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
Following this issue #11161 and related PR, can we support no-copy pickling for recarray, which is a subclass of ndarray? I think right now we don't support this feature, since recarray is non-contiguous arrays.
If it's possible in your experienced view, I guess we need to do some changes in numpy/core/src/multiarray/methods.c and related tests.
Yes, you can probably do that, not sure what the rules are around non-contiguous buffers for pickling. But your examples are contiguous, the problem is just that it is an array subclass and that doesn't go down the buffer pickling path. Not sure there is actually a good reason for that.
EDIT: I.e. if you use a normal array with structured dtypes and not a recarray and it should work fine.
Uh oh!
There was an error while loading. Please reload this page.
Proposed new feature or change:
Following this issue #11161 and related PR, can we support no-copy pickling for recarray, which is a subclass of ndarray? I think right now we don't support this feature, since recarray is non-contiguous arrays.
If it's possible in your experienced view, I guess we need to do some changes in numpy/core/src/multiarray/methods.c and related tests.
Example is here: ray-project/ray#37573 (comment). I can make up a version without Ray later.
And I saw a demo for how to handle pd DataFrame #12011 (comment), don't know if we can have a local trick for recarray too.
The text was updated successfully, but these errors were encountered: