10 minutes to flox

GroupBy single variable

import numpy as np
import xarray as xr

from flox.xarray import xarray_reduce

labels = xr.DataArray(
    [1, 2, 3, 1, 2, 3, 0, 0, 0],
    dims="x",
    name="label",
)
labels

<xarray.DataArray 'label' (x: 9)>
array([1, 2, 3, 1, 2, 3, 0, 0, 0])
Dimensions without coordinates: x

xarray.DataArray

'label'

• x: 9

• 1 2 3 1 2 3 0 0 0

  array([1, 2, 3, 1, 2, 3, 0, 0, 0])

• Coordinates: (0)
• Indexes: (0)
• Attributes: (0)

With numpy

da = xr.DataArray(
    np.ones((9,)), dims="x", name="array"
)

Apply the reduction using flox.xarray.xarray_reduce() specifying the reduction operation in func

xarray_reduce(da, labels, func="sum")

<xarray.DataArray 'array' (label: 4)>
array([3., 2., 2., 2.])
Coordinates:
  * label    (label) int64 0 1 2 3

xarray.DataArray

'array'

• label: 4

• 3.0 2.0 2.0 2.0

  array([3., 2., 2., 2.])

• Coordinates: (1)
  • label
    (label)
    int64
    0 1 2 3

    array([0, 1, 2, 3])

• Indexes: (1)
  • label
    PandasIndex

    PandasIndex(Index([0, 1, 2, 3], dtype='int64', name='label'))

• Attributes: (0)

With dask

Let’s first chunk da and labels

da_chunked = da.chunk(x=2)
labels_chunked = labels.chunk(x=3)

Grouping a dask array by a numpy array is unchanged

xarray_reduce(da_chunked, labels, func="sum")

<xarray.DataArray 'array' (label: 4)>
dask.array<groupby_nansum, shape=(4,), dtype=float64, chunksize=(4,), chunktype=numpy.ndarray>
Coordinates:
  * label    (label) int64 0 1 2 3

xarray.DataArray

'array'

• label: 4

• dask.array<chunksize=(4,), meta=np.ndarray>

  Array Chunk Bytes 32 B 32 B Shape (4,) (4,) Dask graph 1 chunks in 6 graph layers Data type float64 numpy.ndarray

• Coordinates: (1)
  • label
    (label)
    int64
    0 1 2 3

    array([0, 1, 2, 3])

• Indexes: (1)
  • label
    PandasIndex

    PandasIndex(Index([0, 1, 2, 3], dtype='int64', name='label'))

• Attributes: (0)

When grouping by a dask array, we need to specify the “expected group labels” on the output so we can construct the result DataArray. Without the expected_groups kwarg, an error is raised

xarray_reduce(da_chunked, labels_chunked, func="sum")

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[6], line 1
----> 1 xarray_reduce(da_chunked, labels_chunked, func="sum")

File ~/checkouts/readthedocs.org/user_builds/flox/checkouts/latest/flox/xarray.py:340, in xarray_reduce(obj, func, expected_groups, isbin, sort, dim, fill_value, dtype, method, engine, keep_attrs, skipna, min_count, reindex, *by, **finalize_kwargs)
    335     if isbin_:
    336         raise ValueError(
    337             f"Please provided bin edges for group variable {idx} "
    338             f"named {group_name} in expected_groups."
    339         )
--> 340     expect1 = _get_expected_groups(b_.data, sort=sort)
    341 else:
    342     expect1 = expect

File ~/checkouts/readthedocs.org/user_builds/flox/checkouts/latest/flox/core.py:105, in _get_expected_groups(by, sort)
    103 def _get_expected_groups(by: T_By, sort: bool) -> T_ExpectIndex:
    104     if is_duck_dask_array(by):
--> 105         raise ValueError("Please provide expected_groups if not grouping by a numpy array.")
    106     flatby = by.reshape(-1)
    107     expected = pd.unique(flatby[~isnull(flatby)])

ValueError: Please provide expected_groups if not grouping by a numpy array.

