K-Means Clustering for functional data

Perform k-means clustering on functional data.

Usage

kmeans.center.ini(
  fdataobj,
  ncl = 2,
  metric = metric.lp,
  draw = TRUE,
  method = "sample",
  max.iter = 100,
  max.comb = 1e+06,
  par.metric = NULL,
  ...
)

kmeans.fd(
  fdataobj,
  ncl = 2,
  metric = metric.lp,
  dfunc = func.trim.FM,
  max.iter = 100,
  par.metric = NULL,
  par.dfunc = list(trim = 0.05),
  method = "sample",
  cluster.size = 5,
  draw = TRUE,
  ...
)

Arguments

fdataobj: fdata class object.
ncl: See details section.
metric: Metric function, by default metric.lp.
draw: =TRUE, draw the curves in the color of the centers.
method: Method for selecting initial centers. If method="Sample" (by default) takes n times a random selection by the ncl centers. The ncl curves with greater distance are the initial centers. If method="Exact" calculated all combinations (if < 1e+6) of ncl centers. The ncl curves with greater distance are the initial centers (this method may be too slow).
max.iter: Maximum number of iterations for the detection of centers.
max.comb: Maximum number of initial selection of centers (only used when method="exact").
par.metric: List of arguments to pass to the metric function.
...: Further arguments passed to or from other methods.
dfunc: Type of depth measure, by default FM depth.
par.dfunc: List of arguments to pass to the dfunc function .
cluster.size: Minimum cluster size (by default is 5). If a cluster has fewer curves, it is eliminated and the process is continued with a less cluster.

Value

Return:

cluster: Indexes of groups assigned.
centers: Curves centers.

Details

The method searches the locations around which are grouped data (for a predetermined number of groups).

If ncl=NULL, randomizes the initial centers, ncl=2 using kmeans.center.ini function.
If ncl is an integer, indicating the number of groups to classify,
are selected ncl initial centers using kmeans.center.ini function.
If ncl is a vector of integers, indicating the position of the initial centers with length(ncl) equal to number of groups.
If ncl is a fdata class objecct, ncl are the initial centers curves with nrow(ncl) number of groups.

References

Hartigan, J. A. and Wong, M. A. (1979). A K-means clustering algorithm. Applied Statistics 28, 100 \-108.

Author

Manuel Febrero-Bande, Manuel Oviedo de la Fuente manuel.oviedo@udc.es

Examples

if (FALSE) { # \dontrun{
library(fda.usc)
data(phoneme)
mlearn <- phoneme$learn[c(1:50,101:150,201:250),]
# Unsupervised classification
out.fd1 <- kmeans.fd(mlearn,ncl=3,draw=TRUE)
out.fd2 <- kmeans.fd(mlearn,ncl=3,draw=TRUE,method="exact")
# Different Depth function
ind <- c(17,77,126)
out.fd3 <- kmeans.fd(mlearn,ncl=mlearn[ind,],draw=FALSE,
dfunc <- func.trim.FM,par.dfunc=list(trim=0.1))
out.fd4 <- kmeans.fd(mlearn, ncl=mlearn[ind,], draw=FALSE, 
dfunc = func.med.FM)
group <- c(rep(1,50), rep(2,50),rep(3,50))
table(out.fd4$cluster, group)
} # }