Search code examples
cluster-analysisdbscanelki

Change scatterplot labels in ELKI result visualization

Is it possible to easily change the scatterplot labels in the ELKI result visualization, maybe using parameters?

The default axes are labeled Column 0 and Column 1 for 2-dimensional data. It would be nice to change this to Latitude and Longitude for my case. I'm running DBSCAN with ELKI version 0.6.5.

Solution

  • There is no parameter for this. ELKI already has way too many parameters, doesn't it?

    But (I believe there is a bug in 0.6.5 there, you may need to build the current git version yourself) if there is a column label available, ELKI will use the column label instead of enumerating axes. If the first row of a file does not contain numbers, the default parser will interpret is as column labels.

    This is a visualization of clustering 1 day of coordinates in the 1% Twitter sample with HDBSCAN*:

    enter image description here

    Or you just export the graph to SVG and use a SVG editor and change the label as desired. I recommend Inkscape. The ability to fine-tune graphics with Inkscape was the key reason why we chose to use SVG.