# Nick Zinns wrapper around the neural net function

# ----------------------------------------------------------------
testnnet<-function(x, y, out = 0.5, ...)
{
	if(is.vector(x) == T){x <- as.matrix(x,ncol=1)}
	if(nrow(x) != length(y))
		stop("x and y must be equal length")
	if(out >= 1 || out <= 0)
		stop("Out must be a decimal value between 0 and 1")
	training <- sample(length(y), floor(abs(length(y) * (1 - out))))
	validation <- seq(1, length(y))[ -training]	## Run nnet
	fit <- nnet(x[training,  ], y[training], ...)
	
       	
	##make out of sample prediction
       if(ncol(x) == 1){
	outsample.pred <- ifelse((predict.nnet(fit, matrix(x[validation,  ],ncol=1))
                                  > 0.5), 1, 0)	}
	if(ncol(x) != 1){
       outsample.pred <- ifelse((predict.nnet(fit, x[validation,  ])
                                  > 0.5), 1, 0)	}
   

	##assemble comparison
	validate.data <- as.matrix(y[validation])
 	
	##print the results
	print(fit)
	print(paste(length(training), "observations in training set"))
	print(paste(length(validate.data), "observations in validation set"))
	output.table <- table(validate.data, outsample.pred)




	cat("\nOut of sample prediction table:\n")
       print(output.table)
	##return the nnet object
	
	invisible(fit)
}