Unnecessary check of dimensions for input?

[BioTurboNick]

Is this check necessary?

Would be nicer to emit a trivial clustering rather than to error.

Clustering.jl/src/dbscan.jl

Line 104 in 74d5d43

num_points <= dim && throw(ArgumentError("points has $dim rows and $num_points columns. Must be a D x N matric with D < N"))

[lwabeke]

Bump on this, I don't have a lot of experience with DBSCAN, but I don't understand the point of this exception.
From what I can determine, the code should excecute without this check.
Not sure if I would call clustering any more trivial in this case, consider that expanding dim from 1 to 100 does not change the original inherent problem of deciding if the 2 points are close together or not (assuming a version of dbscan that ignores this check):

dim = 100
points = zeros(dim)
points[1,:] .= 1.0
clustersTogether = dbscan(points, 1)

points[1,2] = 10
clustersApart  = dbscan(points, 1)
length(clustersTogether.counts) != length(clustersApart.counts)

[BioTurboNick]

I was assuming that the error was added because incorrect results can be produced if the number of dimensions is greater than the number of points present. So by "trivial clustering" I meant "each point is a cluster". or an empty set if min_cluster_size is greater than 1.

[alyst]

Thank you for the report!
Can you submit the PR with the fix and the tests case (D >= N) confirming that the algorithm works correctly?