GuideAdvanced

Python Performance and Profiling

Intuition about what makes a program slow is almost always wrong, and optimizing without measuring is the most common way to waste time fixing the part that was never the problem. This guide teaches the complete optimization process under the one rule that matters: measure first, change second. You work through a single realistic case — Nómada, a command-line tool that processes trip logs for a vehicle fleet and takes four minutes to generate a monthly report — and take it down to a few seconds using the real method: timing with the correct clock, profiling with `cProfile` to find where the time actually goes (almost never where the team thinks), measuring memory cost, choosing the right data structure for the problem, and applying the specific optimization that fits each real bottleneck. The guide also teaches you to recognize when pure Python no longer cuts it and work needs to be pushed to vectorized code, and how to document and protect an optimization so it doesn't get lost in the next refactor. The final project requires an honest report of what changed, how much each change gained, and what was deliberately NOT optimized and why.