clusterkernel: Bagged cluster kernel

In 3inar/einr: Different Functions That I Keep Needing

Description

Implements the bagged cluster kernel desctibed by Weston et al. Calculates a distance metric between samples based on co-occurence in k-means clustering. The algorithm clusters the data using a range of different values for k in k-means and then uses the fraction of times two data points occur in the same cluster as their similarity.

Usage

clusterkernel(x, ncluster, ninit = 100)

Arguments

x	A matrix of data points, samples in the rows.
ncluster	Maximum number of clusters to use (maximum k for kmeans).
ninit	Number of times to repeat the clusterings. As k-means has some randomness to it, it might be a good idea to run it several times.

Value

A nrow(x) by nrow(x) matrix of pairwise similarities in the range [0, 1]

References

Weston, J., Leslie, C., Ie, E., Zhou, D., Elisseeff, A., & Noble, W. S. (2005). Semi-supervised protein classification using cluster kernels. Bioinformatics, 21(15), 3241-3247.

3inar/einr documentation built on May 5, 2019, 10:43 a.m.