AnchorGraph: Generate an Anchor Graph from an n x p data matrix

In SemiSupervised: Safe Semi-Supervised Learning Tools

Description

The AnchorGraph function is designed for:

1) Graph generation: creates an ‘anchor’ graph object.

2) Graph modification: modify an existing ‘anchor’ graph using the fit.g argument, which is used primarily for predicting new observations.

The getAnchor function is a wrapper to kmeans clustering.

Usage

AnchorGraph(x,metric="cosine",anchor=NULL,fit.g=NULL,control=SemiSupervised.control())
getAnchor(x,control)

Arguments

x	the n x p ‘vector’, ‘matrix’ or ‘data.frame’.
fit.g	an existing ‘anchor’ object to be updated with new observations given by argument x.
anchor	an optional matrix of user provided anchor points. If NULL then link{kmeans} is used to generate the anchors.
metric	the metric either cosine or sqDist for graph construction.
control	the SemiSupervised.control is used to initialize various LAE settings. This control should also be used when executing the agraph function but is not enforced.

Details

The AnchorGraph function creates an informal S3-object of class ‘anchor’. This is required as input to the S4 generic function agraph when using the ‘anchor’ version or agraph.default version.

This call is performed in the ‘formula’, ‘data.frame’, ‘matrix’, or ‘vector’ S4 generic instances of agraph.

Value

Z	the n x k Z-matrix where each row has at most sfrac (refer to SemiSupervised.control) non-zero entries.
rL	reduced Laplacian matrix.
g.scaling	the scaling used to scale the x data prior to graph construction.
anchor	the anchor points.
metric	the metric.

Note

An S3-object was sufficient for our purposes. There is no need for the overhead or flexibility of a S4-class which is why it is programmed this way.

Author(s)

Mark Vere Culp

See Also

agraph, aG

Examples

## Set up Sonar data with 20% labeled
library(mlbench)
data(Sonar)

n=dim(Sonar)[1]
p=dim(Sonar)[2]

nu=0.2
set.seed(100)
L=sort(sample(1:n,ceiling(nu*n)))
U=setdiff(1:n,L)

y.true<-Sonar$Class
Sonar$Class[U]=NA

g.agraph1<-agraph(Class~.,data=Sonar) 

##The following gives an equivalent output to the g.agraph1<-agraph(Class~.,data=Sonar) call.

ctrl<-SemiSupervised.control()
g<-AnchorGraph(x.scaleL(Sonar[,-p],L),control=ctrl)
g.agraph2<-agraph(Class~.+aG(g),data=Sonar,control=ctrl) 

## For performance comparison check against agraph with graph only
tab=table(fitted(g.agraph2)[U],y.true[U])
1-sum(diag(tab))/sum(tab)

## Fit agraph to Sonar but graph only 
g.agraph3<-agraph(Class~aG(g),data=Sonar,control=ctrl) 
g.agraph3
tab=table(fitted(g.agraph3)[U],y.true[U])
1-sum(diag(tab))/sum(tab)

SemiSupervised documentation built on May 11, 2018, 5:03 p.m.