Check non-numeric columns with dataframe_regression

[DrGFreeman]

It would be very useful if the dataframe_regression fixture could check non-numeric columns in dataframes.

One simple work around is to use data_regression and convert the dataframe to a dictionary:

data_regression.check(df.to_dict("records"))

However, this does not allow application of tolerances to numerical values.

As a workaround, I am currently defining a fixture in my conftest.py that leverages data_regression for the non-numeric columns and dataframe_regression for the numeric columns (with tolerances):

# conftest.py

@pytest.fixture()
def check_df(dataframe_regression, data_regression):
    """Fixture to check dataframe against expected values leveraging pytest-regression
    dataframe_regression and data_regression fixtures. This fixture allows verification
    of non-numeric columns as well as application of tolerances on numeric columns."""

    def check(df, basename=None, fullpath=None, tolerances=None, default_tolerance=None):
        data_regression.check(
            df.select_dtypes(exclude="number").to_dict("records"),
            basename=basename,
            fullpath=fullpath,
        )
        dataframe_regression.check(
            df.select_dtypes(include="number"),
            basename=basename,
            fullpath=fullpath,
            tolerances=tolerances,
            default_tolerance=default_tolerance,
        )

    yield check

# test_something.py

def test_something(check_df):
    df = some_operation()

    check_df(df, default_tolerance=dict(atol=1e-8, rtol=1e-5)

While this works, it is less elegant and requires to be run twice to generate the yaml and csv files of expected results.

[nicoddemus]

Hi @DrGFreeman,

Indeed this would be a nice feature. Right now there's no plans to implement this, but we would be glad to review a PR adding this feature. 👍