80.80. Perceptron

Implement a **Single-Layer Perceptron** classifier, the simplest neural network model.

Algorithm:

1. Initialize weights to zeros and bias to 0

2. For each training sample:

a. Compute linear output: z = w @ x + b

b. Apply step activation: y_hat = 1 if z >= 0 else 0

c. Update rule: w = w + lr * (y - y_hat) * x, b = b + lr * (y - y_hat)

3. Repeat for n_iterations epochs

Example:

p = Perceptron(learning_rate=0.01, n_iterations=1000)

p.fit(X_train, y_train) # y must be 0 or 1

predictions = p.predict(X_test)

**Explanation:** The Perceptron learns a linear decision boundary. It updates weights only when a prediction is wrong. It converges for linearly separable data but cannot solve non-linear problems (like XOR).

Constraints: