python: Greatest common divisor (gcd) for floats, preferably in numpy

Question

I am looking for an efficient way to determine the greatest common divisor of two floats with python. The routine should have the following layout

gcd(a, b, rtol=1e-05, atol=1e-08)
"""
Returns the greatest common divisor of a and b

Parameters
----------
a,b : float
    two floats for gcd
rtol, atol : float, optional
    relative and absolute tolerance

Returns
-------
gcd : float
    Greatest common divisor such that for x in [a,b]:
    np.mod(x,gcd) < rtol*x + atol 

.. _PEP 484:
    https://www.python.org/dev/peps/pep-0484/

"""

Example: gcd of rational and irrational number

The gcd(1., np.pi, rtol=0, atol=1e-5) should return (roughly) 1e-5, as

In [1]: np.mod(np.pi,1e-5)
Out[1]: 2.6535897928590063e-06

In [2]: np.mod(1.,1e-5)
Out[2]: 9.9999999999181978e-06

I would prefer to use a library implementation and not to write it myself. The fractions.gcd function does not seem appropriate to me here, as I do not want to work with fractions and it (obviously) does not have the tolerance parameters.

Answer 1

Seems like you could just modify the code of fractions.gcd to include the tolerances:

def float_gcd(a, b, rtol = 1e-05, atol = 1e-08):
    t = min(abs(a), abs(b))
    while abs(b) > rtol * t + atol:
        a, b = b, a % b
    return a

Answer 2

The following code may be useful. The function to call is float_gdc(a, b).

from math import gcd, log10, pow

def float_scale(x):
    max_digits = 14
    int_part = int(abs(x))
    magnitude = 1 if int_part == 0 else int(log10(int_part)) + 1
    if magnitude >= max_digits:
        return 0
    frac_part = abs(x) - int_part
    multiplier = 10 ** (max_digits - magnitude)
    frac_digits = multiplier + int(multiplier * frac_part + 0.5)
    while frac_digits % 10 == 0:
        frac_digits /= 10
    return int(log10(frac_digits))


def float_gcd(a, b):
    sc = float_scale(a)
    sc_b = float_scale(b)
    sc = sc_b if sc_b > sc else sc
    fac = pow(10, sc)

    a = int(round(a*fac))
    b = int(round(b*fac))

    return round(gcd(a, b)/fac, sc)

A part of the code is taken from here: Determine precision and scale of particular number in Python