A Java applet demonstrating the piecewise
linear classification algorithm of Tenmoto
et al. should appear below. If it does not, Java is most likely
disabled in your browser. Despite the platform independence of Java, this
applet is known to be buggy on some versions on Linux. Details on the interesting
points of the algorithm implementation can be found here.
How It Works
The applet is designed to show the construction of the hyperplanes used
to create piecewise
linear classifier. Enter some prototype
points from several different classes by clicking on the left window with
your mouse. Pressing "Links" adds random data points associated with each
and identifies the Tomek links.
Pressing "Next Hyperplane" identifies the next hyperplane. The left window
displays each hyperplane individually along with the Tomek links it cuts.
This is enough to get you started, for a more detailed description of the
settings click here.
Max Local Error: Use the scroll bar to set the maximum acceptable
classification error rate when constructing hyperplanes. This ranges from
0.0 to 0.5. A higher acceptable error results in more hyperplanes required
for classification. A lower acceptable error results in fewer hyperplanes
required for classification.
Points Per Prototype: Use the scroll bar to set the number of
random points assigned to each prototype. Warning: the more
points, the slower the identification of the hyperplanes!
This web page prepared by
Matt Toews (email@example.com)
as a term project for the course
Science 644 - Pattern Recognition
McGill Center for