Clustering longitudinal data using the likelihood as a metric of distance

Description

'kmlcov' Cluster longitudinal data using the likelihood as a metric of distance. The generalised linear model allow the user to introduce covariates with different level effects (2 levels).

Details

Package: kmlcov
Type: Package
Version: 1.0.1
Date: 2013-08-21
Authors@R: c(person("Mamoun O", "Benghezal", role = c("aut", "cre"), email = "mobenghezal@gmail.com"), person("Christophe", "Genolini", role = c("ctb"), email = "christophe.genolini@u-paris10.fr"), )
License: GPL-2
Depends: methods
Collate: 'functions4glmClust.R' 'GlmCluster.R' 'glmclust-internal.R' 'glmClust.R' 'kmlcov-package.R' 'kmlCov.R'

Index:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
glmClust                Clustering longitudinal data
kmlCov                  Clustering longitudinal data from different
                        starting conditions
which_best              Seek the best partitions

Converge-class          Class '"Converge"'
GlmCluster-class        Class '"GlmCluster"'
KmlCovList-class        Class '"KmlCovList"'
addIndic                Create the new formula with the indicator
                        covariates
affect_rand             Affect randomly the individuals to the clusters
getNomCoef              Get the name of the coefficients in the 'glm'
                        object according to the current cluster
log_lik                 Calculate the log-likelihood
lowcyclo                Measures of creatinine and time among cardiac
                        transplant patients.
majIndica               Calculate an indicator vector
observance              Measures of obsevance.
plot,GlmCluster-method
                        Plot the main trajectories
plot,KmlCovList-method
                        Plot the main trajectories
predict_clust           Creates a character string expression to
                        calculate the predicted values
rwFormula               Rewrite the formula with all the covariates
seperateFormula         Separate the covariates in a formula
show,Converge-method    Method for function 'show'

Overview

To cluster longitudinal data, 'kmlcov' implement an ECM type algorithm which assign the trajectories to the cluster which maximise the likelihood.
It is possible to introduce covariates via the generalised linear model with different level effects (2 levels) all spedified in one formula.
The package implements the plot function to produce the diagrams at the condition of not having more than 2 different effects (although the program can deal with more than two effects) for e.g. time and treatment or time and sex q.v. the help of linkglmClust or kmlCov. To plot the main trajectories with more than two effects we recommand to use ggplot of the package ggplot2.

To cluster longitudinal data, 2 functions have to be remembered glmClust and kmlCov, the first run the algorithm for clustering one time and the second run the same algorithm multiple times with different starting conditions. The method is greatly sensitive to the initial conditions, we therefore recommand to use kmlCov although it takes much more time.

See Also

kmlCov
glmClust
which_best