Cluster Evaluation using Pair Segments

At ICDE 2012 we presented an experimental cluster evaluation tool based in pair segments. Details can be found in (see also Publications):

0.5

Elke Achtert, Sascha Goldhofer, Hans-Peter Kriegel, Erich Schubert, Arthur Zimek
Evaluation of Clusterings – Metrics and Visual Support.
Proceedings of the 28th International Conference on Data Engineering (ICDE), Washington, DC, 2012.

Setup

For this experiment, you will need at least version 0.5.0 (Releases).

Launch the MiniGUI as usual and setup your clustering algorithms as usual (if you want to compare to a ground truth labeling, add a clustering.trivial.TrivialByLabelClustering!).

As evaluator choose clustering.pairsegments.ClusterPairSegmentAnalysis, and the UI should automatically include the circle segments visualization. Also add clustering.EvaluateClustering to see cluster scores.

In order to simplify the user interface, you can set for example

-vis.window.single
-visualizer.stylesheet presentation
-vis.enable (ScatterPlot|CircleSegments|Evaluation)

The following two detailed settings also include

• Detailed configurations for two clustering algorithms on this data set
• Embedding visualizers, to allow interaction in the overview (-vis.window.single)
• Customized visualization by selectively enabling just some visualizers (by default, much more visualizers are used, making the overview a bit complicated)

Figure 5+6

-dbc.in pov.csv
-algorithm clustering.EM,clustering.DBSCAN
# Parameters for EM clustering
-em.k 3
# Parameters for DBSCAN clustering
-dbscan.epsilon 0.1
-dbscan.minpts 5
# Parameters for Evaluation and Visualization
-evaluator clustering.EvaluateClustering,clustering.pairsegments.ClusterPairSegmentAnalysis
-vis.window.single
-visualizer.stylesheet presentation
-vis.enable (scatterplotfact|emcluster|voronoi|marker|axis|Evalu|Circle)

Figure 7+8

-dbc.in different-densities-2d.ascii
-algorithm clustering.EM,clustering.kmeans.KMeansLloyd
# Parameters for EM clustering
-em.k 6
-kmeans.initialization KMeansPlusPlusInitialMeans
-kmeans.seed 1
# Parameters for kmeans clustering
-kmeans.k 5
-kmeans.initialization KMeansPlusPlusInitialMeans
-kmeans.seed 1
# Parameters for Evaluation and Visualization
-evaluator clustering.EvaluateClustering,clustering.pairsegments.ClusterPairSegmentAnalysis
-vis.window.single
-visualizer.stylesheet presentation
-vis.enable (scatterplotfact|voronoi|marker|axis|Evalu|Circle)