Now we specify expected_groups:

dask_result = xarray_reduce(
    da_chunked, labels_chunked, func="sum", expected_groups=[0, 1, 2, 3],
)
dask_result

<xarray.DataArray 'array' (label: 4)>
dask.array<groupby_nansum, shape=(4,), dtype=float64, chunksize=(4,), chunktype=numpy.ndarray>
Coordinates:
  * label    (label) int64 0 1 2 3

xarray.DataArray

'array'

• label: 4

• dask.array<chunksize=(4,), meta=np.ndarray>

  Array Chunk Bytes 32 B 32 B Shape (4,) (4,) Dask graph 1 chunks in 9 graph layers Data type float64 numpy.ndarray

• Coordinates: (1)
  • label
    (label)
    int64
    0 1 2 3

    array([0, 1, 2, 3])

• Indexes: (1)
  • label
    PandasIndex

    PandasIndex(Index([0, 1, 2, 3], dtype='int64', name='label'))

• Attributes: (0)

Note that any group labels not present in expected_groups will be ignored. You can also provide expected_groups for the pure numpy GroupBy.

numpy_result = xarray_reduce(
    da, labels, func="sum", expected_groups=[0, 1, 2, 3],
)
numpy_result

<xarray.DataArray 'array' (label: 4)>
array([3., 2., 2., 2.])
Coordinates:
  * label    (label) int64 0 1 2 3

xarray.DataArray

'array'

• label: 4

• 3.0 2.0 2.0 2.0

  array([3., 2., 2., 2.])

• Coordinates: (1)
  • label
    (label)
    int64
    0 1 2 3

    array([0, 1, 2, 3])

• Indexes: (1)
  • label
    PandasIndex

    PandasIndex(Index([0, 1, 2, 3], dtype='int64', name='label'))

• Attributes: (0)

The two are identical:

numpy_result.identical(dask_result)

True

Binning by a single variable

For binning, specify the bin edges in expected_groups using pandas.IntervalIndex:

import pandas as pd

xarray_reduce(
    da,
    labels,
    func="sum",
    expected_groups=pd.IntervalIndex.from_breaks([0.5, 1.5, 2.5, 6]),
)

<xarray.DataArray 'array' (label_bins: 3)>
array([2., 2., 2.])
Coordinates:
  * label_bins  (label_bins) object (0.5, 1.5] (1.5, 2.5] (2.5, 6.0]

xarray.DataArray

'array'

• label_bins: 3

• 2.0 2.0 2.0

  array([2., 2., 2.])

• Coordinates: (1)
  • label_bins
    (label_bins)
    object
    (0.5, 1.5] (1.5, 2.5] (2.5, 6.0]

    array([Interval(0.5, 1.5, closed='right'), Interval(1.5, 2.5, closed='right'),
           Interval(2.5, 6.0, closed='right')], dtype=object)

• Indexes: (1)
  • label_bins
    PandasIndex

    PandasIndex(Index([(0.5, 1.5], (1.5, 2.5], (2.5, 6.0]], dtype='object', name='label_bins'))

• Attributes: (0)

Similarly for dask inputs

xarray_reduce(
    da_chunked,
    labels_chunked,
    func="sum",
    expected_groups=pd.IntervalIndex.from_breaks([0.5, 1.5, 2.5, 6]),
)

<xarray.DataArray 'array' (label_bins: 3)>
dask.array<groupby_nansum, shape=(3,), dtype=float64, chunksize=(3,), chunktype=numpy.ndarray>
Coordinates:
  * label_bins  (label_bins) object (0.5, 1.5] (1.5, 2.5] (2.5, 6.0]

xarray.DataArray

'array'

• label_bins: 3

• dask.array<chunksize=(3,), meta=np.ndarray>

  Array Chunk Bytes 24 B 24 B Shape (3,) (3,) Dask graph 1 chunks in 9 graph layers Data type float64 numpy.ndarray

