creating AUC

gistfile1.r

library("ROCR")
library(plyr)

meshClasses <- list.files("partition0")
extractClass <- function(x) {  a <- strsplit(x, "\\."); a[[1]][1] }

meshClasses <- sapply(meshClasses, function(x) { extractClass(x) })

getOriginal <- function(i, mesh) { 
	originalFile <- paste("partition", i, "/", mesh, ".libsvm", sep="")
	x <- system(paste("cat ", originalFile ," | cut -c1-2", sep=""), intern=TRUE)
	m <- as.table(matrix(x, byrow=TRUE, ncol=1)) 
	as.data.frame(list(value=as.numeric(m[,1])))
}
getPredictions <- function(i, mesh) { 
	predictionsFile <- paste("out/", mesh, ".", i, ".out", sep="")
	read.table(predictionsFile)	
}
maxFold <- 7
lapply(meshClasses, function(mesh) { 
	frame <- data.frame(labels=rep(NA, maxFold), predictions=rep(NA, maxFold))
	for (i in 0:maxFold) {
		frame$labels[i] <- getOriginal(i, mesh)
		frame$predictions[i] <- getPredictions(i, mesh)
	}
	frame
}) -> pred

lapply(pred, function(d) { 
	p <- prediction(d$predictions, d$labels)
	perf <- performance(p, "auc")
	vals <- unlist(attr(perf, 'y.values'))
	data.frame(mean=mean(vals), sd.dev=sd(vals))
}) -> allAUC
df<- ldply(allAUC, data.frame)
df$.id <- sapply(df$.id, function(x) { substr(extractClass(x), 5, 1000L) } )

run SVM

CATEGORIES := $(patsubst partition0/%.libsvm,%,$(wildcard partition0/*.libsvm))
FOLDS := 0 1 2 3 4 5 6 7

all: $(patsubst %,%.out,$(CATEGORIES))

clean:
	rm -f *.train *.test *.model

dist-clean: clean
	rm -f *.out

define category_targets
$(1).out: $$(patsubst %,$(1).%.out,$$(FOLDS))
	touch $(1).out

$(1).%.out: $(1).%.test $(1).%.model 
	predict $$^ $$@ > $(1).$$*.log 

.SECONDARY: $$(patsubst %,$(1).%.out,$$(FOLDS))

$(1).%.model: $(1).%.train
	train -s 2 $$< $$@ 

.SECONDARY: $$(patsubst %,$(1).%.model,$$(FOLDS))

$(1).%.test: partition%/$(1).libsvm
	cp $$< $$@

$(1).%.train: $$(patsubst %,partition%/$(1).libsvm,$$(FOLDS))
	cat $$(filter-out partition$$*/$(1).libsvm,$$^) > $$@
endef

$(foreach category,$(CATEGORIES),$(eval $(call category_targets,$(category))))