Performance Optimization

Introduction

Profile first, optimize second. Python's tools: cProfile, timeit, py-spy. Speed wins: vectorize, cache, drop to C/Rust extensions.

Understanding the topic

Core concepts to understand:

• cProfile — full call profile.
• py-spy — sampling, no-overhead, prod-safe.
• Vectorize with NumPy/Pandas.
• @lru_cache for pure functions.
• Cython / Rust (PyO3) for hot loops.

Syntax reference

Visual flow / code:

# cProfile
python -m cProfile -s tottime main.py

# Timeit a snippet
python -m timeit -n 1000 "sum(range(1000))"

# py-spy on a running process
py-spy record -o flame.svg --pid 1234

# Memoization
from functools import lru_cache

@lru_cache
def fib(n):
    return n if n < 2 else fib(n-1) + fib(n-2)

Execution workflow

Real-world use

When pure Python isn't fast enough: Cython (typed Python), Numba (JIT for NumPy), or Rust extensions via maturin + PyO3.

Best practices

• Profile before optimizing.
• Pick the right data structure first.
• Vectorize and cache before going native.

Common mistakes

• Premature optimization — measure first.

Hands-on exercise

Interview preparation — practice these questions:

• How do you profile Python?
• Vectorization — why fast?
• When use Cython/Rust?

Summary

In summary: Measure → optimize → re-measure. Vectorize, cache, then go native.