• Coordinates: (1)
  • label_bins
    (label_bins)
    object
    (0.5, 1.5] (1.5, 2.5] (2.5, 6.0]

    array([Interval(0.5, 1.5, closed='right'), Interval(1.5, 2.5, closed='right'),
           Interval(2.5, 6.0, closed='right')], dtype=object)

• Indexes: (1)
  • label_bins
    PandasIndex

    PandasIndex(Index([(0.5, 1.5], (1.5, 2.5], (2.5, 6.0]], dtype='object', name='label_bins'))

• Attributes: (0)

For more control over the binning (which edge is closed), pass the appropriate kwarg to pandas.IntervalIndex:

xarray_reduce(
    da_chunked,
    labels_chunked,
    func="sum",
    expected_groups=pd.IntervalIndex.from_breaks([0.5, 1.5, 2.5, 6], closed="left"),
)

<xarray.DataArray 'array' (label_bins: 3)>
dask.array<groupby_nansum, shape=(3,), dtype=float64, chunksize=(3,), chunktype=numpy.ndarray>
Coordinates:
  * label_bins  (label_bins) object [0.5, 1.5) [1.5, 2.5) [2.5, 6.0)

xarray.DataArray

'array'

• label_bins: 3

• dask.array<chunksize=(3,), meta=np.ndarray>

  Array Chunk Bytes 24 B 24 B Shape (3,) (3,) Dask graph 1 chunks in 9 graph layers Data type float64 numpy.ndarray

• Coordinates: (1)
  • label_bins
    (label_bins)
    object
    [0.5, 1.5) [1.5, 2.5) [2.5, 6.0)

    array([Interval(0.5, 1.5, closed='left'), Interval(1.5, 2.5, closed='left'),
           Interval(2.5, 6.0, closed='left')], dtype=object)

• Indexes: (1)
  • label_bins
    PandasIndex

    PandasIndex(Index([[0.5, 1.5), [1.5, 2.5), [2.5, 6.0)], dtype='object', name='label_bins'))

• Attributes: (0)

Grouping by multiple variables

arr = np.ones((4, 12))
labels1 = np.array(["a", "a", "c", "c", "c", "b", "b", "c", "c", "b", "b", "f"])
labels2 = np.array([1, 2, 2, 1])

da = xr.DataArray(
    arr, dims=("x", "y"), coords={"labels2": ("x", labels2), "labels1": ("y", labels1)}
)
da

<xarray.DataArray (x: 4, y: 12)>
array([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
Coordinates:
    labels2  (x) int64 1 2 2 1
    labels1  (y) <U1 'a' 'a' 'c' 'c' 'c' 'b' 'b' 'c' 'c' 'b' 'b' 'f'
Dimensions without coordinates: x, y

xarray.DataArray

• x: 4
• y: 12

• 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 ... 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

  array([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
         [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
         [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
         [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])

• Coordinates: (2)
  • labels2
    (x)
    int64
    1 2 2 1

    array([1, 2, 2, 1])

  • labels1
    (y)
    <U1
    'a' 'a' 'c' 'c' ... 'c' 'b' 'b' 'f'

    array(['a', 'a', 'c', 'c', 'c', 'b', 'b', 'c', 'c', 'b', 'b', 'f'],
          dtype='<U1')

• Indexes: (0)
• Attributes: (0)

To group by multiple variables simply pass them as *args:

xarray_reduce(da, "labels1", "labels2", func="sum")

<xarray.DataArray (labels1: 4, labels2: 2)>
array([[ 4.,  4.],
       [ 8.,  8.],
       [10., 10.],
       [ 2.,  2.]])
Coordinates:
  * labels1  (labels1) object 'a' 'b' 'c' 'f'
  * labels2  (labels2) int64 1 2

xarray.DataArray

• labels1: 4
• labels2: 2

• 4.0 4.0 8.0 8.0 10.0 10.0 2.0 2.0

  array([[ 4.,  4.],
         [ 8.,  8.],
         [10., 10.],
         [ 2.,  2.]])

