Single-neuron trainable classifier. Click the grid to add points, then train.
Tap or click inside the square to place a point. Choose the class with the selector below the canvas. The perceptron learns weights w1, w2 and bias b from your data.
The neuron computes z = w1·x + w2·y + b and predicts +1 if z ≥ 0, otherwise -1. For every misclassified point, weights update by: w += rate · target · (x, y) and b += rate · target. Training stops when all points are correct or the boundary keeps oscillating.