EMIS/CSE 8331: DBSCAN Examples
Michael Hahsler
Wed Feb 13 21:05:54 2019
Introduction
Compare DBSCAN and other methods on a dataset with two clusters of different density with and without noise.
Dataset 1: Two squares with different density
x <- rbind(
cbind(x = runif(100, -2,2), y = runif(100, -2,2)),
cbind(x = runif(200, 2.5,3.5), y = runif(200, -.5,.5))
)
plot(x)
k-means
cl <- kmeans(x, 2)
plot(x, col = cl$cluster)
PAM
library(cluster)
cl <- pam(x, 2)
plot(x, col = cl$cluster)
hclust (complete link)
dend <- hclust(dist(x))
plot(dend)
plot(x, col = cutree(dend, k =2))
hclust (complete link)
dend <- hclust(dist(x), method = "single")
plot(dend)
plot(x, col = cutree(dend, k =2))
DBSCAN
library("dbscan")
kNNdistplot(x, 4)
cl <- dbscan(x, eps = .15, minPts = 4)
cl## DBSCAN clustering for 300 objects.
## Parameters: eps = 0.15, minPts = 4
## The clustering contains 2 cluster(s) and 96 noise points.
##
## 0 1 2
## 96 4 200
##
## Available fields: cluster, eps, minPtsplot(x, col = cl$cluster+1L)
cl <- dbscan(x, eps = .7, minPts = 4)
cl## DBSCAN clustering for 300 objects.
## Parameters: eps = 0.7, minPts = 4
## The clustering contains 2 cluster(s) and 1 noise points.
##
## 0 1 2
## 1 99 200
##
## Available fields: cluster, eps, minPtsplot(x, col = cl$cluster+1L)
Jarvis-Patrick clustering
cl <- jpclust(x, k = 10, kt = 5)
plot(x, col = cl$cluster+1L)
Dataset 2: Added noise
x <- rbind(x, cbind(x = runif(50, -5, 5), y = runif(50, -5, 5)))
plot(x)
k-means
cl <- kmeans(x, 2)
plot(x, col = cl$cluster)
PAM
library(cluster)
cl <- pam(x, 2)
plot(x, col = cl$cluster)
hclust (complete link)
dend <- hclust(dist(x))
plot(dend)
plot(x, col = cutree(dend, k =2))
hclust (complete link)
dend <- hclust(dist(x), method = "single")
plot(dend)
plot(x, col = cutree(dend, k =2))
DBSCAN
library("dbscan")
kNNdistplot(x, 4)
cl <- dbscan(x, eps = .15, minPts = 4)
cl## DBSCAN clustering for 350 objects.
## Parameters: eps = 0.15, minPts = 4
## The clustering contains 2 cluster(s) and 145 noise points.
##
## 0 1 2
## 145 4 201
##
## Available fields: cluster, eps, minPtsplot(x, col = cl$cluster+1L)
cl <- dbscan(x, eps = .5, minPts = 4)
cl## DBSCAN clustering for 350 objects.
## Parameters: eps = 0.5, minPts = 4
## The clustering contains 2 cluster(s) and 47 noise points.
##
## 0 1 2
## 47 102 201
##
## Available fields: cluster, eps, minPtsplot(x, col = cl$cluster+1L)
Jarvis-Patrick clustering
cl <- jpclust(x, k = 10, kt = 5)
plot(x, col = cl$cluster+1L)