| Title: | k-Nearest Neighbor Mutual Information Estimator |
| Version: | 1.0 |
| Description: | This is a 'C++' mutual information (MI) library based on the k-nearest neighbor (KNN) algorithm. There are three functions provided for computing MI for continuous values, mixed continuous and discrete values, and conditional MI for continuous values. They are based on algorithms by A. Kraskov, et. al. (2004) <doi:10.1103/PhysRevE.69.066138>, BC Ross (2014)<doi:10.1371/journal.pone.0087357>, and A. Tsimpiris (2012) <doi:10.1016/j.eswa.2012.05.014>, respectively. |
| License: | GPL (≥ 3) |
| Depends: | R (≥ 4.1.0) |
| Suggests: | spelling, testthat (≥ 3.0.0) |
| Config/testthat/edition: | 3 |
| Encoding: | UTF-8 |
| LazyData: | true |
| RoxygenNote: | 7.3.1 |
| Language: | en-US |
| NeedsCompilation: | yes |
| Packaged: | 2024-03-29 17:51:49 UTC; bgregor |
| Author: | Brian Gregor 
[aut, cre],
Katia Bulekova
[aut],
Reina Chau [aut],
Stefano Monti
[aut],
Benoit Jacob [cph] (Author of included Eigen library),
Gael Guennebaud [cph] (Author of included Eigen library),
Jose Luis Blanco [cph] (Author of included nanoflann library),
Pranjal Kumar Rai [cph] (Author of included nanoflann library) |
| Maintainer: | Brian Gregor <bgregor@bu.edu> |
| Repository: | CRAN |
| Date/Publication: | 2024-04-02 12:32:06 UTC |
Conditional mutual information estimation
Description
Conditional mutual information estimation CMI(X;Y|Z) where X is a continuous vector. The input Y and conditional input Z can be vectors or matrices. If Y and Z are discrete then they must be numeric or integer valued.
Usage
cond_mutual_inf(X, Y, Z, k = 3L)
Arguments
X |
input vector. |
Y |
input vector or matrix. |
Z |
conditional input vector or matrix. |
k |
number of nearest neighbors. |
Details
Argument Y is a vector of the same size as vector X, or a matrix whose column dimension matches the size of X. Argument Z is also a vector of the same size as vector X, or a matrix whose column dimension matches the size of X. If Y and Z are both matrices they must additionally have the same number of rows. If Y and/or Z are discrete values they must have a numeric or integer type.
Value
Returns the estimated conditional mutual information. The return value is a vector of size 1 if both Y and Z are vectors. If either Y or Z are matrices the return value is a vector whose size is the number of rows in the matrix.
References
Alkiviadis Tsimpiris, Ioannis Vlachos, Dimitris Kugiumtzis, Nearest neighbor estimate of conditional mutual information in feature selection, Expert Systems with Applications, Volume 39, Issue 16, 2012, Pages 12697-12708 doi:10.1016/j.eswa.2012.05.014
Examples
data(mutual_info_df)
set.seed(654321)
cond_mutual_inf(mutual_info_df$Zc_XcYc,
mutual_info_df$Xc, t(mutual_info_df$Yc))
M <- cbind(mutual_info_df$Xc, mutual_info_df$Yc)
ZM <- cbind(mutual_info_df$Yc, mutual_info_df$Wc)
cond_mutual_inf(mutual_info_df$Zc_XcYcWc, t(M), t(ZM))
Mutual information estimation
Description
Estimate the mutual information MI(X;Y) of the target X and features Y
where X and Y are both continuous using k-nearest neighbor distances.
Usage
mutual_inf_cc(target, features, k = 3L)
Arguments
target |
input vector. |
features |
input vector or matrix. |
k |
Integer number of nearest neighbors. The default value is 3. |
Details
The features argument is a vector of the same size as the target vector, or a matrix whose column dimension matches the size of the target vector.
Value
Returns the estimated mutual information. The return value is a vector of size 1 if the features argument is a vector. If the features argument is a matrix then the return value is a vector whose size matches the number of rows in the matrix.
References
Alexander Kraskov, Harald Stögbauer, and Peter Grassberger. Phys. Rev. E 69, 066138 (2004). doi:10.1103/PhysRevE.69.066138
Examples
data(mutual_info_df)
set.seed(654321)
mutual_inf_cc(mutual_info_df$Yc, t(mutual_info_df$Zc_XcYc))
mutual_inf_cc(mutual_info_df$Xc, t(mutual_info_df$Zc_XcYc), k=5)
Mutual information estimation
Description
Estimate the mutual information MI(X;Y) of the target X and features Y
where X is continuous or discrete and Y is discrete using k-nearest neighbor distances.
Usage
mutual_inf_cd(target, features, k = 3L)
Arguments
target |
input vector. |
features |
input vector or matrix. |
k |
Integer number of nearest neighbors. The default value is 3. |
Details
The features argument is a vector of the same size as the target vector, or a matrix whose column dimension matches the size of the target vector. Discrete values for the features or targets must be numeric or integer types.
Value
Returns the estimated mutual information. The return value is a vector of size 1 if the features argument is a vector. If the features argument is a matrix then the return value is a vector whose size matches the number of rows in the matrix.
References
Ross BC (2014) Mutual Information between Discrete and Continuous Data Sets. PLoS ONE 9(2): e87357. doi:10.1371/journal.pone.0087357
Examples
data(mutual_info_df)
set.seed(654321)
mutual_inf_cd(mutual_info_df$Zc_XdYd, t(mutual_info_df$Xd))
M <- cbind(mutual_info_df$Xd, mutual_info_df$Yd)
mutual_inf_cd(mutual_info_df$Zc_XdYdWd, t(M))
Toy Dataset for knnmi package
Description
Toy Dataset for knnmi package
Usage
data(mutual_info_df)
Format
A data frame with 100 rows and 10 columns