circlust: Model-based clustering for two-dimensional linear-circular...

In gilles-guillot/HMMVMS: Hidden Markov Models for the analysis of Vessel Monitoring data

Description

Model-based clustering with two clusters for two-dimensional data, where one of the dimensions is the hour of the day. Expectation maximization algorithm is implemented that takes into account that the hour of the day data are circular, i.e. 00:00 is the same as 24:00. The circularity is being handled by defining a truncated normal distribution.

Usage

circlust(x, diag = F)

Arguments

x	a matrix with the linear data in the first column and the circular data in the second one.
diag	a boolean indicating if the covariance matrix of the joint distribution is diagonal, i.e. the linear and the circular variables are independent.

Value

loglik Final log-likelihood estimate of the EM algorithm. parameters Parameters inferred by the algorithm: pro: Mixing proportion of each distribution. mean: Means of the two clusters. sigma: Covariance matrices of the two clusters. z: Responsibilities. classification: Classification of the datapoints to the clusters. iter: Number of iterations of the algorithm.

Examples

require(mixtools)
x1 <- mixtools::rmvnorm(n= 250, mu= c(3.5,12), sigma= matrix(c(1,0,0,4),nrow=2,ncol=2))
x2 <- mixtools::rmvnorm(n= 350, mu= c(3,22), sigma= matrix(c(1,0,0,6),nrow=2,ncol=2))
x2[x2[,2]>24, 2] <- x2[x2[,2]>24, 2] - 24
x <- rbind(x1,x2,deparse.level=0)
x <- x[sample.int(nrow(x)),]
res <- circlust(x)
cat('mixing proportions:',res$parameters$pro,'\nmean of cluster 1:\n',res$parameters$mean[,1],
    '\ncovariance matrix of cluster 1:\n',res$parameters$sigma[,,1],
    '\nmean of cluster 2:\n',res$parameters$mean[,2],
    '\ncovariance matrix of cluster 2:\n',res$parameters$sigma[,,2])

gilles-guillot/HMMVMS documentation built on Dec. 23, 2019, 6:30 p.m.