Non-linear Cluster Enhancement: Forcing Clusters into a compact shape
Code, data, results



Implementation:

Included in the download  is 1) the sourcecode of our algorithm PCE and 2) the data sets used in the experiments. 
The code is implemented in Java and should be easy to implement.
One has to add the path of the JdistLib-Library ("jdistlib-0.4.5-bin.jar") to get the Dip-test running, included in the download. 
For the eigenvalue decomposition of PCA we used the Apache Commons Math library ("commons-math3-3.6.1.jar"). It also needs to get
added as a path so that PCE can be executed and is included in the download. 

K-means is implemented on Source code found on the web, whose author has nothing to do with us. We simply used it, as it was 
practical at that time. Instead of that one might use any other k-means code.

The JdistLib-Library is found https://sourceforge.net/projects/jdistlib/ and is used for the Dip-Test.
The main class is run.java. Details are explained there, but simply executing the code should perform PCE on the running Example 
including creating various plots. Other aspects like re-computing NMI or re-running the experiments can also be done in the source code.



Real-World Data

The real-world data we used are in the directory "/dataset". The data sets are as follows:

Iris: https://archive.ics.uci.edu/ml/datasets/iris
Seeds: https://archive.ics.uci.edu/ml/datasets/seeds
Vertebral Column: https://archive.ics.uci.edu/ml/datasets/vertebral+column
Wifi Localization: https://archive.ics.uci.edu/ml/datasets/Wireless+Indoor+Localization"
Breast Cancer Wisconsin: https://archive.ics.uci.edu/ml/datasets/Breast+Cancer+Wisconsin+(Diagnostic)
Breast Tissue: https://archive.ics.uci.edu/ml/datasets/breast+tissue
Wine: https://archive.ics.uci.edu/ml/datasets/wine

We also saved the transformations of these datasets in the folder. Those transformations are saved with the suffix "AAA", as the name 
was the last thing we decided on, long after the experiments were done. Since we also run experiments on other transformations, these 
are also saved there. Eg. the PCA variant of Iris is called "4 IrisnormPCA". "norm" is added to all data sets, as all experiments are 
run in the normalized (in the [0,1]-range) version. The transformed version have the Suffixes "PCA", "ICA", "zero", "Dip", "AAA" for 
PCA, ICA, Z-Transformation, DipTransformation and PCE respectively.

The Running Example goes by the name of "runEx".

Labels
The results for the compared techniques, i.e. the labels, can be found in the directory "\label" in the source code.
Included in the source code main part is a routine that loads the labels and computed the NMI.

AMI/NMI
The measure for clustering quality is NMI. One could also use AMI instead, but as the two measures often do not differ relevantly, 
only the NMI results are included.