FAQ

If your question or problem is not in this list, please send me a mail.

Q: Which browser should I use?
A: Any browser is fine as long as it has a java plug-in installed that has a jre version 1.4.1_07 or later. If your java plug-in is older, it is recommended to install a newer version. Newer versions are backward compatible and your older applications will still work. Newer java runtime environments are usually faster too.
Q: I get an 'Applet notinited' error!
A: It is likely that you have an old jre version running. Refer to the answer above.
Q: The applet doesn't load at all!
A: Your browser doesn't support java. Try using another browser that is java enabled or install a recent java plug-in into your browser.
Q: The comboboxes (drop down options) don't work when I want to select another data set.
A: I don't know why that is the case, but you can use arrows and press ENTER to make your selection. The rest should work fine.
Q: The backpropagation algorithm does not seem to converge even after 10,000 iterations... I keep getting 33%, 50% or 66% accuracy.
A: It is possible that the algorithm is 'stuck' in a local minimum that doesn't produce a very good solution. What happens is that the gradient is very small in that particular point of the error surface. This means that the error does not decrease because whichever way the algorithm traverses the error space, the error doesn't get any smaller anymore (only slightly larger) and just 'surfs' around that area where the error is minimal. The best thing to do is to restart the experiment and because the weights are initialized at random it is likely a different outcome will result this time. You can use the 'fast training' option to speed up training.
Q: Why does the Perceptron algorithm not converge for the XOR data set?
A: Aha! Read the book :-). (and look at the demo carefully)

References:
Machine Learning, Tom Mitchell, McGraw Hill, 1997
(1) Section 4.4
(2) Section 4.5 and 4.6

If you have any questions or comments, let me know:
freeisms

gmail.com
It would help improve these demos and the FAQ!