Replies: 2 comments
-
Sorry for missing this earlier @DrTodd13 ! The That said, if you still want to experiment with a custom backend for fun/learning, I'll be happy to help :) My initial guess is that some numpy-specific logic is being used somewhere in |
Beta Was this translation helpful? Give feedback.
-
Hi @rjzamora, I'm currently working on a prototype for an array backend. I was wondering if it's feasible to create a backend for array implementation where the data type is not treated as an alias of |
Beta Was this translation helpful? Give feedback.
-
I have been experimenting with ArrayBackendEntrypoint to provide an alternative backing store instead of NumPy. I have arrays A and B that I create directly through dask and internally those seem to use my configured backing store. I then do:
B += da.overlap.overlap(A, depth=r, boundary='none').map_blocks(lambda x: stencil(x,W,pattern,r),chunks=csB)
Inside the stencil function, the type of A is numpy.ndarray instead of my alternate backing store like I expected. Can someone help me understand what is going on in this process to I presume at some point to have converted from my alternate backing store back to NumPy?
@rjzamora
Beta Was this translation helpful? Give feedback.
All reactions