We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

stat.ML

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Statistics > Machine Learning

Title: Conditional independence testing based on a nearest-neighbor estimator of conditional mutual information

Authors: Jakob Runge
Abstract: Conditional independence testing is a fundamental problem underlying causal discovery and a particularly challenging task in the presence of nonlinear and high-dimensional dependencies. Here a fully non-parametric test for continuous data based on conditional mutual information combined with a local permutation scheme is presented. Through a nearest neighbor approach, the test efficiently adapts also to non-smooth distributions due to strongly nonlinear dependencies. Numerical experiments demonstrate that the test reliably simulates the null distribution even for small sample sizes and with high-dimensional conditioning sets. The test is better calibrated than kernel-based tests utilizing an analytical approximation of the null distribution, especially for non-smooth densities, and reaches the same or higher power levels. Combining the local permutation scheme with the kernel tests leads to better calibration, but suffers in power. For smaller sample sizes and lower dimensions, the test is faster than random fourier feature-based kernel tests if the permutation scheme is (embarrassingly) parallelized, but the runtime increases more sharply with sample size and dimensionality. Thus, more theoretical research to analytically approximate the null distribution and speed up the estimation for larger sample sizes is desirable.
Comments: 17 pages, 12 figures, 1 table
Subjects: Machine Learning (stat.ML); Information Theory (cs.IT); Methodology (stat.ME)
Cite as: arXiv:1709.01447 [stat.ML]
  (or arXiv:1709.01447v1 [stat.ML] for this version)

Submission history

From: Jakob Runge [view email]
[v1] Tue, 5 Sep 2017 15:21:25 GMT (655kb,D)

Link back to: arXiv, form interface, contact.