You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Ironically, IndexedMaskedArray (from FillableArray) and BitMaskedArray (from Arrow/Parquet) are more important than ByteMaskedArray. The ByteMaskedArray might actually lag behind and not get implemented for a while.
I should do ByteMaskedArray before BitMaskedArray, as preparation. BitMaskedArray is just like it but with the additional complication of having to deal with bits. Probably just generate bytes when needed, leaving direct bit-manipulation as a later optimization.
As soon as ByteMaskedArray is ready, reducers should get a semigroup parameter to turn placeholder-identities into None (expected for min/max, and probably necessary before #70).
At that time, also push None values in a reduction down to just above the primitive level with ByteMaskedArray, and use that to propagate Nones in the result. But it will be much easier with a ByteMaskedArray than an IndexedOptionArray.
Ironically,
IndexedMaskedArray
(fromFillableArray
) andBitMaskedArray
(from Arrow/Parquet) are more important thanByteMaskedArray
. TheByteMaskedArray
might actually lag behind and not get implemented for a while.Might need to change the BitMaskedArray wiki page.
The text was updated successfully, but these errors were encountered: