Skip to content

Instantly share code, notes, and snippets.

Created January 21, 2011 23:25
What would you like to do?
R k-nearest neighbors example
#I'm using two non-standard packages; one for nearest neighbors (FNN) and one for doing split-combine-apply (think map reduce) type operations.
#you'll have to run this: install.packages(c('FNN', 'plyr'))
#you can also checkout documentation in R using the question mark; run this: ?knn
#Here are some random points on the plane to show you the interface
neartop = ldply(1:50, function(i){
c( x = rnorm(1)
, y = 1 + rnorm(1, sd=0.5))
nearbottom = ldply(1:50, function(i){
c( x = rnorm(1)
, y = -1 + rnorm(1, sd=0.5))
# the first 50 are near the top, the rest are near the bottom
points = rbind(neartop, nearbottom)
#the vector of classification labels is easy to construct. Since the labels are TRUE/FALSE, the question we're asking is, "is this point in the set near the top?"
points[,'label'] = (1:100 < 51)
plot(points$x, points$y)
#split into training/testing sets by sampling 60 numbers from the set of 1--100, using them for training, and the rest for testing.
train_i = sample(1:100, 60)
train = points[train_i,c('x','y')]
test = points[-train_i,c('x','y')]
results = knn(
, test
, cpt_vecs
, cl = points[train_i,'label']
, k=1) #k = 1 just assigns a point to its nearest neighbor.
#percent correct
sum(points[-train_i,'label'] == results) / nrow(test)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment