rok commented on issue #16264: URL: https://github.com/apache/arrow/issues/16264#issuecomment-2310097455
@sjperkins looking at the past discussion I agree extension type is the way to go. I'm not sure about the storage type though. It would be nice to have zero-copy path to numpy which, as per [this](https://github.com/apache/arrow/issues/39753#issuecomment-1908608726), would need fixed_size_list approach. That said, wouldn't it be possible to interpret Fixed/VariableShapeTensor as a complex tensor if an extra dimension of size 2 was added (and strides were done correctly, namely the complex dimension had the smallest stride)? I think the memory layout in this case would match numpy's. FixedShapeTensor can be zero-copied to numpy today and can probably be cast to complex in numpy. It would of course be better to add a new extension type to not make users do this manually. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
