AffinityParam-class: Affinity propogation

In LTLA/bluster: Clustering Algorithms for Bioconductor

Affinity propogation

Description

Use affinity propagation from the apcluster package to cluster observations. Note that this requires the installation of the apcluster package.

Usage

AffinityParam(
  s = NULL,
  p = NA,
  q = NA,
  maxits = 1000,
  convits = 100,
  lam = 0.9,
  nonoise = FALSE
)

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

Arguments

s	A function that accepts a matrix of observations by dimensions and returns a similarity matrix. If NULL, defaults to the output of negDistMat with r=2.
p, q	Numeric scalars controlling the input preference, i.e., the resolution of the clustering. These are passed to the apcluster function, where values of NA are the default.
maxits, convits, lam, nonoise	Further arguments to pass to the apcluster function.
x	A numeric matrix-like object where rows represent observations and columns represent variables.
BLUSPARAM	A AffinityParam object.
full	Logical scalar indicating whether the full affinity propagation statistics should be returned.

Details

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

Setting q (and less typically, p) allows us to tune the resolution of the clustering. In particular, when p=NA, it is computed based on the setting of q:

• If the specified q lies in [0, 1], p is defined as the q-quantile of the finite similarities across all pairs of observations. When q=NA, it defaults to 0.5.

• If q is negative, p is defined as the M + abs(M) * q where M is the smallest finite similarity across all pairs. This yields smaller p values while still responding to the scale of the similarities.

The resulting value is used as the self-preference, i.e., the diagonal of the availability matrix. Larger values yield more clusters as each data point is more inclined to form its own cluster.

Value

The AffinityParam constructor will return a AffinityParam 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 similarity, the similarity matrix; and apcluster, the direct output of apcluster).

Author(s)

Aaron Lun

See Also

apcluster from the apcluster package, which does all of the heavy lifting.

Examples

clusterRows(iris[,1:4], AffinityParam())
clusterRows(iris[,1:4], AffinityParam(q=0.9))
clusterRows(iris[,1:4], AffinityParam(s=apcluster::expSimMat()))

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