SomParam-class: Clustering with self-organizing maps

In LTLA/bluster: Clustering Algorithms for Bioconductor

Clustering with self-organizing maps

Description

Use the self-organizing map implementation in the kohonen package to cluster observations into the specified number of nodes. Note that this requires the installation of the kohonen package.

Usage

SomParam(
  centers,
  dim.ratio = 1,
  topo = "rectangular",
  neighbourhood.fct = "bubble",
  toroidal = FALSE,
  rlen = 100,
  alpha = c(0.05, 0.01),
  radius = NULL,
  dist.fct = "sumofsquares"
)

## S4 method for signature 'ANY,SomParam'
clusterRows(x, BLUSPARAM, full = FALSE)

Arguments

centers	An integer scalar specifying the number of centers. Alternatively, a function that takes the number of observations and returns the number of centers.
dim.ratio	A positive numeric scalar in specifying how centers should be distributed between the x and y dimensions. Defaults to equal distribution, i.e., both dimensions will be of length equal to the square root of centers. Values above 1 will distribute more nodes to x while values below 1 will distribute mode nodes to y.
topo, neighbourhood.fct, toroidal	Further arguments to pass to the somgrid function in the kohonen package.
rlen, alpha, radius, dist.fct	Further arguments to pass to the som function in the kohonen package.
x	A numeric matrix-like object where rows represent observations and columns represent variables.
BLUSPARAM	A SomParam object.
full	Logical scalar indicating whether the full SOM statistics should be returned.

Details

This class usually requires the user to specify the number of clusters beforehand. However, we can also allow the number of clusters to vary as a function of the number of observations. The latter is occasionally useful, e.g., to allow the clustering to automatically become more granular for large datasets.

Note that the final number of clusters may not be exactly equal to centers, depending on how dim.ratio is specified. For example, if centers is a perfect square and dim.ratio=1, we will get exactly the requested number of points.

To modify an existing SomParam object x, users can simply call x[[i]] or x[[i]] <- value where i is any argument used in the constructor.

For radius, a value of NULL means that the default argument in the som function signature is used. This is are data-dependent and so cannot be specified during construction of the SomParam object.

For dist.fct, users can specify any string that can be used in the dist.fcts arguments in som. In practice, the only real alternative is "manhattan".

Value

The SomParam constructor will return a SomParam object with the specified parameters.

The clusterRows method will return a factor of length equal to nrow(x) containing the cluster assignments. If full=TRUE, a list is returned with clusters (the factor, as above) and objects (a list containing som, the direct output of som).

Author(s)

Aaron Lun

See Also

som from the kohonen package, which does all of the heavy lifting.

FixedNumberParam, the parent of the SomParam class.

Examples

clusterRows(iris[,1:4], SomParam(centers=16))
clusterRows(iris[,1:4], SomParam(centers=12, dim.ratio=3/4))

LTLA/bluster documentation built on Aug. 20, 2023, 5:39 a.m.