Testing discipline vs folklore speedups

Beginner Accelerating numerics & developer hygiene

Created by Pavel · 29.04.2026 at 19:11 UTC

Unit tests lock behaviour while you refactor: changing a feature encoder should not silently shift training targets. Tests do not improve Big-O complexity—they improve confidence and iteration speed.

Confusing “we added tests” with “we made the algorithm asymptotically faster” is a category error. Likewise, micro-benchmarks without correctness checks optimise the wrong thing.

pytest is the de facto runner in many DS repos; property-based testing (hypothesis) finds edge cases random unit tests miss.

pytest docs: [1].

Sources

[1]https://docs.pytest.org/en/stable/Return to text

University approvals: 0

Tasks

Question 1

Systematic automated tests primarily:

Hint

Quality gate, not hardware boost.

Raise CPU clock frequencies

Catch regressions and document intended behaviour

Replace Big-O analysis

Eliminate need for type hints

Question 2

A model trains faster after you vectorise a loop, but validation AUC drops. What failed?

Hint

Optimisation needs regression checks.

Big-O analysis

Correctness/equivalence—speed without behavioural checks

Docker only

Type hints

Question 3

add(a, b) returns a + b but raises RuntimeError('no_bool') if either argument is bool (use type is bool).

Hint

Explicit type guard—bool is subclass of int.

def add(a, b):
    pass

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

Runtime output (stdout/stderr)

1 test case will be used for grading

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

Card Info

Topic: Accelerating numerics & developer hygiene
Difficulty: Beginner
Completed: 0 users

Creator

Pavel