xarray-regressions 0.0.2

pytest-regressions plugin for xarray

A pytest-regressions plugin for identifying regressions in Xarray objects.

[!WARNING] xarray-regressions is in early development and might have breaking changes.

Install

pip install xarray-regressions

Usage

[!TIP] If you're unfamiliar with pytest-regressions, check out their documentation first.

Once installed, this package registers a test fixture xarray_regression for detecting regressions in the data or metadata of xr.DataArray and xr.Dataset objects.

Say you have a function make_dataarray that should always return the same output:

import xarray as xr

def make_dataarray(name: str) -> xr.DataArray:
    """A dummy method that needs to be tested."""
    return xr.DataArray(
        np.full((2, 4, 3), 1),
        dims=["variable", "y", "x"],
        coords={
            "variable": ["var1", "var2"],
            "y": [1, 2, 3, 4],
            "x": [1, 2, 3],
        },
        name=name,
        attrs={"foo": "bar"},
    )

Add the xarray_regression fixture to a new test and call the check method on the returned data array:

from xarray_regressions import XarrayRegressionFixture # Only used for type annotation

def test_make_dataarray(xarray_regression: XarrayRegressionFixture):
    """Test that the function always returns an identical xr.DataArray."""
    da = make_dataarray(name="test_array")
    xarray_regression.check(
        da,
        check_names=True,
        check_attrs=True,
    )

Running the test once will write da to a local NetCDF[^netcdf], and future test runs will compare da with the stored result. Values, dimensions, and coordinates are checked using xr.testing.assert_equal or xr.testing.assert_allclose (if atol or rtol are given). Names and attributes are checked seperately. Encodings are not currently checked.

If make_dataarray(name="test_array") returns a different result in the future, the test will fail:

AssertionError: DataArray names are different. L: foo, R: test_array

[^netcdf]: Because results are stored in NetCDF, all tested objects must be serializable.