Counting one too many clusters? about dbscan

[peekxc]

Perhaps there is something right with the file linked, but could you clarify the question a bit?

Running dbscan with the first two columns of your data results in 32 clusters with 0 noise points for me:

data <- read.csv("~/Downloads/42041320940000.csv")
data <- apply(as.matrix(na.omit(data))[, 1:4], 2, as.numeric)
res <- dbscan(data, eps=50, minPts = 1)

After converting the spreadsheet to csv. Looking at the spreadsheet attached, there are a number of things I notice immediately:

• dimensions 3 and 4 are blank for record 1
• dimension 4 is blank for record 2
• There's a random 'x' character for dimension 5 at record 19, I assume this was unintentional?

I used second statement (apply) removes this data and cleans the data set, as dbscan expects a numeric matrix as input.

In your comments, you only mentioned dimensions A and B as the data. What about dimensions C and D? Also, may I ask why you are expecting 11 clusters?

[jdd112]

I apologize for the confusing excel document i have clarified it in the most recent version. I think the cluster count should be 29. I am mirroring the logic in Columns C & D. Column c calculates the distance to the next point and column d counts the clusters when these points are more than 50 away. could you please look this over and see if i am missing something?
Copy of 42041320940000.xlsx

[peekxc]

DBSCAN does 29 clusters with 0 noise.

data <- read.csv("~/Documents/Copy.of.42041320940000.csv")
res <- dbscan(as.matrix(data[, 1:2]), eps=50, minPts = 1) # also works with just column 1
res 

res
DBSCAN clustering for 191 objects.
Parameters: eps = 50, minPts = 1
The clustering contains 29 cluster(s) and 0 noise points.
...

It matches your fourth column as well

all(res$cluster == data[, 4]) # true 

[jdd112]

yup, you are right ... I had max(res$cluster)+1 on my output ... so sorry to waste your time. Great work on this project.