reasonable

Reasonable float handling for Hypothesis testing

Hypothesis is an excellent package, however sometimes we want to trade the purist approach towards floating-point numbers for a more pragmatic approach.

reasonable complements hypothesis by limiting the range and precision of generated floats, defaulting to 8d.p.

As a result, reasonable does not generate very small (10^-300) floats, which makes approximate comparisons (of the sort used in unit-tests) much easier to perform.

For example, by replacing floats with the version in reasonable, you can perform approximate comparisons:

from hypothesis import given
from pytest import approx
from reasonable import floats, x

@given(floats(x < 0.0))
def test_minimizer(n):
    result = minimize(my_func, x0=n)
    assert result == approx(0.23456789)

Documentation

reasonable.readthedocs.io

Installation

Requirements: You will need hypothesis installed (obviously) -

pip install hypothesis

then install reasonable with

pip install reasonable

Contributing

All contributions are welcome :)

If you would like to contribute anything, open a PR (issues are turned off):

• To report a bug, open a PR with a unit-test that fails.
• To request an API change/new functionality, open a PR with a unit-test that fails showing your preferred API.
• To submit a fix, open a PR with passing unit-tests and doctests.

Doctests should be minimal and serve as API docs for the most common use cases.

Unit-tests should be exhaustive, and include basic sanity tests as well as generative tests using hypothesis.

How to report a bug

• Create a new python file in the tests folder, make sure the filename starts with test_.
• Create a single function in the file which replicates the bug. Make sure the function name starts with test_.
• Open a PR with the new file and a brief description - the unit-tests should fail.

At that point either the reporter or a fastats developer can commit on that PR to fix the bug.