Labs / Computer Science
Perceptron
Click anywhere to drop a point of the selected colour, then press Train. The perceptron nudges a straight decision boundary a little every time it meets a point on the wrong side, and misclassified points flash until the line finally separates blue from orange.
acc 100%epochs 0learning
What to try
- Can the perceptron ever separate two classes that overlap, or does the line just keep swinging back and forth?
- What happens to the number of epochs when you crank the learning rate all the way up versus all the way down?
- Add a single blue point deep inside the orange cluster. How much does one stubborn outlier disturb the boundary?
- Arrange your points in a ring, blue inside and orange outside. Why can a single straight line never solve this?
- After it converges, drag the learning rate and add a new point on the wrong side — how quickly does it recover?