How to force FindClusters to return a specific number of clusters?

Asked Sep 01 '20 at 00:22

Active Sep 06 '20 at 21:01

Viewed 154 times

I am using FindClusters with the Method -> "KMedoids" option. I am specifying a desired number of clusters. It returns with many fewer than the desired number. Is there a way to force it to return the desired number? For reference a code snippet is

    nclusters = 1207;
    Length /@ {vectors, weights}
    weighteddata = WeightedData[vectors, weights];
    clusters = FindClusters[weighteddata, nclusters, Method -> "KMedoids"];
    Length[clusters];

output is

{58852, 58852}
453

All of the elements in vectors are distinct, so it is unclear why it finds only 453 clusters. The documentation says it will partition into "at most" nclusters, but does not say whether or how one can force a specific number of clusters.

Any help would be appreciated.

In response to a request, here is a worked example.

n = 10000;
k = 1000;
vectors = RandomReal[{-1, 1}, {n, 3}];
weights = RandomInteger[{1, 1000}, n];
clusters = 
  FindClusters[WeightedData[vectors, weights], k, 
   Method -> "KMedoids"];
Length[clusters]

output is

edited Sep 06 '20 at 21:01

asked Sep 01 '20 at 00:22

abwatson

1,919
1
12
19

Please provide a runnable example. – C. E. Sep 01 '20 at 07:04
The example above requires 58852 vectors of length 3. I am not sure how to provide that here. – abwatson Sep 01 '20 at 13:26
1

RandomInteger, RandomReal or some kind of function like that can be used to generate data. SeedRandom can make sure that everyone sees the same data. – C. E. Sep 01 '20 at 13:30
1

I have added a worked example. – abwatson Sep 07 '20 at 14:05

How to force FindClusters to return a specific number of clusters?

0 Answers0