GitHub - scoder/quicktions: Fast fractions reimplementation in Cython (original) (raw)
quicktions
Python's Fraction data type is an excellent way to do exact calculations with unlimited rational numbers and largely beats Decimal in terms of simplicity, accuracy and safety. Clearly not in terms of speed, though, given the cdecimal accelerator in Python 3.3+.
quicktions is an adaptation of the original fractions module (as included in CPython 3.13) that is compiled and optimised withCython into a fast, native extension module.
Compared to the standard library fractions module of CPython 3.12, computations in quicktions are about 2-4x faster.
Instantiation of a Fraction in quicktions is also
- 5-15x faster from a floating point string value (e.g.
Fraction("123.456789")) - 3-5x faster from a floating point value (e.g.
Fraction(123.456789)) - 2-5x faster from an integer numerator-denominator pair (e.g.
Fraction(123, 456))
We provide a set of micro-benchmarks here:
https://github.com/scoder/quicktions/tree/master/benchmark
As of quicktions 1.19, the different number types and implementations compare as follows in CPython 3.12 (measured on Ubuntu Linux):
Average times for all 'create' benchmarks: float : 19.69 us (1.0x) Fraction : 58.63 us (3.0x) Decimal : 84.32 us (4.3x) PyFraction : 208.20 us (10.6x)
Average times for all 'compute' benchmarks: float : 1.79 us (1.0x) Decimal : 10.11 us (5.7x) Fraction : 39.24 us (22.0x) PyFraction : 96.23 us (53.9x)
While not as fast as the C implemented decimal module in Python 3,quicktions is about 15x faster than the Python implemented decimalmodule in Python 2.7.
For documentation, see the Python standard library's fractions module: