The toolkit can be downloaded from Attachments

PbSOMBox Version 1.0:  This toolbox includes Matlab implementations of probabilistic self-organizing maps (PbSOM) learning algorithms, i.e., KohonenGaussian, SOCEM, SOEM, and SODAEM, which are presented in the paper: "Model-based Clustering by Probabilistic Self-Organizing Maps," IEEE Trans. on Neural Networks, 2009. This toolbox is released by Shih-Sian Cheng, who is the corresponding author of the paper.      

Demostration

  • Simuliations on 2D data (You need to click on the (GIF) figures to see the demo)
    • Data set: The data set consists of 2000 points uniformly distributed in a unit square.
    • Network structure: In the experiments, an 5 by 5 equally spaced square lattice in a unit square is used as the structure of the SOM network.
  • Visualization of high-dimentional data
    • Data set: The Ecoli data set from UCI machine learning repository, which consists of 336 8-dimensional feature vectors.  It is comprised of eight classes, namely cp: C, im: I, pp: P, imU: U, om: O, omL: M, imL: L, and imS: S. The numbers of data samples are 143, 77, 52, 35, 20, 5, 2, and 2, respectively.
    • Network structure: In the experiments, an 7 by 7 equally spaced square lattice in a unit square is used as the structure of the SOM network.
    • Results: For each algorithm we see that the topological relationships among data clusters can be visualized on a two-dimensional network (lattice).
ċ
PbSOMBox_Version1.0.zip
(523k)
鄭士賢,
Aug 16, 2010, 1:04 AM
Comments