Mercator (version 1.1.6)

setClusters: Changing Cluster and Color Assignments in Mercator

Description

Cluster assignments from unsupervised analyses typically assign arbitrary integers to the classes. When comparing the results of different algorithms or different distance metrics, it is helpful to match the integers in order to use colors and symbols that are as consistent as possible. These functions help achieve that goal.

Usage

setClusters(DV, clusters)
recolor(DV, clusters)
remapColors(fix, vary)
recluster(DV, K)

Value

Both setClusters and remapColors return an object of class Mercator in which only the cluster labels and associated color and symbol representations (but not the distance metric used, the number of clusters, the cluster assignments, nor the views) have been updated.

The recolor function is currently an alias for setClusters. However, using recolor is deprecated, and the alias will likely be removed in the next version of the package.

Arguments

DV

An object of the Mercator class.

clusters

An integer vector specifiny the cluster membership of each element.

fix

An object of the Mercator class, used as the source of color and cluster assignments.

vary

An object of the Mercator class, used as the target of color and cluster assignments.

K

An integer, the number of clusters.

Author

Kevin R. Coombes <[e-mail]>

Details

In the most general sense, clustering can be viewed as a function from the space of "objects" of interest into a space of "class labels". In less mathematical terms, this simply means that each object gets assigned an (arbitrary) class label. This is all well-and-good until you try to compare the results of running two different clustering algorithms that use different labels (or even worse, use the same labels -- typically the integers \(1, 2, \dots, K\) -- with different meanings). When that happens, you need a way to decide which labels from the different sets are closest to meaning the "same thing".

The functions setClusters and remapColors solve this problem in the context of Mercator objects. They accaomplish this task using the greedy algorithm implemented and described in the remap function in the Thresher package.

See Also

remap

Examples

#Form a BinaryMatrix
data("iris")
my.data <- t(as.matrix(iris[,c(1:4)]))
colnames(my.data) <- 1:ncol(my.data)
my.binmat <- BinaryMatrix(my.data)

# Form a Mercator object; Set K to the number of known species
my.vis <- Mercator(my.binmat, "euclid", "tsne", K=3)
table(getClusters(my.vis), iris$Species)
summary(my.vis)

# Recolor the Mercator object with known species
DS <- recolor(my.vis, as.numeric(iris$Species))
table(getClusters(DS), iris$Species)

# Use a different metric
my.vis2 <- Mercator(my.binmat, "manhattan", "tsne", K=3)
table(getClusters(my.vis2), iris$Species)
table(Pearson = getClusters(my.vis2), Euclid = getClusters(my.vis))

# remap colors so the two methods match as well as possible
my.vis2 <- remapColors(my.vis, my.vis2)
table(Pearson = getClusters(my.vis2), Euclid = getClusters(my.vis))

# recluster with K=4
my.vis3 <- recluster(my.vis, K = 4)

# view the results
opar <- par(mfrow=c(1,2))
plot(my.vis, view = "tsne", main="t-SNE plot, Euclid")
plot(my.vis2, view = "tsne",  main="t-SNE plot, Pearson")
par(opar)