Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!
Get Started for Free

pdfCluster-class

Class "pdfCluster"


Description

This class pertains to results of the application of function pdfCluster.

Objects from the Class

Objects can be created by calls of the form new("pdfCluster", ...) or as a result to a call to pdfCluster.

Slots

call:

Object of class "call" representing the matched call;

x:

Object of class "matrix" representing the clustered data points;

pdf:

Object of class "list" reporting details about the kernel density estimate at data points x.

nc:

Object of class "list" summarizing the result of the connected components search for different sections of the estimated density.

graph:

An object of class "list" defining details about the graph built to find the connected sets of high density regions.

cluster.cores:

Object of class "ANY" reporting the group labels of the data allocated to the cluster cores.

tree:

Object of class "ANY", namely class dendrogram if the procedure detects more than one group, list otherwise. It reports the cluster tree structure associated to the different connected components for different density levels.

noc:

Object of class "numeric" giving the number of clusters.

stages:

Object of class "ANY", being NULL if the cluster cores only are detected, "list" when also the lower density data are allocated. The elements of the list correspond to the group labels at the different stages of the classification procedure. NA values correspond to unlabeled data.

clusters:

Object of class "ANY" being NULL if the cluster cores only are detected, "numeric" when all the data are clustered. This slot is obsolete. Groups can be extracted by a call to function groups.

See pdfCluster for further details.

Methods

dbs

signature(x = "pdfCluster", clusters = "missing")

Computes the density based Silhouette diagnostics of clustered data. See dbs for further details.

pdfCluster

signature(x="pdfCluster")

Speeds up time for re-running the pdfCluster procedure with different values of tau when graphtype = "pairs"

plot

signature(x = "pdfCluster", y = "missing")

Plots objects of pdfCluster-class. plot-methods are available for:

  • the mode function: gives the number of connected components when the proportion of data points with density above a threshold varies. Set argument which to 1 to display this plot.

  • the cluster tree: plot the hierarchical structure associated to the clusters detected by different sections of the density estimate. Set argument which to 2 to display this plot.

  • the data points: scatterplot of data or of all the possible couples of coordinates reporting the label group. Set argument which to 3 to display this plot.

  • the density-based Silhouette information: graphical diagnostics of the clustering. See plot,dbs-method. Set argument which to 4 to display this plot. Not available when noc=1.

See plot,pdfCluster-method for further details.

show

signature(object = "pdfCluster").

Prints the following elements:

  • the matched Call;

  • the type of kernel estimator;

  • the type of graph built;

  • the groups tree (if available);

  • the cluster cores;

  • the cluster labels at the different stages of the classification procedure;

  • the final clustering.

summary

signature(object = "pdfCluster").

Provides a summary of pdfCluster-class objects by printing the following elements:

  • the matched call to pdfCluster function

  • the frequency table of the cluster cores;

  • the frequency table of the final grouping;

  • the tree of clusters.

See Also

Examples

showClass("pdfCluster")

data(wine)
x <-wine[ ,-1]
gr <- wine[ ,1]

# clustering
cl <- pdfCluster(x, graphtype="pairs", bwtype="adaptive")
summary(cl)
cl
plot(cl)

pdfCluster

Cluster Analysis via Nonparametric Density Estimation

v1.0-3
GPL-2
Authors
Adelchi Azzalini, Giovanna Menardi for the current version; Adelchi Azzalini, Giovanna Menardi and Tiziana Rosolin up to version 0.1-13.
Initial release
2018-12-04

We don't support your browser anymore

Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.