; Zhang, K. & Visweswaran, S. Approximate Kernel-based Conditional Independence Tests for Fast Non-Parametric Causal Discovery. arXiv preprint arXiv:1702.03877, 2017.