zcollection.view.View.map_overlap#

View.map_overlap(func, /, *args, depth=1, delayed=True, filters=None, partition_size=None, npartitions=None, selected_variables=None, **kwargs)[source]#

Map a function over the partitions of the view with some overlap.

Parameters:

  • func (PartitionCallable) – The function to apply to every partition of the view.

  • depth (int) – The depth of the overlap between the partitions. Default is 0 (no overlap). If depth is greater than 0, the function is applied on the partition and its neighbors selected by the depth. If func accepts a partition_info as a keyword argument, it will be passed a tuple with the name of the partitioned dimension and the slice allowing getting in the dataset the selected partition without the overlap.

  • *args – The positional arguments to pass to the function.

  • delayed (bool) – Whether to load data in a dask array or in memory.

  • filters (str | Callable[[Dict[str, int]], bool] | None) – The predicate used to filter the partitions to process. To get more information on the predicate, see the documentation of the zcollection.Collection.partitions() method.

  • partition_size (int | None) – The length of each bag partition.

  • npartitions (int | None) – The number of desired bag partitions.

  • selected_variables (Iterable[str] | None) – A list of variables to retain from the view. If None, all variables are loaded. Useful to load only a subset of the view.

  • **kwargs – The keyword arguments to pass to the function.

Returns:

A bag containing the tuple of the partition scheme and the result of the function.

Return type:

Bag

Example

>>> futures = view.map_overlap(
...     lambda x: (x["var1"] + x["var2"]).values,
...     depth=1)
>>> for item in futures:
...     print(item)
[1.0, 2.0, 3.0, 4.0]