Empirical timings with perf_counter

Beginner Asymptotics & empirical timing

Created by Pavel · 29.04.2026 at 19:08 UTC

time.perf_counter() returns a float timestamp from a monotonic clock suitable for deltas: subtract end from start to measure elapsed seconds. It is what you want for micro-benchmarks of NumPy calls, sklearn fit, or small pure-Python kernels—not time.time() (NTP adjustments can move backwards) and usually not time.process_time() alone if you care about wall-clock waits.

Empirical measurement is noisy: cold caches, JIT warm-up, other processes, and Python’s allocator all jitter results. Discard the first iteration, repeat, and report median or mean over several runs when comparing approaches.

For coursework, pairing asymptotic reasoning with a few timed curves (doubling n) builds intuition that Big-O alone cannot supply.

time.perf_counter docs: [1].

Sources

[1]https://docs.python.org/3/library/time.html#time.perf_counter Return to text

University approvals: 0

Tasks

Question 1

time.perf_counter() is primarily intended for:

Hint

Subtract two readings.

Reading the Unix epoch in UTC

Measuring elapsed intervals with a monotonic performance counter

Sleeping the current thread

Querying SQL transaction timestamps

Question 2

You time a function twice: first call after process start, second call after 1000 identical warm-up calls. The first call is often slower mainly because:

Hint

First execution pays setup costs.

perf_counter is inaccurate on Windows

Cold caches, import side effects, and allocator warmup dominate once

Python forbids timing the first invocation

The GIL doubles first-call latency

Question 3

Implement timing_ms(run_once) that calls run_once() once and returns elapsed wall time in milliseconds as a float, using time.perf_counter().

Hint

Delta in seconds, then multiply by 1000.

import time


def timing_ms(run_once):
    pass

Starter code is prefilled; replace TODO blocks with your solution.

Runtime output (stdout/stderr)

2 test cases will be used for grading

Run checks runtime behavior only. Final correctness is evaluated when you submit.

Card Info

Topic: Asymptotics & empirical timing
Difficulty: Beginner
Completed: 0 users

Creator

Pavel