Utilizing Structure-rich Features to improve Clustering



The source code for DipExt/DipInit is included here and can easily be executed in R-Studio (or any other R editor) and should be executable 
without any additional tweaking. The file main.R contains all necessary steps, i.e. the libraries which need to be installed and the functions
which are imported from function.R with source("functions.R"). The script is set up that it executes DipExt/DipInit on the data sets. All steps 
are further explained in the comments in the script.



Real-World Data and Results

The real-world data we used can be found as a original under the following links. We converted them into .txt files, to make them compatible to 
our code. No values or other things were changed. 

    SkinSegmentation
https://archive.ics.uci.edu/ml/datasets/skin+segmentation
    Banknote Authentication
https://archive.ics.uci.edu/ml/datasets/banknote+authentication
    Iris
https://archive.ics.uci.edu/ml/datasets/iris
    User Knowledge Modeling
https://archive.ics.uci.edu/ml/datasets/User+Knowledge+Modeling
    Breast Tissue
http://archive.ics.uci.edu/ml/datasets/breast+tissue
    Foresttypes training
https://archive.ics.uci.edu/ml/datasets/Forest+type+mapping
    Mice Protein Expression
https://archive.ics.uci.edu/ml/datasets/Mice+Protein+Expression
    SonyAIBORobotSurface1
http://timeseriesclassification.com/description.php?Dataset=SonyAIBORobotSurface1
    Proximal Phalanx Outline AgeGroup
http://timeseriesclassification.com/description.php?Dataset=ProximalPhalanxOutlineAgeGroup
    MoteStrain
http://timeseriesclassification.com/description.php?Dataset=MoteStrain
    DiatomSizeReduction	
http://timeseriesclassification.com/description.php?Dataset=DiatomSizeReduction

The Running Example goes by the name of "runEx". The data sets can all be found in the folder "datasets", though we could not include them in the 
Supplement, due to the 10mb-size restriction.


Labels/AMI

The results for the compared techniques, i.e. the labels, are be in the "\label" directory. Due to the extensive testing, 
which produced an extensive amount of labels (20.000+), are the labels in a separate .zip file, as it is otherwise extremly unpractical to handle. 
These labels are only included in the download, not in the Supplement, due to the size restriction.
Simply extract the labels, and run the pythonskript, if you want to check the AMI- values.
This script to compute the AMI-values can also be found in the download. Its a simple python-skript that loads the labels and computes the AMI. 
The used AMI implementation is from the SKLEARN-library. R has currently no implementation which seems satisfying to us (the aricode-version 
seems erronous to us, as the NMI- values are identical to the AMI-values which should not be)
