abs(a - b) <= (atol + rtol * abs(b))
vs
(abs(a - b) <= atol
or
abs(a - b) / max(abs(a), abs(b)) < rtol)
Docs: numpy.allclose, mpmath.almosteq -- Source: numpy, mpmath
How do these schemes differ? An obvious one is that allclose is and while almosteq is or with respect to the tolerances. To me the and makes more sense, yet mpmath is a library dedicated to precision, so I find this strange.
When is almosteq preferred over allclose, speed/memory aside? (Optional question, can one's rtol be converted to another's rough equivalent)?