Function test_reduction_method_split_every

dask/dataframe/tests/test_dataframe.py:2620–2676

()

def test_reduction_method_split_every():
    df = pd.Series([1] * 60)
    ddf = dd.from_pandas(df, npartitions=15)

    def chunk(x, constant=0):
        return x.sum() + constant

    def combine(x, constant=0):
        return x.sum() + constant + 1

    def agg(x, constant=0):
        return x.sum() + constant + 2

    f = lambda n: ddf.reduction(
        chunk,
        aggregate=agg,
        combine=combine,
        chunk_kwargs=dict(constant=1.0),
        combine_kwargs=dict(constant=2.0),
        aggregate_kwargs=dict(constant=3.0),
        split_every=n,
    )

    assert_max_deps(f(3), 3)
    assert_max_deps(f(4), 4, False)
    assert_max_deps(f(5), 5)
    assert f(15).dask.keys() == f(ddf.npartitions).dask.keys()

    r3 = f(3)
    r4 = f(4)
    assert r3._name != r4._name

    # Keywords are different for each step
    assert f(3).compute() == 60 + 15 + 7 * (2 + 1) + (3 + 2)
    # Keywords are same for each step
    res = ddf.reduction(
        chunk, aggregate=agg, combine=combine, constant=3.0, split_every=3
    )
    assert res.compute() == 60 + 15 * 3 + 7 * (3 + 1) + (3 + 2)
    # No combine provided, combine is agg
    res = ddf.reduction(chunk, aggregate=agg, constant=3.0, split_every=3)
    assert res.compute() == 60 + 15 * 3 + 8 * (3 + 2)

    # split_every must be >= 2
    with pytest.raises(ValueError):
        f(1)

    # combine_kwargs with no combine provided
    with pytest.raises(ValueError):
        ddf.reduction(
            chunk,
            aggregate=agg,
            split_every=3,
            chunk_kwargs=dict(constant=1.0),
            combine_kwargs=dict(constant=2.0),
            aggregate_kwargs=dict(constant=3.0),
        )