Currently argmin and argmax (and their nan equivalents) implement the axis=None case by flattening the array first. This is potentially very expensive since it does a full rechunk. It should be possible to use block_id and offset calculations to figure out the array index of the min/max element.
Currently argmin and argmax (and their nan equivalents) implement the
axis=Nonecase by flattening the array first. This is potentially very expensive since it does a full rechunk. It should be possible to useblock_idand offset calculations to figure out the array index of the min/max element.