spw = stepwise.res[[cj]]
vars = spw$model[,2]

my.lda <- function(dad,stepwisevars) {
	library(MASS)
	library(stringr)
	#cria um training e um test set
	rnn = rownames(dad)
	rnn = t(as.data.frame(str_split(rnn,"_")))
	rownames(rnn) = rownames(dad)
	spun = unique(as.vector(rnn[,1]))
	test.n = NULL
	train.n = NULL
	for (j in 1:length(spun)) {
		rnu = rownames(rnn[rnn[,1]%in%spun[j],])
		srnu = length(rnu)
		ntest = floor(srnu*0.3)
		ntr = srnu-ntest
		test.n = c(test.n,sample(rnu,ntest))
		train.n = c(train.n,sample(rnu,ntr))
	}
	#seleciona o fator especies
	rz = as.vector(rnn[,1])
	spp = gsub("([0123456789])", "", rz)
	spp = str_trim(spp)
	spp = as.factor(spp)
	dad$species = spp
	dad.test = dad[test.n,]
	spp.test = as.factor(dad.test$species)
	dad.test = dad.test[,vars]
	dad.train = dad[train.n,]
	spp.train = as.factor(dad.train$species)
	dad.train = dad.train[,vars]
		
	#faz a discriminante com o training set
	ldd = lda(as.matrix(dad.train),spp.train)
	
	#prediz com o test set
	ldp = predict(ldd,dad.test)

	#checa a predicao
	pred = as.vector(ldp$class)
	org = as.vector(spp.test)
	pp = as.data.frame(cbind(org,pred))
	pp$ACERTO = 0
	pp[pp[,1]==pp[,2],'ACERTO'] = 1
	mistakes = pp[pp$ACERTO==0,1:2]
	pacert = (nrow(pp[pp$ACERTO==1,])/nrow(pp))*100
	return(list(pacert,mistakes))
}