hClust_llr: Hierarchical Clustering based on log-likelihood ratio

In ziLRhClust: Hierarchical Clustering Based on Log-Likelihood Ratio for Zero-Inflated Data

Description

hClust_llr returns an object of type hclust, results of the clustering on individuals, using the dissimilarity based on log-likelihood ratio test.

Usage

hClust_llr(X, ZI = TRUE, reorder = TRUE)

Arguments

X	matrix n x p with observations (i.e. individuals) in rows and variables in columns
ZI	boolean, indicating whether using zero-inflated Gaussian model (if TRUE) or just gaussian model (if FALSE), default is TRUE
reorder	boolean, indicating whether individuals should be reorder or not (for heatmap aesthetic), if TRUE, use a re-implementation of reorder.hclust

Details

Algorithm to process hierarchical clustering on individuals (from a matrix individuals x variables). The dissimilarity used in the observation space is the log-likelihood ratio, i.e. D(i,j) = log f(Y_i) + log f(Y_j) - log f(Y_i U Y_j) where Y_i is the vector of observations from individual i. The statistical model defining the considered likelihood is standard Gaussian or Zero-Inflated Gaussian (i.e. a Gaussian-Bernoulli mixture). The agglomeration method is also the log-likelihood ratio, however between clusters in this case.

Wrapper for Cpp function

Value

an object of type hclust, see hclust

diss	the n x n dissimilarity matrix between individuals
	(not included in hclust object)
height	hclust
merge	hclust
order	hclust
labels	vector of n individual labels

Author(s)

Ghislain Durif, [e-mail] Franck Picard, [e-mail]

See Also

plot_heatmap

ziLRhClust documentation built on May 2, 2019, 5:24 p.m.