Many existing implementations (like NeuroShell 2) only use 1D SOM. That simplified algorithm can be used as a classifier, but 2D SOM is certainly more powerful. It can still be used for classification purposes (and does a much better job at it than 1D SOM), but using 2D SOM allows for more accurate data mapping of the original data onto a 2D grid, hence reducing its dimensionality. For instance, if you use a training file composed of, say 10 technical indicators, their values can be placed anywhere in a 10-dim hyperspace. It is virtually impossible to visualize nor analyze their position in that high dimension space. WinSOM will clarify that picture by projecting the probability density function onto a 2D grid, and in doing so will cluster 10-dim vectors (each made of those 10 technical indicators). Those clusters are placed on the XY grid, and X and Y values are calculated. As a matter of fact, WinSOM can serve as a fuzzy classifier. In addition, clusters are inter-related. For instance, the cluster (2,2) in a 5x6 grid, is likely to be close to the cluster (1,2), (2,1), (2,3), and (3,2), and populated with indicator vectors which are similar to its cluster neighbors. As indicators often change continuously from one bar to the next, one is likely to see the bar indicators (10-dim vector in our case), possibly change from one cluster to one of its neighbors. We have therefore built a fuzzy pattern, which will be more easily picked in your trading strategy optimizer.
I may have lost you a little now, but don't panic: using WinSOM is easier than it looks, samples are provided, and a PowerPoint walk-through is also available on our download page. Theses samples will help you get WinSOM off the ground, and you'll soon be able to use it with your own data files.
For more information on SOM, please check our links page. There are hundreds of sites on the Internet about Kohonen SOM. We have posted two Adobe Acrobat PDF documents for your perusal: Neural Nets and Kohonen
WinSOM uses text data files (PAT extension by default). For compatibility with other operating systems, files must have been saved using ANSI coding. Microsoft NotePad in Windows does that.
WinSOM is supplied on a "as is" basis. Please contact us for optional email support or product customization.
To download WinSOM, please click here.
Click here to read the Getting Started document (<30KB RTF document readable from your browser)
to download or view the Powerpoint Presentation (<600KB) It requires
either Microsoft PowerPoint or at least the free PowerPoint Viewer.
The current download V1.12 is up-to-date (March 2003). Source code available (Borland C++ Builder 5)
Page last modified:
May 08, 2008