Finding the k-largest clusters in dbscan result - r

Question

I have a dataframe df, consists of 2 columns: x and y coordinates.
Each row refers to a point.
I feed it into dbscan function to obtain the clusters of the points in df.
library("fpc")
db = fpc::dbscan(df, eps = 0.08, MinPts = 4)
plot(db, df, main = "DBSCAN", frame = FALSE)
By using print(db), I can see the result returned by dbscan.
> print(db)
dbscan Pts=13131 MinPts=4 eps=0.08
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
border 401 38 55 5 2 3 0 0 0 8 0 6 1 3 1 3 3 2 1 2 4 3
seed 0 2634 8186 35 24 561 99 7 22 26 5 75 17 9 9 54 1 2 74 21 3 15
total 401 2672 8241 40 26 564 99 7 22 34 5 81 18 12 10 57 4 4 75 23 7 18
22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
border 4 1 2 6 2 1 3 7 2 1 2 3 11 1 3 1 3 2 5 5 1 4 3
seed 14 9 4 48 2 4 38 111 5 11 5 14 111 6 1 5 1 8 3 15 10 15 6
total 18 10 6 54 4 5 41 118 7 12 7 17 122 7 4 6 4 10 8 20 11 19 9
45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68
border 2 4 2 1 3 2 1 1 3 1 0 2 2 3 0 3 3 3 3 0 0 2 3 1
seed 15 2 9 11 4 8 12 4 6 8 7 7 3 3 4 3 3 4 2 9 4 2 1 4
total 17 6 11 12 7 10 13 5 9 9 7 9 5 6 4 6 6 7 5 9 4 4 4 5
69 70 71
border 3 3 3
seed 1 1 1
total 4 4 4
From the above summary, I can see cluster 2 consists of 8186 seed points (core points), cluster 1 consists of 2634 seed points and cluster 5 consists of 561 points.
I define the largest cluster as the one contains the largest amount of seed points. So, in this case, the largest cluster is cluster 2. And the 1st, 2nd, 3th largest clusters are 2, 1 and 5.
Are they any direct way to return the rows (points) in the largest cluster or the k-largest cluster in general?
I can do it in an indirect way.
I can obtain the assigned cluster number of each point by
db$cluster.
Hence, I can create a new dataframe df2 with db$cluster as the
new additional column besides the original x column and y
column.
Then, I can aggregate the df2 according to the cluster numbers in
the third column and find the number of points in each cluster.
After that, I can find the k-largest groups, which are 2, 1 and 5
again.
Finally, I can select the rows in df2 with third column value equals to 2 to return the points in the largest cluster.
But the above approach re-computes many known results as stated in the summary of print(db).

Answer 1

The dbscan function doesn't appear to retain the data.
library(fpc)
set.seed(665544)
n <- 600
df <- data.frame(x=runif(10, 0, 10)+rnorm(n, sd=0.2), y=runif(10, 0, 10)+rnorm(n,sd=0.2))
(dbs <- dbscan(df, 0.2))
#dbscan Pts=600 MinPts=5 eps=0.2
# 0 1 2 3 4 5 6 7 8 9 10 11
#border 28 4 4 8 5 3 3 4 3 4 6 4
#seed 0 50 53 51 52 51 54 54 54 53 51 1
#total 28 54 57 59 57 54 57 58 57 57 57 5
attributes(dbs)
#$names
#[1] "cluster" "eps" "MinPts" "isseed"
#$class
#[1] "dbscan"
Your indirect steps are not that indirect (only two lines needed), and these commands won't recalculate the clusters. So just run those commands, or put them in a function and then call the function in one command.
cluster_k <- function(dbs, data, k){
kth <- names(rev(sort(table(dbs$cluster)))[k])
data[dbs$cluster == kth,]
}
cluster_k(dbs=dbs, data=df, k=1)
## x y
## 3 6.580695 8.715245
## 13 6.704379 8.528486
## 23 6.809558 8.160721
## 33 6.375842 8.756433
## 43 6.603195 8.640206
## 53 6.728533 8.425067
## a data frame with 59 rows