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
ND-arrays representing spatial data may have highly related data located spatially in an n-dimensional sense. It's possible that pre-ordering the buffer with some sort of space-filling curve would improve compression.
I think there is overlap with zarr-developers/zarr-python#40 where @shoyer proposed re-ordering each chunk's buffer using the Z-order curve, except here the goal is to improve compression and in zarr-developers/zarr-python#40 the goal is to reduce overhead associated with accessing small data regions.
Activity
alimanfoo commentedon Nov 7, 2016
I think there is overlap with zarr-developers/zarr-python#40 where @shoyer proposed re-ordering each chunk's buffer using the Z-order curve, except here the goal is to improve compression and in zarr-developers/zarr-python#40 the goal is to reduce overhead associated with accessing small data regions.
Looks like there was some discussion of adding support for Z-ordering (Morton ordering) to numpy a while back, but looks like it didn't go anywhere. There is some code by @tgamblin.
cc @benjeffery
dstansby commentedon Dec 12, 2024
I'm going to transfer this to numcodecs, because I think this is more of a request for a new filter or codec. If there's interest in adding this to the zarr v3 spec, then https://github.com/zarr-developers/zarr-specs is the place to propose/discuss.
d-v-b commentedon Sep 24, 2026
various subchunk write orders are now supported by the sharding codec. closing as completed.