rr.approx

Approximate floating point arithmetic library. This simple module can be used to compare numbers using a relative and absolute tolerance, to mitigate (not eliminate!) floating point rounding errors.

from rr.approx import Approx

x = Approx(0.1) * 3
x == 0.3  # True

The Approx class is very simple to use as a replacement for "regular" floats -- you can use Approx objects instead of floats in most (if not all) contexts: arithmetic and comparisons.

The Approx.context() context manager allows temporary modification of relative and absolute tolerance parameters.

print(Approx.rtol, Approx.atol)  # display current parameters
print(Approx(0.1) * 3 == 0.3)  # make a test comparison
with Approx.context(rtol=0, atol=0):  # temporary modification
    print(Approx.rtol, Approx.atol)  # show modified tolerances
    print(Approx(0.1) * 3 == 0.3)  # rerun the comparison
print(Approx.rtol, Approx.atol)  # back to original parameters

Disclaimer

Please note that this module does not solve any problems with floating point numbers. What it does is provide a little (or big, depending on how you configure your tolerance parameters) buffer to compensate for rounding errors, but as these errors accumulate you can always get unexpected results.

Float rounding is inherently inaccurate in all computers and programming languages due to representing a possibly recurring decimal in a finite number of digits. If you really care about exact results and don't mind paying a performance penalty, you should check out the decimal module from the standard library or some other alternative.

Compatibility

Developed and tested in Python 3.6+. The code may or may not work under earlier versions of Python 3 (perhaps back to 3.3).

Installation

From the github repo:

pip install git+https://github.com/2xR/rr.approx.git

License

This library is released as open source under the MIT License.

Copyright (c) 2017 Rui Rei