Testing and validation

Verify that computed floating-point results match expected values within acceptable error bounds.

Testing: unit testing numerical algorithms; Validation: validating outputs of simulations or financial calculations; Scientific computing: comparing results of floating-point operations

Scientific computing, data science, finance, game development, any Python code that performs floating-point arithmetic.

assert math.isclose(actual, expected, rel_tol=tolerance)

assert math.isclose(..., ..., rel_tol=...)

actual: numeric expression (float or int), expected: numeric expression, tolerance: positive float (e.g., 1e-9)

• unittest.TestCase
• pytest
• numpy.testing.assert_allclose
• math.isclose
• assert statements

• using abs_tol instead of rel_tol: better for values near zero
• using both rel_tol and abs_tol: covers both small and large values
• using numpy.isclose: for array comparisons
• using pytest.approx: richer syntax in pytest

forgetting to import math, using == for float comparison, setting tolerance too tight or too loose, omitting the assert keyword

pytest.approx (more expressive), numpy.testing.assert_allclose (array support), manual epsilon check (abs(a-b) < eps)

Coming from C or Java: expecting exact equality with == for floats; may overlook rounding errors.

• Floating-point comparison
• Assertion
• Numeric tolerance testing

Relative tolerance scales with magnitude; for values near zero consider adding an absolute tolerance (abs_tol) to avoid always passing.

Makes floating-point equality intent explicit and prevents false test failures due to rounding.

Assert they're close enough

math.isclose was added in Python 3.5; earlier versions require manual epsilon checks. The default rel_tol is 1e-9, which is sufficient for most scientific computing but may need loosening for iterative algorithms with accumulated error.

Use pytest.approx for richer assertion syntax in test frameworks.