python-performance

Run python -m pyinstrument script.py for readable statistical output, or python -m cProfile -s cumulative script.py for detailed built-in timing. PyInstrument is preferred for quick identification of slow functions.

PyInstrument is a statistical profiler with lower overhead and more readable output, ideal for quick analysis. cProfile is deterministic and built into Python, providing detailed call counts and cumulative timing for deeper investigation.

Install memray and run memray run script.py, then generate a flamegraph with memray flamegraph to visualize allocations. Alternatively, use the built-in tracemalloc module to snapshot and compare memory usage by line number.

Use pytest-benchmark by adding a benchmark fixture to your test and calling benchmark(func, *args). Run pytest with --benchmark-only to execute benchmarks, and --benchmark-compare to detect regressions across runs.

Use sets for O(1) membership checks, deque for queue operations, generators for large data streams, functools.lru_cache for expensive computations, and str.join for string building. Always profile first and improve algorithms before micro-optimizations.