score {supclust}

Wilcoxon Score for Binary Problems

Description

For a set of n observations grouped into two classes (for example n expression values of a gene), the score function measures the separation of the classes. It can be interpreted as counting for each observation having response zero, the number of individuals of response class one that are smaller, and summing up these quantities.

Usage

score(x, resp)

Arguments

x

Numeric vector of length n, for example containing gene or cluster expression values of n different cases.

resp

Numeric vector of length n containing the “binary” class labels of the cases. Must be coded by 0 and 1.

Value

A numeric value, the score. The minimal score is zero, the maximal score is the product of the number of samples in class 0 and class 1. Values near the minimal or maximal score indicate good separation, whereas intermediate score means poor separation.

Author(s)

Marcel Dettling, [e-mail]

See Also

wilma also for references; margin is the second statistic that is used there.

Examples

data(leukemia, package="supclust")
op <- par(mfrow=c(1,3))
plot(leukemia.x[,69],leukemia.y)
title(paste("Score = ", score(leukemia.x[,69], leukemia.y)))

## Sign-flipping is very important
plot(leukemia.x[,161],leukemia.y)
title(paste("Score = ", score(leukemia.x[,161], leukemia.y),2))
x <- sign.flip(leukemia.x, leukemia.y)$flipped.matrix
plot(x[,161],leukemia.y)
title(paste("Score = ", score(x[,161], leukemia.y),2))
par(op)