rclust.dbscan: DBSCAN for Manifold-valued Data

Description

Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is a generally applicable clustering algorithm as long as we have concept of dissimilarity. We adopt dbscan algorithm by dbscan package. See dbscan for more details.

Usage

rclust.dbscan(input, type = c("extrinsic", "intrinsic"), eps, minPts = 5, ...)

Arguments

input

a S3 object of riemdata class. See riemfactory for more details.

type

type of distance, either "intrinsic" or "extrinsic".

eps

size of the epsilon neighborhood.

minPts

number of minimum points in the eps region (for core points). Default is 5 points.

...

extra parameters for DBSCAN algorithms including eps or minPts. See dbscan for more details.

Value

an object of class dbscan_fast containing

eps: value of the eps parameter.
minPts: value of the minPts parameter.
cluster: An integer vector with cluster assignments. Zero indicates noise points.

References

ester_density-based_1996RiemBaseExt

hahsler_dbscan_2019RiemBaseExt

Examples

Run this code

# NOT RUN {
## generate 50 points near (0,0,1)  and
##          50 points near (0,0,-1) on Sphere S^2 
ndata = 50
theta = seq(from=-0.99,to=0.99,length.out=ndata)*pi
tmpx  = cos(theta) + rnorm(ndata,sd=0.1)
tmpy  = sin(theta) + rnorm(ndata,sd=0.1)

## wrap it as 'riemdata' class
data  = list()
for (i in 1:ndata){
  tgt = c(tmpx[i],tmpy[i],1)
  data[[i]] = tgt/sqrt(sum(tgt^2)) # project onto Sphere
}
for (i in 1:ndata){
  tgt = c(tmpx[i],tmpy[i],-1)
  data[[i+ndata]] = tgt/sqrt(sum(tgt^2)) # project onto Sphere
}
data = RiemBase::riemfactory(data, name="sphere")

## compare extrinsic and intrinsic DBSCAN
dbext <- rclust.dbscan(data, eps=0.5, type="extrinsic")
dbint <- rclust.dbscan(data, eps=0.5, type="intrinsic")

## let's visualize the results via MDS in R^2
pdist = stats::as.dist(RiemBase::rbase.pdist(data))
dat2d = stats::cmdscale(pdist, k=2)

opar <- par(no.readonly=TRUE)
par(mfrow=c(1,2))
plot(dat2d[,1], dat2d[,2], col=dbext$cluster, pch=19, main="extrinsic+dbscan")
plot(dat2d[,1], dat2d[,2], col=dbint$cluster, pch=19, main="intrinsic+dbscan")
par(opar)

# }