R help - Jul 2009 - Help with Loop!

Dear experts,

I'm new in R and trying to learn by writing a version of the Perceptron
Algorithm. How can I tell in the code below to stop the iteration when the
condition in the "for loop" is not satisfied for all training
examples?

Thanks in advance for your help!


## Generate a linearly separable data set in R2

sample <- as.data.frame(cbind(runif(n=100),runif(n=100)))
sample$Y <- ifelse(sample$V1>sample$V2,1,-1)
## Plot data;
attach(sample)
plot(V1, V2, type="p", pch=Y,main="Sample Data")

##Perceptron algorithm

sample_m <- as.matrix(sample)
w <- c(0,0); b <- 0; k <- 0; nu <- 1e-3
R <- max(sqrt(V1^2+V2^2))

repeat {
 for (i in 1:nrow(sample_m)){
    if  (sample_m[i,3]*(t(w)%*%sample_m[i,1:2] + b) <= 0)
          w <- w + nu*sample_m[i,3]*sample_m[i,1:2]
          b <- b +nu*sample_m[i,3]*R^2
          k <- k+1
          cat(k, w, b, "\n") }
     }

	[[alternative HTML version deleted]]

Hi,

On Jul 23, 2009, at 7:30 PM, Lars Bishop wrote:
> Dear experts,
>
> I'm new in R and trying to learn by writing a version of the  
> Perceptron
> Algorithm. How can I tell in the code below to stop the iteration  
> when the
> condition in the "for loop" is not satisfied for all training  
> examples?
>
> Thanks in advance for your help!
>
>
> ## Generate a linearly separable data set in R2
>
> sample <- as.data.frame(cbind(runif(n=100),runif(n=100)))
> sample$Y <- ifelse(sample$V1>sample$V2,1,-1)
> ## Plot data;
> attach(sample)
> plot(V1, V2, type="p", pch=Y,main="Sample Data")
>
> ##Perceptron algorithm
>
> sample_m <- as.matrix(sample)
> w <- c(0,0); b <- 0; k <- 0; nu <- 1e-3
> R <- max(sqrt(V1^2+V2^2))
>
> repeat {
> for (i in 1:nrow(sample_m)){
>    if  (sample_m[i,3]*(t(w)%*%sample_m[i,1:2] + b) <= 0)
>          w <- w + nu*sample_m[i,3]*sample_m[i,1:2]
>          b <- b +nu*sample_m[i,3]*R^2
>          k <- k+1
>          cat(k, w, b, "\n") }
>     }

One thing to realize is that by using a variable named "sample",  
you're trampling over the "sample" function ... just don't
trip over
that later in your R career :-)

I whipped up two version of your repeat block (I'm using while  
instead, though). One uses the inner for loop like you have, the  
second one is doing the same thing but w/ no for loops:

set.seed(123)
examples <- as.data.frame(cbind(runif(n=100),runif(n=100)))
examples$Y <- ifelse(examples$V1 > examples$V2, 1, -1)
examples <- as.matrix(examples)

# plot(examples[,1], examples[,2], type="p", pch=examples[,3],  
main="Sample Data")

b <- 0; k <- 0; nu <- 1e-3;
R <- max(sqrt(examples[,1]^2 + examples[,2]^2))
w <- matrix(0, nrow=2, ncol=1)

misclassified <- examples[,3] * (examples[,1:2] %*% w) <= 0
while (any(misclassified)) {
   for (i in which(misclassified)) {
       w <- w + nu * examples[i,3] * examples[i,1:2]
       b <- b + nu * examples[i,3]*R^2
       k <- k + 1
       cat(k, w, b, "\n")
   }
   misclassified <- examples[,3] * (examples[,1:2] %*% w) <= 0
}


## Here's the while loop w/o for loops
misclassified <- examples[,3] * (examples[,1:2] %*% w) <= 0
while (any(misclassified)) {
   w <- w + colSums(nu * examples[misclassified,3] *  
examples[misclassified,1:2])
   b <- b + sum(nu * examples[misclassified,3] * R^2)
   k <- k + 1
   cat(k, w, b, "\n")
   misclassified <- examples[,3] * (examples[,1:2] %*% w) <= 0
}

HTH,
-steve

--
Steve Lianoglou
Graduate Student: Physiology, Biophysics and Systems Biology
Weill Medical College of Cornell University

Contact Info: http://cbio.mskcc.org/~lianos