I'm using ELKI GUI to run DBSCAN algorithm. My input is a CSV file. I create a projection as feature selection:
-dbc.filter transform.ProjectionFilter -projection NumericalFeatureSelection -projectionfilter.selectedattributes 1,2
ELKI gives me some files as clusters. These files contain only projected attributes and other attributes such as UserId not exists. How can I identify UserId for each object in clusters?
The easiest approach is not using projection to remove unwanted columns (which will literally remove these columns), but instead treat them as labels. Because that is what they are, labels.
Usually, the most reliable approach is to include a unique identifier (object label) with each object, if you want to be able to identify points afterwards.
In other words, use
-parser.labelIndices 0
instead of your filter (choose your label columns, of course - assuming your user id is in column 0).
