This is the distribution of the ClusPath code. It is distributed under the GNU General Public License v3.0. This code was developed as a research prototype and it should be treated as such. It comes with no guarantees or support, however questions can be addressed to [email protected].
The ClusPath source code and two datasets (CPDS1 dataset and European Companies). For the description of the ClusPath algorithm, please check the paper:
Marian-Andrei Rizoiu, Julien Velcin, Stéphane Bonnevay, & Stéphane Lallich. ClusPath: A Temporal-driven Clustering to Infer Typical Evolution Paths. Data Mining and Knowledge Discovery (DAMI), 30(5), pp. 1324-1349, (2016). paper here
The current code was developed and tested under Linux using Octave v3.8.1, with the following packages installed:
- control
- econometrics
- general
- io
- missing-functions
- nan
- optim
- struct
- statistics
Install additional tools for plotting (in Ubuntu):
$> sudo apt-get install graphviz epstool
Octave is a fast-changing environment and new functionalities appear constantly, while old functionalities might be retired without notice. Workarounds might be required to make ClusPath function with the newest version of Octave.
Start octave in the main folder of the project. Upon lauching octave, the ClusPath code is loaded into the Octave path (check the .octaverc file).
$> octave
Next, execute ClusPath with parameters: Alpha = 0.95, Beta = 0.0002, Delta = 3, lamb1 = 1, lamb2 = 10, lamb3 = 1000
octave:2> [Adj, clusassign, dtimestamp, centroids, ctimestamp] = temporalKMeans('file', 'data/cpds1.csv', 'typeT', 5, 'Alpha', 0.95, 'Beta', 0.0002, 'Delta', 3, 'lamb1', 1, 'lamb2', 10, 'lamb3', 1000, 'saveResults', true );
Evaluate the clustering using the four measures:
octave:3> evaluateTemporalClustering('inputfile', 'data/cpds1.csv', 'clusassign', clusassign, 'centroids', centroids, 'ctimestamp', ctimestamp, 'Alpha', 0.95);