A 2D perceptron algorithm

This is an implementation of the perceptron algorithm, which is a simple form of learning algorithm. The applet's not very sophisticated - it starts off by picking a random line across the canvas, generates points randomly and labels them (cross or box) according to which side of the line they fall on. It then completely forgets about the original line, and the learning algorithm comes into play - effectively, it uses the sample points as a training set to try and work out what the original line was.

• Click 'Restart' to restart the algorithm with the same dataset
• Click 'Regenerate training set' to generate a new random dataset and restart
• Adjusting the value of η (eta) will control how quickly the perceptron learns. Usually a greater value will cause it to find a solution more quickly, but may sometimes cause it to 'miss' the solution in difficult cases: if it seems to be taking a long time to converge, try a smaller value (e.g. 10x smaller).

If you're interested, you can download the Java source code and Javadocs.