inst/oldFiles/classInterfaces.R

In MLInterfaces: Uniform interfaces to R machine learning procedures for data in Bioconductor containers

setOldClass("knnP")
setOldClass("nnet.formula")
setOldClass("diana")
setOldClass("agnes")
setOldClass("pam")
setOldClass("rpart")
setOldClass("svm")
setOldClass("bclust")
setOldClass("fclust")
setOldClass("cshell")
setOldClass("ica")
setOldClass("lca")
setOldClass("naiveBayes")
setOldClass("pamr")
setOldClass("randomForest")
setOldClass("hclust")
setOldClass("kmeans")
setOldClass("prcomp")
setOldClass("classbagg")
setOldClass("bagging")
setOldClass("ipredknn")
setOldClass("slda")
setOldClass("lda")
setOldClass("qda")
setOldClass("nsc")
setOldClass("gbm")
setOldClass("logitboost")

knnP <- function(train, test, cl, k=1, l=0, prob=FALSE, use.all=TRUE) {
#
# idea here is to allow knn to work with predict method using new data.
# create a closure that knows about the training data, and later evaluate
# it on any conforming test data
#
# october 25 -- seem to need to use name newdata
# for this to work with generic prediction
#
 ans <- class::knn(train,test,cl,k,l,prob,use.all)
 nf <- function(train,cl,k,l,prob,use.all) function(newdata)
	 class::knn(train,newdata,cl,k,l,prob,use.all)
 attr(ans, "predfun") <- nf(train,cl,k,l,prob,use.all)
 class(ans) <- c("knnP", "factor")
 ans
}

predict.knnP <- function(object, ...) 
	attr(object, "predfun")(...)

print.knnP <- function(x, ...)
	{
	cat("instance of knnP [predictable knn object]\n")
	NextMethod()
	}

setGeneric("knnB", function(exprObj, classifLab, trainInd, 
		k=1, l=1, prob=TRUE, use.all=TRUE, metric="euclidean"){
			standardGeneric("knnB")
		})

setMethod("knnB", c("ExpressionSet", "character", "integer", 
			"ANY", "ANY", "ANY", "ANY", "ANY"), 
		function(exprObj, classifLab, trainInd, k, l, 
			prob, use.all, metric){

		cl <- pData(exprObj)[[classifLab]][trainInd]				
		trainDat <- t(exprs(exprObj)[,trainInd])
		testDat <- t(exprs(exprObj)[,-trainInd])
		dis <- dist(testDat, method=metric)
		out <- knnP(trainDat, testDat, cl, k, l, prob, use.all)
                new("classifOutput", method="knn", 
			predLabels=newPredClass(as.character(out)), 
			trainInds=trainInd, allClass=as.character(pData(exprObj)[[classifLab]]),
			predScores=newQualScore(attr(out,"prob")),
                        RObject=out, call=match.call(), distMat=dis)
                                                                                
})


#####################
# title: knn.cvB
# description: interface to knn.cv {class}
# arguments:
#	exprObj		ExpressionSet
#	classifLab	character string specifying what covariate data 
#			to use for classification
#	metric		for distance matrix 
# value:
# 	object of class "classif2Output"
# example:
# knn.cvOut <- knn.cvB(golubMerge[101:140,], "ALL.AML")
####################

#setGeneric("knn.cvB", function(exprObj, classifLab, trainInd=NULL, k=1, l=1, prob=TRUE, use.all=TRUE, metric="euclidean"){
#		standardGeneric("knn.cvB")
#})
#
#setMethod("knn.cvB", c("ExpressionSet", "character", "ANY", "ANY", "ANY", "ANY", "ANY"), 
#			function(exprObj, classifLab, trainInd=NULL, k, l, prob, use.all, metric){
#			if (!is.null(trainInd)) warning("disregarding trainInd for knn.cvB")
#			cl <- pData(exprObj)[[classifLab]]
#			dat <- t(exprs(exprObj))
#			dis <- dist(dat, method=metric)
#			out <- class::knn.cv(dat, cl, k, l, prob, use.all)
#                new("classifOutput", method="knn.cv", 
##			predLabels=newPredClass(as.character(out)), 
#			trainInds=integer(0), allClass=as.character(pData(exprObj)[[classifLab]]),
#			predScores=newQualScore(attr(out,"prob")),
#                        RObject=out, call=match.call(), distMat=dis)
#})

#####################
# title: knn1B
# description: interface to knn1 {class}
# arguments:
#	exprObj		ExpressionSet
#	trainInd	vector of indices for the columns to be 
#			included in the training set
#	classifLab	character string specifying what covariate data 
#			to use for classification
#	metric		for distance matrix 
# value:
# 	object of class "classif1Output"
# example:
# train <- c(sample(1:47, 21), sample(48:72, 12))
# knn1Out <- knn1B(golubMerge[100:200,], "ALL.AML", train)
####################

setGeneric("knn1B", function(exprObj, classifLab, trainInd, metric="euclidean"){
		standardGeneric("knn1B")
})

setMethod("knn1B", c("ExpressionSet", "character", "integer", "ANY"), 
		function(exprObj, trainInd, classifLab, metric){
				
		cl <- pData(exprObj)[[classifLab]][trainInd]
		trainDat <- t(exprs(exprObj)[,trainInd])
		testDat <- t(exprs(exprObj)[,-trainInd])
		dis <- dist(testDat, method=metric)
		out <- class::knn1(trainDat, testDat, cl)
                new("classifOutput", method="knn1", 
			predLabels=newPredClass(as.character(out)), 
			trainInds=trainInd, allClass=as.character(pData(exprObj)[[classifLab]]),
			predScores=newQualScore(attr(out,"prob")),
                        RObject=out, call=match.call(), distMat=dis)
})

#####################
# title: lvq1B
# description: interface to lvq1 {class}
# arguments:
#	exprObj		ExpressionSet
#	trainInd	vector of indices for the columns to be 
#			included in the training set
#	classifLab	character string specifying what covariate data 
#			to use for classification
#	metric		for distance matrix 
# value:
# 	object of class "classifPred"
# example:
# train <- c(sample(1:47, 21), sample(48:72, 12))
# lvq1Out <- lvq1B(golubMerge[100:200,], "ALL.AML", train)
####################

setGeneric("lvq1B", function(exprObj, classifLab, trainInd, size, prior, k=5, niter, alpha=0.03, metric="euclidean"){
		standardGeneric("lvq1B")
})

setMethod("lvq1B", c("ExpressionSet", "character", "integer", "ANY", "ANY", "ANY", "ANY", "ANY", "ANY"), 
		function(exprObj, classifLab, trainInd, size, prior, k, niter, alpha, metric){

		if(missing(size)){ size <- NULL }
		cl <- pData(exprObj)[[classifLab]][trainInd]
		trainDat <- t(exprs(exprObj)[,trainInd])
		testDat <- t(exprs(exprObj)[,-trainInd])
		dis <- dist(testDat, method=metric)

		cbkInit <- class::lvqinit(trainDat, cl, size=size, prior=prior, k=k)
		if(missing(niter)){ niter <- 100 * nrow(cbkInit$x) } 
		cbkTrain <- class::lvq1(trainDat, cl, cbkInit, niter=niter)
		out <- class::lvqtest(cbkTrain, testDat)
                new("classifOutput", method="lvq1", 
			predLabels=newPredClass(as.character(out)), 
			trainInds=trainInd, allClass=as.character(pData(exprObj)[[classifLab]]),
			#predScores=newQualScore(attr(out,"prob")),
                        RObject=cbkTrain, call=match.call(), distMat=dis)
})	

#####################
# title: lvq2B
# description: interface to lvq2 {class}
# arguments:
#	exprObj		ExpressionSet
#	trainInd	vector of indices for the columns to be 
#			included in the training set
#	classifLab	character string specifying what covariate data 
#			to use for classification
#	metric		for distance matrix 
# value:
# 	object of class "classifPred"
# example:
# train <- c(sample(1:47, 21), sample(48:72, 12))
# lvq2Out <- lvq2B(golubMerge[100:200,], "ALL.AML", train)
####################

setGeneric("lvq2B", function(exprObj, classifLab, trainInd, size, 
		prior, k=5, niter, alpha=0.03, win=0.3, metric="euclidean"){
		standardGeneric("lvq2B")
})

setMethod("lvq2B", c("ExpressionSet", "character", "integer", "ANY", "ANY", "ANY", "ANY", 
		"ANY", "ANY", "ANY"),
		function(exprObj, classifLab, trainInd, size, prior, k, 
			niter, alpha, win, metric){

		if(missing(size)){ size <- NULL }
		cl <- pData(exprObj)[[classifLab]][trainInd]
		trainDat <- t(exprs(exprObj)[,trainInd])
		testDat <- t(exprs(exprObj)[,-trainInd])
		dis <- dist(testDat, method=metric)
		cbkInit <- class::lvqinit(trainDat, cl, size=size, 
					prior=prior, k=k)
		if(missing(niter)){ niter <- 100 * nrow(cbkInit$x) } 
		cbkTrain <- class::lvq2(trainDat, cl, cbkInit, niter=niter, 										alpha=alpha, win=win)
		out <- class::lvqtest(cbkTrain, testDat)
                new("classifOutput", method="lvq2", 
			predLabels=newPredClass(as.character(out)), 
			trainInds=trainInd, allClass=as.character(pData(exprObj)[[classifLab]]),
			predScores=newQualScore(attr(out,"prob")),
                        RObject=cbkTrain, call=match.call(), distMat=dis)
	
})	

#####################
# title: lvq3B
# description: interface to lvq3 {class}
# arguments:
#	exprObj		ExpressionSet
#	trainInd	vector of indices for the columns to be 
#			included in the training set
#	classifLab	character string specifying what covariate data 
#			to use for classification
#	metric		for distance matrix 
# value:
# 	object of class "classifPred"
# example:
# train <- c(sample(1:47, 21), sample(48:72, 12))
# lvq3Out <- lvq3B(golubMerge[100:200,], "ALL.AML", train)
####################

setGeneric("lvq3B", function(exprObj, classifLab, trainInd, size, prior, k=5, 
		niter, alpha=0.03, win=0.3, epsilon=0.1, metric="euclidean"){
		standardGeneric("lvq3B")
})

setMethod("lvq3B", c("ExpressionSet", "character", "integer", "ANY", "ANY", "ANY", "ANY", "ANY", "ANY", "ANY", "ANY"),
		function(exprObj, classifLab, trainInd, size, prior, k, niter, alpha, win, epsilon, metric){

		if(missing(size)){ size <- NULL }
		cl <- pData(exprObj)[[classifLab]][trainInd]			
		trainDat <- t(exprs(exprObj)[,trainInd])
		testDat <- t(exprs(exprObj)[,-trainInd])	
		dis <- dist(testDat, method=metric)
		cbkInit <- class::lvqinit(trainDat, cl, size=size, prior=prior, k=k)
		if(missing(niter)){ niter <- 100 * nrow(cbkInit$x) } 
		cbkTrain <- class::lvq3(trainDat, cl, cbkInit, niter=niter, alpha=alpha, 
					win=win, epsilon=epsilon)
		out <- class::lvqtest(cbkTrain, testDat)
                new("classifOutput", method="lvq3", 
			predLabels=newPredClass(as.character(out)), 
			trainInds=trainInd, allClass=as.character(pData(exprObj)[[classifLab]]),
			predScores=newQualScore(attr(out,"prob")),
                        RObject=cbkTrain, call=match.call(), distMat=dis)
	
})	

#####################
# title: olvq1B
# description: interface to olvq1 {class}
# arguments:
#	exprObj		ExpressionSet
#	trainInd	vector of indices for the columns to be 
#			included in the training set
#	classifLab	character string specifying what covariate data 
#			to use for classification
#	metric		for distance matrix 
# value:
# 	object of class "classifPred"
# example:
# train <- c(sample(1:47, 21), sample(48:72, 12))
# olvq1Out <- olvq1B(golubMerge[100:200,], "ALL.AML", train)
####################

setGeneric("olvq1B", function(exprObj, classifLab, trainInd, size, prior, k=5, niter, alpha=0.03, metric="euclidean"){
		standardGeneric("olvq1B")
})

setMethod("olvq1B", c("ExpressionSet", "character", "integer", "ANY", "ANY", "ANY", "ANY", "ANY", "ANY"),
		function(exprObj, classifLab, trainInd, size, prior, k, niter, alpha, metric){

		if(missing(size)){ size <- NULL }
		cl <- pData(exprObj)[[classifLab]][trainInd]
		trainDat <- t(exprs(exprObj)[,trainInd])
		testDat <- t(exprs(exprObj)[,-trainInd])
		dis <- dist(testDat, method=metric)
		cbkInit <- class::lvqinit(trainDat, cl, size=size, prior=prior)
		if(missing(niter)){ niter <- 100 * nrow(cbkInit$x) } 
		cbkTrain <- class::olvq1(trainDat, cl, cbkInit, niter=niter, alpha=alpha)
		out <- class::lvqtest(cbkTrain, testDat)
                new("classifOutput", method="olvq1", 
			predLabels=newPredClass(as.character(out)), 
			trainInds=trainInd, allClass=as.character(pData(exprObj)[[classifLab]]),
			predScores=newQualScore(attr(out,"prob")),
                        RObject=cbkTrain, call=match.call(), distMat=dis)

})	

#####################
# title: SOMB
# description: interface to SOM {class}
# arguments:
#	exprObj		ExpressionSet
#	kx		x dimension
#	ky		y dimension
# 	topo		grid topology 
#	classifLab	character string specifying what covariate data 
#			to use for classification
# value:
# 	object of class "classifPred"
#	where sampLabels are the labels of the original sample
# example:
# somOut <- SOMB(golubMerge[100:200,], "ALL.AML", 2, 2)
####################

setGeneric("SOMB", function(exprObj, classifLab, kx=3, ky=3, topo="hexagonal", rlen=10000, 
		alpha=seq(0.05, 0, len=rlen), 
		radii=seq(4, 1, len = rlen), init, metric="euclidean"){
		standardGeneric("SOMB")
})

# a special container is provided for SOMB in somInterfaces.R