• Coordinates: (2)
  • labels1
    (labels1)
    object
    'a' 'b' 'c' 'f'

    array(['a', 'b', 'c', 'f'], dtype=object)

  • labels2
    (labels2)
    int64
    1 2

    array([1, 2])

• Indexes: (2)
  • labels1
    PandasIndex

    PandasIndex(Index(['a', 'b', 'c', 'f'], dtype='object', name='labels1'))

  • labels2
    PandasIndex

    PandasIndex(Index([1, 2], dtype='int64', name='labels2'))

• Attributes: (0)

Histogramming (Binning by multiple variables)

An unweighted histogram is simply a groupby multiple variables with count.

arr = np.ones((4, 12))
labels1 = np.array(np.linspace(0, 10, 12))
labels2 = np.array([1, 2, 2, 1])

da = xr.DataArray(
    arr, dims=("x", "y"), coords={"labels2": ("x", labels2), "labels1": ("y", labels1)}
)
da

<xarray.DataArray (x: 4, y: 12)>
array([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
       [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])
Coordinates:
    labels2  (x) int64 1 2 2 1
    labels1  (y) float64 0.0 0.9091 1.818 2.727 3.636 ... 7.273 8.182 9.091 10.0
Dimensions without coordinates: x, y

xarray.DataArray

• x: 4
• y: 12

• 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 ... 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

  array([[1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
         [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
         [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.],
         [1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]])

• Coordinates: (2)
  • labels2
    (x)
    int64
    1 2 2 1

    array([1, 2, 2, 1])

  • labels1
    (y)
    float64
    0.0 0.9091 1.818 ... 9.091 10.0

    array([ 0.        ,  0.90909091,  1.81818182,  2.72727273,  3.63636364,
            4.54545455,  5.45454545,  6.36363636,  7.27272727,  8.18181818,
            9.09090909, 10.        ])

• Indexes: (0)
• Attributes: (0)

Specify bins in expected_groups

xarray_reduce(
    da,
    "labels1",
    "labels2",
    func="count",
    expected_groups=(
        pd.IntervalIndex.from_breaks([-0.5, 4.5, 6.5, 8.9]),  # labels1
        pd.IntervalIndex.from_breaks([0.5, 1.5, 1.9]),  # labels2
    ),
)

<xarray.DataArray (labels1_bins: 3, labels2_bins: 2)>
array([[10,  0],
       [ 6,  0],
       [ 4,  0]])
Coordinates:
  * labels1_bins  (labels1_bins) object (-0.5, 4.5] (4.5, 6.5] (6.5, 8.9]
  * labels2_bins  (labels2_bins) object (0.5, 1.5] (1.5, 1.9]

xarray.DataArray

• labels1_bins: 3
• labels2_bins: 2

• 10 0 6 0 4 0

  array([[10,  0],
         [ 6,  0],
         [ 4,  0]])

• Coordinates: (2)
  • labels1_bins
    (labels1_bins)
    object
    (-0.5, 4.5] (4.5, 6.5] (6.5, 8.9]

    array([Interval(-0.5, 4.5, closed='right'), Interval(4.5, 6.5, closed='right'),
           Interval(6.5, 8.9, closed='right')], dtype=object)

  • labels2_bins
    (labels2_bins)
    object
    (0.5, 1.5] (1.5, 1.9]

    array([Interval(0.5, 1.5, closed='right'), Interval(1.5, 1.9, closed='right')],
          dtype=object)

• Indexes: (2)
  • labels1_bins
    PandasIndex

    PandasIndex(Index([(-0.5, 4.5], (4.5, 6.5], (6.5, 8.9]], dtype='object', name='labels1_bins'))

  • labels2_bins
    PandasIndex

    PandasIndex(Index([(0.5, 1.5], (1.5, 1.9]], dtype='object', name='labels2_bins'))

• Attributes: (0)

Resampling

Use the xarray interface i.e. da.resample(time="M").mean().

Optionally pass method="blockwise": da.resample(time="M").mean(method="blockwise")