Import Tree from dask-awkward if not in dask

[martindurant]

Following the disappearance of DataframeTreeReduction in upstream dask, it was copied to dask-awkward, making it a required dependency in the future.

cc @lgray

[martindurant]

Appears to fail on some dataframe-specific functions:

FAILED tests/test_boost.py::test_histogramdd_series - NotImplementedError: The legacy implementation is no longer supported
FAILED tests/test_boost.py::test_histogramdd_arrays_and_series - NotImplementedError: The legacy implementation is no longer supported
FAILED tests/test_boost.py::test_histogramdd_dataframe - NotImplementedError: The legacy implementation is no longer supported
FAILED tests/test_core.py::test_df_input[True] - NotImplementedError: The legacy implementation is no longer supported
FAILED tests/test_core.py::test_df_input[None] - NotImplementedError: The legacy implementation is no longer supported

I didn't even realise we supported this. I suppose we drop support for now, as with dak.to_dataframe?

[martindurant]

I haven't yet figured out why dask thinks we are trying to use pre-expr, there must be a config set somewhere.

Locally, I am still seeing some failures, but the following without dask is mystifying me:

>>> x = np.random.standard_normal(size=(3_000,))
>>> h2 = bh.Histogram(bh.axis.Regular(10, -3, 3))
>>> h2.fill(x)
>>> h2.to_numpy(dd=True, flow=True)
(array([  5.,  17.,  79., 246., 456., 711., 666., 487., 241.,  70.,  20.,
          2.]),
 [array([1.79769313e+308, 1.79769313e+308, 1.79769313e+308, 1.79769313e+308,
         1.79769313e+308, 1.79769313e+308, 1.79769313e+308, 1.79769313e+308,
         1.79769313e+308, 1.79769313e+308, 1.79769313e+308, 1.79769313e+308,
         1.79769313e+308])])

why are the edges messed up?? The values look right. Is there some copy thing with numpy 2 I should care about?

^ this only happens for flow=True

[pfackeldey]

I haven't yet figured out why dask thinks we are trying to use pre-expr, there must be a config set somewhere.

Locally, I am still seeing some failures, but the following without dask is mystifying me:

>>> x = np.random.standard_normal(size=(3_000,))
>>> h2 = bh.Histogram(bh.axis.Regular(10, -3, 3))
>>> h2.fill(x)
>>> h2.to_numpy(dd=True, flow=True)
(array([  5.,  17.,  79., 246., 456., 711., 666., 487., 241.,  70.,  20.,
          2.]),
 [array([1.79769313e+308, 1.79769313e+308, 1.79769313e+308, 1.79769313e+308,
         1.79769313e+308, 1.79769313e+308, 1.79769313e+308, 1.79769313e+308,
         1.79769313e+308, 1.79769313e+308, 1.79769313e+308, 1.79769313e+308,
         1.79769313e+308])])

why are the edges messed up?? The values look right. Is there some copy thing with numpy 2 I should care about?

^ this only happens for flow=True

I can not reproduce this boost-histogram (v1.5.0) issue with either numpy v2.0.0 or numpy v2.2.2 (latest). Maybe @henryiii has seen this behavior in the past?

[martindurant]

Updating boost-histogram to v1.5.0 fixed this.