DensityCut: The densityCut algorithm

In poipou/cytosee: automatic computation and evaluation of cytometry data

Description

The densityCut algorithm

Usage

DensityCut(X, K, knn.index, knn.dist, V, D, G, threshold, alpha = 0.9,
  nu = seq(0, 1, by = 0.05), adjust = TRUE, maxit = 50, eps = 1e-05,
  col, show.plot = TRUE, show.tip.label = FALSE, debug = FALSE,
  xlab = TRUE, text = NULL, ...)

Arguments

X	A data matrix (columns are features and rows are data points)
K	A integer to specify the number of neighbours in building the Knn graph. Default to K=\log_2(N), where N is the number of data points
knn.index	An N*K data matrix for the nearest neighbour indices
knn.dist	An N*K data matrix for the nearest neighbour distances
V	The initial density vector of length N
D	The dimensionality of data
G	A sparse Knn graph, reseaved for extension
threshold	A number between 0 and 1 specifying the saliency index to cut the tree. If not specified, it is selecting by stability analysis of the clustering tree
alpha	The damping factor between 0 and 1, default to 0.90
nu	The saliency index in merging trees, default to seq(0.0, 1.0, by=0.05)
adjust	Lotical, whether to ajdust valley height or not
maxit	The maximum number of iteration allowed in density refinement, default to 50
eps	The threshold in density refinement, default to 1e-5
col	A vector of clours
show.plot	Logical, whether to draw clustering results
show.tip.label	Logical, whether to draw the tip labels of trees
debug	Logical, whether to print debug information
xlab	Logical, whether to show the xlab
text	subplot label
...	Reserved for extension

Value

A list contains the clustering memberships, the modes of each cluster, and the estimated densities at each point

Examples

library(mvtnorm)

data(distinct.col)
set.seed(0)

N = 2^12
number.cluster = 64
N = N / number.cluster

i  = j = seq(-3.5, 3.5, by=1)
mu = expand.grid(i, j)
mu = as.matrix(mu)

sigma = matrix(c(1, 0, 0, 1)*0.05, byrow=TRUE, nrow=2)

x = lapply(seq(number.cluster), function(z) rmvnorm(N, mu[z,], sigma))
x = do.call(rbind, x)

label = lapply(1:number.cluster, function(z) rep(z, N))
col = AssignLabelColor(distinct.col, unlist(label))
NeatPlot(x, col=col, pch=4, cex=0.5)

K = ceiling(log2(N * number.cluster))
a = DensityCut(X=x, K=K, alpha=0.85, nu=seq(0.0, 1.0, by=0.05),
               debug=FALSE, show.plot=TRUE,
               col=distinct.col)

col = AssignLabelColor(distinct.col, a$cluster)
NeatPlot(x, col=col, pch=4, cex=0.5)

poipou/cytosee documentation built on May 28, 2019, 4:39 p.m.