The DTRN program (Mazanec, 2006) performs topology-sensitive vector quantization like the TRN with the additional feature of dynamically changing the number of classes in the partition. A new class is added and an existing class is deleted during the training when needed.

DTRN is based on Si, J., S. Lin, and M.-A. Vuong (2000). Dynamic Topology Representing Networks. Neural Networks, 13, 617-27. DTRN is a Matlab routine that requires the installation of the Matlab Runtime Machine v7.8 which is available in the table below. The download file dtrn.zip includes the VQ_DTRN.exe file and the batch file DTRN.bat for starting a DTRN run. It also contains a test data file named cassini791.dat. Here the data are two-dimensional so you can watch the network training. You may like to experiment with the 791 data points in the cassini791.dat file trying initial values for the vigilance parameter ranging between 10 (default) and .5. See read_me.txt for installation instructions and dtrn_outline.pdf for algorithmic details.

Note: You have to read and accept our download conditions before!