Check if multiple collections have compatible partitions.


*args (Array) – Any number of array collections to check.


Result of the check.

Return type:



Starting with an absolutely compatible comparison:

>>> import dask_awkward as dak
>>> import awkward as ak
>>> concrete = ak.Array([[1, 2, 3], [4], [5, 6], [0, 0, 0, 0]])
>>> lazy = dak.from_awkward(concrete, npartitions=2)
>>> selection = dak.sum(lazy, axis=1) == 0
>>> dak.partition_compatibility(lazy, selection)
<PartitionCompatibility.YES: 0>

The selection doesn’t change the length of the arrays at each partition, so the divisions are known to be conserved for those operations (the sum on axis=1 along with the equality comparison).

In general we have no way of knowing what the resulting divisions will be after a boolean selection, but the total number of partitions will be conserved, so we have to report MAYBE:

>>> selected_lazy = lazy[selection]
>>> dak.partition_compatibility(lazy, lazy_selection)
<PartitionCompatibility.MAYBE: 2>

Due the simple nature of this example we know that after the selection the partitions will not be compatible (because it’s clear only 1 element of the original array will survive the selection, so the divisions will change after that compute). Now we can eagerly compute what the divisions will be on the lazy_selection collection and get a NO result:

>>> lazy_selection.eager_compute_divisions()
>>> dak.partition_compatibility(lazy, lazy_selection)
<PartitionCompatibility.NO: 1>

Remember that Array.eager_compute_divisions() is going to trigger a compute to determine the divisions (to know divisions we need to know the length of each partition)