Frequently Asked Questions¶
Roadmap¶
Should this be merged into pandas
one day?¶
Yes, we definitely want to have parts of fletcher
as part of pandas
.
There are plans for a native string type and a list type where Apache Arrow would be the preferable data structure.
If this is merged into pandas
, is there still a need for fletcher
?¶
Definitely! fletcher
functions as glue for all available types in Apache Arrow as pandas.ExtensionDtype
.
Only a subset of these types will make its way into pandas
, thus the string functions that return a boolean result in pandas
will return a pandas.BooleanArray
(bytemask-backed numpy
array) whereas the ones in fletcher
will keep returning a fletcher.FletcherBaseArray
(bitmask-backed Arrow array).
Furthermore, an additional goal of fletcher
is to support working with numba
on top of pyarrow.Array
structures.
These utilities will not make their way into pandas
as there we don’t wont to introduce a hard numba
dependency.