Quick Tip: Benchmark Any Python Function with `timeit` the Right Way (CLI One-Liner)

Quick Tip

Don't write start = time.time() boilerplate for micro-benchmarks. The stdlib does it better, from the CLI:

python3 -m timeit -s "import re" "re.findall(r'\d+', 'abc123def456')"
# 5000000 loops, best of 5: 0.421 usec per loop

Compare two approaches directly:

python3 -m timeit "'-'.join(['a','b','c'])"
python3 -m timeit "'a' + '-' + 'b' + '-' + 'c'"

Why it's better than time.time():

• Runs the snippet millions of times — one-off timings are noise (OS scheduling, CPU frequency scaling)
• Disables garbage collection during measurement
• best of 5 reports the minimum, which is the closest to true cost
• -s flag sets up imports once, outside the timed loop

In a script, same API:

import timeit
t = timeit.timeit("sorted(x)", setup="import random; x=[random.random() for _ in range(1000)]", number=1000)
print(f"{t/1000*1e6:.2f} usec per call")

If your two implementations differ by less than 2x after timeit, the "optimization" probably isn't worth the readability cost.

Powered by MonkeyCode (free AI coding assistant): https://ly.cyberserval.tech/iIETXiF

What micro-benchmark surprised you the most when you actually measured it?