Extension to hypothesis for testing numpy general universal functions
.. image:: https://api.travis-ci.com/uber/hypothesis-gufunc.png?token=RSemjpisB7uiZv78DVwd&branch=master :target: https://travis-ci.com/uber/hypothesis-gufunc
This project is experimental and the APIs are not considered stable.
Only Python>=3.6
is officially supported, but older versions of Python likely work as well.
This package includes support for strategies which generate arguments to
functions that follow the numpy general universal function API. So, it can
automatically generate the matrices with shapes that follow the shape
constraints. For example, to generate test inputs for np.dot
, one can use,
.. code-block:: python
import numpy as np from hypothesis import given from hypothesis.strategies import floats from hypothesis_gufunc.gufunc import gufunc_args
easy_floats = floats(min_value=-10, max_value=10)
@given(gufunc_args('(m,n),(n,p)->(m,p)', dtype=np.float_, elements=easy_floats)) def test_np_dot(args): x, y = args assert np.allclose(np.dot(x, y), np.dot(y.T, x.T).T)
test_np_dot() # Run the test
We also allow for adding extra dimensions that follow the numpy broadcasting conventions. This allows one to test that the broadcasting follows the vectorization conventions:
.. code-block:: python
@given(gufunc_args('(m,n),(n,p)->(m,p)', dtype=np.float_, elements=easy_floats, max_dims_extra=3)) def test_np_matmul(args): x, y = args f_vec = np.vectorize(np.matmul, signature='(m,n),(n,p)->(m,p)', otypes=[np.float_]) assert np.allclose(np.matmul(x, y), f_vec(x, y))
test_np_matmul() # Run the test
Providing max_dims_extra=3
gives up to 3 broadcast compatible dimensions on each of the arguments.
Simply install with pip:
.. code-block::
pip install hypothesis-gufunc
If one would like the same pinned requirements we use during testing, then one can install from the repo with:
.. code-block::
git clone [email protected]:uber/hypothesis-gufunc.git cd hypothesis-gufunc pip install -r requirements/base.txt pip install -e .
The tests for this package can be run by first doing a cd
to its root directory and then
.. code-block::
./test.sh
The script creates a conda environment using the requirements found in requirements/test.txt
.
This package also contains an extension to hypothesis to generate xarray data structures.
To install the package with the xarray dependencies install it with pip as
.. code-block::
pip install hypothesis-gufunc[xarray]
Once installed, one can generate a data array as follows:
.. code-block:: python
from hypothesis.strategies import integers, lists from hypothesis_gufunc.extra.xr import fixed_dataarrays
S = fixed_dataarrays(("a", "b"), coords_st={"a": lists(integers(0, 3))}) S.example()
Here, coords_st
allows one to specify a custom strategy for the coordinates on a per-dimension basis. Likewise, if
one has known coordinates one can call fixed_coords_dataarrays
; or
dataarrays
if one wants both the dimensions and coordinates determined by the
strategy.
The package also has the ability to generate a dataset:
.. code-block:: python
from hypothesis_gufunc.extra.xr import fixed_datasets
S = fixed_datasets({5: ("a", "b"), "bar": ("b"), "baz": ()}, coords_st={"a": lists(integers(0, 3))}) S.example()
One can use fixed_coords_datasets
when the coordinates are determined; or simply
datasets
to have both the dimensions and coordinates generated.
The source <https://github.com/uber/hypothesis-gufunc>
_ is hosted on GitHub.
The documentation <https://hypothesis-gufunc.readthedocs.io/en/latest/>
_ is hosted at Read the Docs.
The main hypothesis project <https://hypothesis.readthedocs.io/en/latest/>
_.
A description of the numpy
Generalized Universal Function API <https://docs.scipy.org/doc/numpy/reference/c-api.generalized-ufuncs.html>
_.
Likewise, the numpy broadcasting rules are described
here <https://docs.scipy.org/doc/numpy/user/basics.broadcasting.html>
_.
The xarray <http://xarray.pydata.org/en/stable/index.html>
_ project describes data arrays and datasets.
This project is licensed under the Apache 2 License - see the LICENSE file for details.