library(class)

names <- c("os",							#  1. Operating system
           "topos",						#  2. Generic operating system
			  "ipnum",						#  3. IP ID number
			  "npackets",					#  4. Number of packets seen
			  "meanttl",					#  5. Mean TTL
			  "meantos",					#  6. Mean type of service
			  "ntos",						#  7. Number distinct type of service
			  "meanlog2win",				#  8. mean(log(window size,2))
			  "selog2win",					#  9. sd(log(window size,2))
			  "nnwin",						# 10. Number distinct window sizes
			  "meandf",						# 11. mean(don't fragment flag set)
			  "meanincripid",				# 12. mean(diff(ipid))
			  "stdincripid",				# 13. sd(diff(ipid))
			  "meanlog2seq",				# 14. mean(log(sequence number,2))
			  "rangelog2seq",				# 15. range(log(sequence number,2))
			  "meanlog2incrseq",			# 16. mean(log(diff(sequence number),2))
			  "stdlog2incrseq",			# 17. sd(log(diff(sequence number),2))
			  "meanlog2sport",			# 18. mean(log(source port,2))
			  "maxlog2sport",				# 19. max(log(source port,2))
			  "minlog2sport",				# 20. min(log(source port,2))
			  "meanincrsport",			# 21. mean(diff(source port))
			  "stdincrsport",				# 22. sd(diff(source port))
			  "meannops", 					# 23. mean(number options)
			  "modeMSS",					# 24. mode maximum segment size
			  "ndiff_MSS",					# 25. number distinct maximum segment size
			  "ntimepstamps",				# 26. number timestamps
			  "nwscale",					# 27. number window scale
			  "nsackok",					# 28. number sack ok
			  "nendtcp",					# 29. number end tcp
			  "meaniptotlen",				# 30. mean(total length)
			  "stdiptotlen")				# 31. sd(total length)


osnames <- c("dos","irix","linux","apple","mac","solaris","windows")

a <- matrix(scan("ptrain.txt",comment.char="#"),byrow=T,ncol=31)
b <- matrix(scan("ptest.txt",comment.char="#"),byrow=T,ncol=31)

data <- rbind(a,b)

cls <- unique(data[,2])
nums <- rep(0,length(cls))
for(i in 1:length(cls)){
   nums[i] <- sum(data[,2]==cls[i])
}

training.data <- NULL
testing.data <- NULL

for(i in 1:length(cls)){
   x <- 1:nums[i]
	y <- seq(1,nums[i],by=2)
	z <- setdiff(x,y)
	q <- data[data[,2]==cls[i],]
	training.data <- rbind(training.data,q[y,])
	testing.data <- rbind(testing.data,q[z,])
}

variates <- c(5:24,26:30)

training.data2 <- training.data[,variates]
testing.data2 <- testing.data[,variates]
training.cls <- training.data[,2]
testing.cls <- testing.data[,2]

nam <- names[variates]


confusion.matrix <- function (truth, class) 
{
    n <- length(truth)
    if (n != length(class)) {
        stop("\nTruth and class must be of the same length")
    }
    nc <- max(truth) + 1
    conf <- matrix(0, ncol = nc, nrow = nc)
    for (i in 1:n) {
        if (class[i] >= 0) {
            conf[1 + truth[i], 1 + class[i]] <- conf[1 + truth[i], 
                1 + class[i]] + 1
        }
    }
    conf
}



mn <- apply(training.data2,2,min)
mx <- apply(training.data2,2,max)
td <- t((t(training.data2)-mn)/(mx-mn))


test1 <- function()
{
	require(class)
	require(utilities)

	tc <- as.factor(training.cls)

	ttd <-  t((t(testing.data2)-mn)/(mx-mn))

	ttc <- as.factor(testing.cls)
	k <- knn(td,ttd,tc,k=3)
	sum(k!=ttc)

	con <- confusion.matrix(testing.cls,as.integer(as.vector(k)))
	rownames(con) <- osnames
	colnames(con) <- osnames
	con
}

test2 <- function()
{
	require(class)
	require(utilities)

	x <- training.cls
	x[x==6] <- 0
	x[x==4] <- 3
	x[x==5] <- 4
	newosnames <- c("windows","irix","linux","mac","solaris")

	tc <- as.factor(x)

	ttd <-  t((t(testing.data2)-mn)/(mx-mn))

	y <- testing.cls
	y[y==6] <- 0
	y[y==4] <- 3
	y[y==5] <- 4
	ttc <- as.factor(y)

	k <- knn(td,ttd,tc,k=3)
	sum(k!=ttc)

	con <- confusion.matrix(y,as.integer(as.vector(k)))
	rownames(con) <- newosnames
	colnames(con) <- newosnames
	con
}


#        dos irix linux apple mac solaris windows
#dos       0    0     0     0   2       0      32
#irix      0   16     0     0   0       0       1
#linux     0    0    25     0   0       0       0
#apple     0    0     0     0   3       0       3
#mac       0    0     0     0  31       0       0
#solaris   0    0     0     0   0      27       0
#windows   1    0     6     1   3       0    1752

kf <- NULL
training.clsf <- NULL
testing.clsf <- NULL
k.crude <- as.integer(as.vector(k))
for(i in 0:6){
   z <- ttd[k.crude==i,]
	if(is.vector(z)) z <- matrix(z,nrow=1)
   testing.clsf <- c(testing.clsf,testing.data[k.crude==i,1])
   w <- td[training.cls==i,]
   k <- knn(w,z,as.factor(training.data[training.cls==i,1]))
	kf <- c(kf,as.integer(as.vector(k)))
}

con2 <- confusion.matrix(testing.clsf,kf)

write(t(con2),file="confusion2",ncol=30)