#!/usr/bin/env python3
# Source: q1-ml-fundamentals.html
import numpy as np

def sigmoid(z):
    # stable form: avoid overflow in exp for large |z|
    z = np.asarray(z, dtype=float)
    e = np.exp(-np.abs(z))
    return np.where(z >= 0, 1 / (1 + e), e / (1 + e))

def fit_logreg(X, y, lr=0.1, epochs=1000, l2=0.0):
    """Batch gradient descent on L2-regularized log loss."""
    n, d = X.shape
    w, b = np.zeros(d), 0.0
    for _ in range(epochs):
        p = sigmoid(X @ w + b)
        err = p - y                          # dL/dz for every row
        w -= lr * (X.T @ err / n + l2 * w)   # do not regularize b
        b -= lr * err.mean()
    return w, b

def log_loss(y, z):
    """Log loss from logits, stable: log(1 + e^z) - y*z."""
    return np.mean(np.logaddexp(0, z) - y * z)
X = np.array([[-2.0], [-1.0], [1.0], [2.0]])
y = np.array([0, 0, 1, 1])
w, b = fit_logreg(X, y, l2=0.1)
print((sigmoid(X @ w + b) >= 0.5).astype(int))  # [0 0 1 1]
