Implement np.tile #5785
Implement np.tile #5785
Conversation
Add basic support for np.tile, limited to when the first input (the tiles) have at most 1 dimension. Includes documentation and tests. Could easily scale to arbitrary-dimensional inputs if the tuple constructor were supported (so the output shape could be computed at runtime). Fixes numba#2327 Fixes numba#3456
|
Thanks for the patch. There's a few tricks to do tuple generation. Do you want to perhaps get this restricted version in first and then work on expanding it? |
|
@stuartarchibald Sure, if you think that would be best. What are the tricks? |
|
Btw, shouldn't this be implemented in |
Yeah, probably. I think my rationale was that |
… once in second implementation
|
what is the status on merging this in, this feature would be great |
|
I'd also like to join the chorus of requestors. Tile is such a useful utility function it can entail quite a bit of refactoring to replace it! |
|
After merging master into this branch there is a small conflict which is trivial to resolve, and the added tests of |
AndrewEckart
requested review from
esc,
sklam and
stuartarchibald
as code owners
| @@ -1712,7 +1712,7 @@ def np_triu_indices_from_impl(arr, k=0): | |||
| return np_triu_indices_from_impl | |||
|
|
|||
|
|
|||
| def _prepare_array(arr): | |||
| def _prepare_array(arr) -> np.ndarray: | |||
There was a problem hiding this comment.
Not sure how this type annotation got in here - my guess is that it was removed some time in the last several months after I originally branched, and then git put it back when I merged master into the feature branch.
|
Thanks for the update! I've added this to the review queue, and put it in the 0.54 milestone to make sure it doesn't get passed over or omitted from consideration for the next release without proper consideration. |
|
Excited to use this in anger! One less impediment to decorating a function and being off to the races. Thanks a lot guys, very much appreciated! |
|
It would be great to have this functionality. Is this functionality going to be merged sometime soon? Seems like this PR was added for a review for 0.54 release. Is anyone still working on this or this PR stale? |
|
Hello, it would be useful to have a built-in implementation of np.tile. Hope it will be merge in the future. |
Add basic support for np.tile, limited to when the first input (the tiles) have at most 1 dimension. Includes documentation and tests.
I'd like to support higher-dimensional inputs, but I spent a while trying to figure out how to dynamically allocate a new tuple at runtime for the output shape with little success (the number of dimensions needs to be equal to
max(len(reps), a.ndim). It seems like the only option iscontext.make_tupleusing thelower_builtinAPI, but that didn't seem like a great idea. A possible workaround is using a list/array for the shape, but unfortunately that's not yet supported bynp.reshapeorarray.reshape.Fixes #2327
Fixes #3456