This is an empirical probability metric estimation problem, with especially cruel error properties. There are a few different versions of this problem corresponding to various different information: Mutual information between two variables, KL divergence between two distributions, information of one variable; discrete variables, continuous variable… In the mutual information case this is an independence test.

Say I would like to know the mutual information of the laws of processes generating two streams \(X,Y\) of observations, with weak assumptions on the laws of the generation process. Better, suppose further that each observation from each process is i.i.d. In the case that they have a continuous state space and joint densities \(p_{X,Y}\), marginal densities \(p_{X},p_{Y},\)

\[ \operatorname {I} (X;Y)=\int _{\mathcal {Y}}\int _{\mathcal {X}}{p_{X,Y}X,Y\log {\left({\frac {p_{X,Y}X,Y}{p_{X}(x)\,p_{Y}(y)}}\right)}}\;dx\,dy\]

Information is harder than the many metrics, because observations with low frequency have high influence on that value but are by definition rarely observed. It is easy to get a uselessly biased — or even inconsistent — estimator, especially in the nonparametric case.

## Histogram estimator

The obvious one for discrete data. For continuous data where the histogram bins must also be learned, this method is highly sensitive and can be inconsistent if you don’t do it right (Paninski 2003).

## Parametric

🏗

## Monte Carlo parametric

One case you might want to estimate this value which is one where there is no nonparametric estimation problem *per se* but the integral to solve it is inconvenient.
In which case, we might use a Monte Carlo method.

John Schulmann explicates a good trick for estimating KL divergence in the case that you can simulate from \(x_i\sim q\) and calculate \(p(x)\) and \(q(x_i),\) The following estimator is good despite looking unrelated:

\[\begin{aligned} KL[q, p] &= \int_x q(x) \log \frac{q(x)}{p(x)} \mathrm{d}x\\ &= E_{ x \sim q}\left[\log \frac{q(x)}{p(x)} \right]\\ &\approx \frac1N \sum_{i=1}^N \frac12(\log p(x)−\log q(x))^2 \end{aligned}\]

He also introduced a simple debiased one that does even better. The mechanics are interesting. (If you actually want a mutual information this notionally calculates it if we find the KL divergence between joint and product densities; But that is not totally trivial I shall concede.)

## Incoming

Inform: A C library for information analysis of complex systems

Inform is a cross-platform C library designed for performing information analysis of complex system.

- The inform_dist struct provides discrete, emperical probability distributions. These form the basis for all of the information-theoretic measures.
- A collection of information measures built upon the distribution struct provide the core algorithms for the library, and are provided through the shannon.h header.
- A host of measures of the information dynamics on time series are built upon the core information measures. Each measure is housed in its own header, e.g. active_info.h.

See also ITE toolbox (estimators).

Information Theoretical Estimators (ITE) in PythonIt

- is the redesigned, Python implementation of the Matlab/Octave ITE toolbox.
- can estimate numerous entropy, mutual information, divergence, association measures, cross quantities, and kernels on distributions.
- can be used to solve information theoretical optimization problems in a high-level way.
- comes with several demos.
- is free and open source: GNU GPLv3(>=).
Estimated quantities:

`entropy (H)`

: Shannon entropy, Rényi entropy, Tsallis entropy (Havrda and Charvát entropy), Sharma-Mittal entropy, Phi-entropy (f-entropy).`mutual information (I)`

: Shannon mutual information (total correlation, multi-information), Rényi mutual information, Tsallis mutual information, chi-square mutual information (squared-loss mutual information, mean square contingency), L2 mutual information, copula-based kernel dependency, kernel canonical correlation analysis, kernel generalized variance, multivariate version of Hoeffding’s Phi, Hilbert-Schmidt independence criterion, distance covariance, distance correlation, Lancaster three-variable interaction.`divergence (D)`

: Kullback-Leibler divergence (relative entropy, I directed divergence), Rényi divergence, Tsallis divergence, Sharma-Mittal divergence, Pearson chi-square divergence (chi-square distance), Hellinger distance, L2 divergence, f-divergence (Csiszár-Morimoto divergence, Ali-Silvey distance), maximum mean discrepancy (kernel distance, current distance), energy distance (N-distance; specifically the Cramer-Von Mises distance), Bhattacharyya distance, non-symmetric Bregman distance (Bregman divergence), symmetric Bregman distance, J-distance (symmetrised Kullback-Leibler divergence, J divergence), K divergence, L divergence, Jensen-Shannon divergence, Jensen-Rényi divergence, Jensen-Tsallis divergence.`association measures (A)`

: multivariate extensions of Spearman’s rho (Spearman’s rank correlation coefficient, grade correlation coefficient), multivariate conditional version of Spearman’s rho, lower and upper tail dependence via conditional Spearman’s rho.`cross quantities (C)`

: cross-entropy,`kernels on distributions (K)`

: expected kernel (summation kernel, mean map kernel, set kernel, multi-instance kernel, ensemble kernel; specific convolution kernel), probability product kernel, Bhattacharyya kernel (Bhattacharyya coefficient, Hellinger affinity), Jensen-Shannon kernel, Jensen-Tsallis kernel, exponentiated Jensen-Shannon kernel, exponentiated Jensen-Rényi kernels, exponentiated Jensen-Tsallis kernels.`conditional entropy (condH)`

: conditional Shannon entropy.`conditional mutual information (condI)`

: conditional Shannon mutual information.

## References

*Proceeding of the Second International Symposium on Information Theory*, edited by Petrovand F Caski, 199–213. Budapest: Akademiai Kiado.

*Neural Computation*16 (4): 717–36.

*Frontiers in Neuroscience*5: 90.

*Diversity and Distributions*17 (1): 43–57.

*Journal of Machine Learning Research*, 277–86.

*arXiv:cond-Mat/0202383*, February.

*Physics Letters A*128 (6–7): 369–73.

*Neural Computation*22 (1): 121–57.

*Journal of Machine Learning Research*10: 1469.

*arXiv:1411.4342 [Stat]*, November.

*Physical Review E*69: 066138.

*Automatica*26 (3): 545–55.

*arXiv:2005.03750 [Cond-Mat, Physics:nlin, Stat]*, May.

*NIPS 2014*.

*Physical Review E*69 (5): 056111.

*arXiv:physics/0108025*.

*Neural Computation*15 (6): 1191–1253.

*Physica D: Nonlinear Phenomena*125 (3-4): 285–94.

*XRDS: Crossroads, The ACM Magazine for Students*19 (1): 24–28.

*Neural Computation*27 (10): 2097–2106.

*Statistica Sinica*7: 375–94.

*Arxiv Preprint arXiv:0712.4382*.

*arXiv:comp-Gas/9403002*, March.

*arXiv:comp-Gas/9403001*, March.

*Neural Computation*26 (11): 2570–93.

## No comments yet. Why not leave one?