Last data update: 2014.03.03
R: Difference measures for multivariate Gaussian pdfs
Difference measures for multivariate Gaussian pdfs
Description
Various difference measures for Gaussian pdfs are
implemented: Euclidean distance of the means, Mahalanobis
distance, Kullback-Leibler divergence, J-Coefficient,
Minkowski L2-distance, Chi-square divergence and the Hellinger
coefficient which is a similarity measure.
Usage
normdiff(mu1,sigma1=NULL,mu2,sigma2=sigma1,inv=FALSE,s=0.5,
method=c("Mahalanobis","KL","J","Chisq",
"Hellinger","L2","Euclidean"))
Arguments
mu1
mean value of pdf 1, a vector
sigma1
covariance matrix of pdf 1
mu2
mean value of pdf 2, a vector
sigma2
covariance matrix of pdf 2
method
difference measure to be used, see below
inv
if TRUE, 1-Hellinger is reported, default: inv=FALSE
s
exponent for Hellinger coefficient, default: s=0.5
Details
Equations can be found in H.-H. Bock, Analysis of
Symbolic Data , Chapter Dissimilarity Measures for Probability
Distributions
Value
A scalar object of class normdiff
reporting the distance.
Author(s)
Henning Rust, henning.rust@met.fu-berlin.de
References
H.-H. Bock, Analysis of Symbolic Data , Chapter
Dissimilarity measures for Probabilistic Distributions
Examples
library(gaussDiff)
mu1 <- c(0,0,0)
sig1 <- diag(c(1,1,1))
mu2 <- c(1,1,1)
sig2 <- diag(c(0.5,0.5,0.5))
## Euclidean distance
normdiff(mu1=mu1,mu2=mu2,method="Euclidean")
## Mahalanobis distance
normdiff(mu1=mu1,sigma1=sig1,mu2=mu2,method="Mahalanobis")
## Kullback-Leibler divergence
normdiff(mu1=mu1,sigma1=sig1,mu2=mu2,sigma2=sig2,method="KL")
## J-Coefficient
normdiff(mu1=mu1,sigma1=sig1,mu2=mu2,sigma2=sig2,method="J")
## Chi-sqr divergence
normdiff(mu1=mu1,sigma1=sig1,mu2=mu2,sigma2=sig2,method="Chisq")
## Minkowsi L2 distance
normdiff(mu1=mu1,sigma1=sig1,mu2=mu2,sigma2=sig2,method="L2")
## Hellinger coefficient
normdiff(mu1=mu1,sigma1=sig1,mu2=mu2,sigma2=sig2,method="Hellinger")
Results