I implemented the DBSCAN algorithm in R, and I map the cluster assignments to the DBSCAN fpc library implementation . Testing is performed on synthetic data that is generated, as indicated in the dbscan example of the fpc library:
n <- 600
x <- cbind(runif(10, 0, 10)+rnorm(n, sd=0.2), runif(10, 0, 10)+rnorm(n, sd=0.3))
Clustering is performed with parameters as shown below:
eps = 0.2
MinPts = 5
I map cluster assignments fpc::dbscanto my implementation dbscan. The maximum number of runs shows that each point was classified equally by both implementations.
But there are cases when 1 or 2 points and some rare times 5 or 6 points are assigned to different clusters in my implementation than in the fpc implementation. I noticed that only the classification of border points is different. After plotting the graph, I saw that the points whose cluster membership does not coincide in the implementations are in such a position that it can be assigned to any of its surrounding clusters, depending on which point of the cluster seed it was found first.
I am showing an image with 150 points (to avoid clutter) where 1 point classification is different. Please note that the cluster number of difference points is always greater in my implementation than the fpc implementation.
Cluster graph.
The top insert is fpc :: dbscan, the bottom insert is my dbscan implementation
. , , (!)
:
dbscan
+
o -
-
!
fpc:: dbscan
-
-
-
:
dbscan
fpc:: dbscan
x-y
-
, , , fpc . . :
fpc:: dbscan ( ), ,
dbscan ( + ), ,
, : ?
, fpc::dbscan . ( !) . , , fpc::dbscan , . , ?
.