Skip to content

numpy/almost-equal-assert

Rule almost-equal-assert
Pack NumPy
Severity info
Fix unsafe, with xpd check --fix --unsafe-fixes
Library versions every version

assert_almost_equal and its kin are on NumPy's not recommended list, and their default test hides an absolute tolerance of 1.5e-7. Use assert_allclose with the tolerance written out.

np.testing.assert_array_almost_equal(result, expected, decimal=4)
np.testing.assert_allclose(result, expected, rtol=0, atol=1.5e-4)

The rule reports assert_almost_equal, assert_array_almost_equal and assert_approx_equal, which NumPy lists under "Asserts (not recommended)".

xpd check --fix --unsafe-fixes applies the fixes of this rule. Check the result, because an unsafe fix can change what the code does.

The unsafe fix writes np.testing.assert_allclose(a, b, rtol=0, atol=1.5e-7), with 1.5e-6 for assert_array_almost_equal and 1.5e-<decimal> for a literal decimal. A difference of exactly atol failed before and passes after, dicts and ragged lists are no longer compared item by item, and complex values are compared by magnitude. A bare name that the file imports from numpy.testing gets assert_allclose(a, b, rtol=0, atol=1.5e-7) in a file that also imports assert_allclose from numpy.testing. assert_approx_equal(a, b) gets assert_allclose(a, b, rtol=1e-6), which can fail where a difference up to ten times rtol passed. Any other bare name, an err_msg argument, a computed decimal and significant= have no fix.

xpd prints this message with the issue. A name that starts with $ stands for code from your file.

assert_almost_equal is on NumPy's not recommended list, and its default test is an absolute tolerance of 1.5e-7. Write $M.assert_allclose($A, $B, rtol=0, atol=1.5e-7).

Browse the other rules of the NumPy pack.