spa (version 2.0)

update.spa: Update procedure for transductive prediction with the SPA

Description

This implements the transductive prediction for an spa object. It performs regularization/region approach for transductive prediction. In addition it can also updates an existing spa object with new transductive estimate.

Usage

"update"(object,ynew,xnew,gnew, type=c("vector","probs","coef","all"), reg=c("ridge","hlasso"),trans.update=FALSE, dat=list(k=0,l=Inf),verbose=FALSE,FUN=sum,...)

Arguments

object

an object of type spa

ynew

an object of type spa

xnew

an object of type spa

gnew

an object of type spa

type

the type of predictions in classification, classes, probabilities or both. In the case of both the object will return an additional penalty vector corresponding to the rate function for each case.

reg

for regression it is automatically taken as a ridge penalty. In the case of classification one can use either ridge or the hyperbolic l1 penalty (hlasso).

trans.update

comming soon

verbose

comming soon

dat

data driven estimation routine consists of list k for the number of vertex sets, and l for the regularization (see reference). default dat=list(k=0,l=Inf)

FUN

measure used to sort WUL, the unlabeled-labeled partition. The FUN=sum multiplies WUL times a vector of ones, others may include max.

...

additional arguments passed into the function

References

M. Culp (2011). spa: A Semi-Supervised R Package for Semi-Parametric Graph-Based Estimation. Journal of Statistical Software, 40(10), 1-29. URL http://www.jstatsoft.org/v40/i10/.

Examples

## Use simulated example (two moon)
set.seed(100)
dat=spa.sim(type="moon") 

##Use spa to train with a supervised/transductive kernel smoother
gsemi<-spa(dat$y,graph=as.matrix(daisy(dat[,-1])))

##Use spa to update the object with new data
dat<-rbind(dat,spa.sim(100,0))
gsemi<-update(gsemi,ynew=dat$y,,as.matrix(daisy(dat[,-1])),trans.update=TRUE)