getQ: getQ

In aucm: AUC Maximization

Description

getQ calculates Q or Q.pred matrix depending on the value of do.pred.

Usage

getQ (K,n1,n2,call.C=TRUE,do.pred=FALSE)

Arguments

K	kernel matrix of dimension (n1+n2) by (n1+n2) or n.pred by (n1+n2). The (n1+n2) observations must be ordered case followed by non-case
n1	number of cases
n2	number of non-cases
call.C	boolean. If TRUE, make .C call, otherwise compute Q in R.
do.pred	boolean. If TRUE, K is a n.pred by (n1+n2) matrix; otherwise, it is a (n1+n2) by (n1+n2) matrix.

Value

A n1*n2 by n1*n2 matrix if do.pred is FALSE, or n.pred by n1*n2 matrix if do.pred is TRUE

Author(s)

Shuxin Yin
Youyi Fong [e-mail]
Krisztian Sebestyen

Examples

dat = sim.dat.1(n=100,seed=1)
dat=rbind(subset(dat, y==1), subset(dat, y==0))
X = as.matrix(subset(dat, select=c(x1,x2)))
n1=sum(dat$y)
n2=sum(1-dat$y)

K = kyotil::getK(X,"linear", 1)
Q1 = getQ(K,n1=n1,n2=n2,call.C=FALSE)
Q2 = getQ(K,n1=n1,n2=n2,call.C=TRUE)
all(Q2-Q1<1e-6)

# compare to a direct computation
X.diff=get.X.diff(X[1:n1,], X[1:n2+n1,])
Q3 = tcrossprod(X.diff, X.diff)
all(Q3-Q1<1e-6)

# two printouts of Q2 should not be different
Q2[1:3,1:3]
K = kyotil::getK(X,"rbf", 1)
Q4 = getQ(K,n1=n1,n2=n2,call.C=TRUE)
Q2[1:3,1:3]
Q4[1:3,1:3]

K = kyotil::getK(X[1:10,],"linear", 1, X2=X)
Q5 = getQ(K,n1=n1,n2=n2,call.C=FALSE,do.pred=TRUE)
Q6 = getQ(K,n1=n1,n2=n2,call.C=TRUE,do.pred=TRUE)
dim(Q5)
dim(Q6)
all(Q5-Q6<1e-6)

aucm documentation built on Jan. 11, 2020, 9:43 a.m.