Perceptron learning rule

The problem, from first principles

The perceptron is the simplest learning neuron: it fires (outputs 1) when a weighted sum of its inputs passes zero, and it learns by nudging its weights only when it gets a sample wrong.

From theory to code

Implement perceptron_train(X, y, lr=1.0, epochs=100). Sweep the samples in order each epoch; whenever the prediction 1 if w @ x + b > 0 else 0 is wrong, move the weights and bias towards the correct answer. Stop early after an epoch with no mistakes.

Constraints

• X has shape (n_samples, n_features); y holds labels in {0, 1}.
• Start from w = 0 and b = 0.0; process samples in the given order, updating immediately after each mistake.
• Return (w, b) where w is a float array and b a Python float.
• Do not modify X or y.

Hints

Open one at a time. Each gives away a little more than the last.