Options

"Empty clusters"

jeszyjeszy Member Posts: 13 Contributor II
edited June 2019 in Help
Clustering operators such as DBSCAB may produce empty clusters. Despite the fact that DBSCAN can identify all natural clusters in my data I got an extra empty cluster, that makes it impossible to use the "Map Clustering on Labels" operator to map cluster labels to class labels, because I got the following message:

        
The example sets label only contain 4 values while there are existing 5 clusters. Mapping impossible.
Without the empty cluster each cluster could be mapped to one of the classes naturally.

Is it possible to get rid of this annoying emtpy cluster?

Best regards,

Peter
Tagged:

Answers

  • Options
    MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869 Unicorn
    Hi Peter,

    if you apply the "Remove unused Values" operator on the output of DBSCAN it should remove the empty cluster. Don't forget to activate the parameter "include special attributes".

    Regards,
    Marius
  • Options
    jeszyjeszy Member Posts: 13 Contributor II
    Hi Marius,

    Unfortunately, that does not help since the Remove Unused Values operator can be applied to ExampleSets only. The empty cluster should be removed from the ClusterModel object.

    I think so it would be quite straightforward to add an "Ignore empty clusters" parameter to the "Map Clustering on Labels" operator that could solve the problem.

    Best regards.

    Peter
Sign In or Register to comment.