infomeasure-rs: A High-Performance Rust Library for Information-Theoretic Measures and Estimators
Abstract
Information theory has become a standard tool across the natural and social sciences, yet practical use is limited by estimator choice, fragmented software, reproducibility, and computational cost. We present infomeasure-rs, an open-source Rust library that brings a broad set of information-theoretic measures and estimators into one self-contained, type-safe package. It covers entropy, mutual information, conditional mutual information, transfer entropy, conditional transfer entropy, cross-entropy, and the Kullback-Leibler and Jensen-Shannon divergences, with discrete (plug-in and bias-corrected), kernel, ordinal, and nearest-neighbour estimators, including the Rényi and Tsallis families, and exposes local and global values and significance testing through a consistent trait interface. It ships as a single portable binary with no runtime dependency and is continuously validated against a reference implementation. On a single central processing unit (CPU) core, and even without its GPU or parallel paths, it is the fastest of the packages we compared across languages, matching the closest packages only for plug-in discrete entropy and substantially faster for mutual information, the conditional and transfer measures, and the kernel, ordinal, and generalised-entropy estimators. The optional GPU and multi-threaded paths accelerate selected continuous estimators further. The library is open source and, to our knowledge, the most complete information-theory package for Rust.
Related
- Invariant Measures as Estimators: Second-Order Stochastic Expansions and Bias Reduction
- Bias-Corrected Estimators for Joint Entropy, Conditional Entropy, and Mutual Information in the Discrete Bivariate Case: Theory and an Application to Motor Insurance
- Design of Privacy-Preserving Estimators Under Prior Information Exposure
- Bayes Estimators with Performance Comparable to Empirical Bayes Estimators and Improved Local Robustness
- Measuring Learned Equivariance: An Information Theoretic Approach
BibTeX
@misc{buth2026infomeasurers,
title = {{infomeasure-rs: A High-Performance Rust Library for Information-Theoretic Measures and Estimators}},
author = {Carlson Moses Büth},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.27951},
eprint = {2610.27951},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.27951},
}
RIS
TY - UNPB
TI - infomeasure-rs: A High-Performance Rust Library for Information-Theoretic Measures and Estimators
AU - Carlson Moses Büth
PY - 2026
DA - 2026/10/20
PB - arXiv
DO - 10.48550/arxiv.2610.27951
UR - https://arxiv.org/abs/2610.27951
AB - Information theory has become a standard tool across the natural and social sciences, yet practical use is limited by estimator choice, fragmented software, reproducibility, and computational cost. We present infomeasure-rs, an open-source Rust library that brings a broad set of information-theoretic measures and estimators into one self-contained, type-safe package. It covers entropy, mutual information, conditional mutual information, transfer entropy, conditional transfer entropy, cross-entropy, and the Kullback-Leibler and Jensen-Shannon divergences, with discrete (plug-in and bias-corrected), kernel, ordinal, and nearest-neighbour estimators, including the Rényi and Tsallis families, and exposes local and global values and significance testing through a consistent trait interface. It ships as a single portable binary with no runtime dependency and is continuously validated against a reference implementation. On a single central processing unit (CPU) core, and even without its GPU or parallel paths, it is the fastest of the packages we compared across languages, matching the closest packages only for plug-in discrete entropy and substantially faster for mutual information, the conditional and transfer measures, and the kernel, ordinal, and generalised-entropy estimators. The optional GPU and multi-threaded paths accelerate selected continuous estimators further. The library is open source and, to our knowledge, the most complete information-theory package for Rust.
ER -