Lesson 23 — Gradient descent with early stopping

Train until good enough: early stopping

A training loop rarely runs a fixed number of steps. It watches the loss and stops early once the loss is small enough (or stops improving). This saves time and is how real training scripts work. Python's break leaves a loop immediately.

for i in range(500):
    ...
    if error_now < 0.01:
        break

Why it matters: training big models costs real money per step. Knowing when to stop is part of the job.