Scribd
Upload
267 views3 pages

Clustering On Boston Dataset

The document describes k-means clustering and provides code to demonstrate it. It explains k-means clustering with inputs for the number of clusters k, number of initializing trials nstart, and maximum number of iterations iter.max. It then performs k-means clustering on Boston housing data for 3 clusters as a test case, and calculates the within-sum-of-squares for each cluster and overall to validate the k-means calculations.

Uploaded by

anubhav582
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
267 views3 pages

Clustering On Boston Dataset

The document describes k-means clustering and provides code to demonstrate it. It explains k-means clustering with inputs for the number of clusters k, number of initializing trials nstart, and maximum number of iterations iter.max. It then performs k-means clustering on Boston housing data for 3 clusters as a test case, and calculates the within-sum-of-squares for each cluster and overall to validate the k-means calculations.

Uploaded by

anubhav582
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as TXT, PDF, TXT or read online on Scribd
You are on page 1/ 3

###

#
# Introduction to Clustering demo
#
# k-means clustering
#
# History:
#
# 2018/07/11 Initial code (copied from other example documents) walter
johnston
#
###
#
# k-means algorithm description/explanation:
#
# user- supplied input parameters:
#
# k: number of clusters
# nstart: number of initializing trials
# iter.max: maximum number of iterations (repetitions)
#
# (1) randomize observations into "k" initial groups (keep best of
"nstart" trials)
# (2) calculate centroid (vector of arithmetic means) for each cluster
# (3) calculate within-sum-of-squares (ESS) for each cluster (retain
value)
# (4) re-assign observations into closest cluster
# (distance from centroid; retain count of re-assignments)
# (5) if iterations <= "iter.max" and count of re-assignments > 0, go to
(2)
# (6) finished
#
###
# code to fetch the package if it is not present

if ( !require(MASS ) ) { install.packages('MASS'); library(MASS) }


if ( !require(tidyverse) ) { install.packages('tidyverse'); library(tidyverse) }
if ( !require(broom ) ) { install.packages('broom'); library(broom) }

data(Boston) # from MASS

###
#
# filter for complete cases (no missing data)
#
###
myBoston <- Boston[ complete.cases(Boston), ]

dim(Boston)
dim(myBoston)
###
#
# kmeans: fixed number of clusters
#
# cluster data based on error-sum-of-squares
# random starting points
# user specified number of clusters
# user specified number of starting trials (best one is automatically selected)
# user specified maximum number of iterations
#
###
# user choices
#
seed <- 1 # random number generator seed
minClusters <- 1 # minimum number of clusters (see code)
maxClusters <- 20 # maximum number of clusters (see code)
km.nstart <- 10 # number of starting trials
km.iter.max <- 20 # iteration limit

###
# start skip of code
###

# track the behavior (initialize to an illegal value) [ total within-ss ] (elbow


method)
#sse <- rep(-1, maxClusters)

# test results for each clustering scenario


#for (i in minClusters:maxClusters) {
# set.seed(seed) # reset RNG each time
# sse[i] <- kmeans(myBoston,
# centers=i,
# nstart=km.nstart,
# iter.max=km.iter.max)$tot.withinss
# }

# plot results to find optimal tradeoff (num clusters v. total winthin-ss)


#plot( minClusters:maxClusters,
# sse,
# type="b",
# xlab="Number of Clusters",
# ylab="Aggregate within cluster SSE (sum of squared error)")

# zoom in to identify the choice more easily


#clust1 <- 3
#clust2 <- 8
#sse2 <- sse[ clust1:clust2 ]
#plot( clust1:clust2,
# sse2,
# type="b",
# xlab="Number of Clusters",
# ylab="Aggregate within cluster SSE (sum of squared error)")

###
# skip to here
###

# test case: 3 clusters


set.seed(seed)
t <- kmeans(myBoston,
centers=3,
nstart=km.nstart,
iter.max=km.iter.max)

# measures of interest
table(t$cluster)
t$tot.withinss # overall
t$withinss # by cluster

# reconstruct the error measurements


t2 <- myBoston
t2$cluster <- t$cluster # add cluster number for each row (observation)

# calculate penalty function (within group error sum of squares)


wssf <- function(df) {
t <- scale(df, center=T, scale=F) # center around the mean
return( sum(t^2) )
}

# by cluster
t3 <- t2 %>%
group_by(cluster) %>%
do( data.frame(wss = wssf(.) ) )

sum(t3) # overall
t3 # by cluster

# individual divergences
t$withinss - t3$wss

# overall divergence
t$tot.withinss - sum(t3$wss)

###
#
# k-means WSS calculations woking correctly
#
# now, apply it to hclust() to select a number of clusters
#
# use: squared euclidean distance as metric for clustering
# method="complete"
#
###

You might also like