Implement np.intersect1d #3726
Implement np.intersect1d #3726
Conversation
|
|
||
| if return_indices: | ||
| indices = argsort(ar) | ||
| aux = ar[indices] |
There was a problem hiding this comment.
It's possible to use lax.sort directly to get the sorted array and the indices in a single pass (if I recall correctly, argsort uses this and then discards the sorted array).
|
I think Jake and I were reviewing at the same time! His suggestion sounds good; maybe you can make another PR to incorporate it? |
* Implement np.intersect1d * Add jitable helper to function * Fix argsort failing tests * Fix linter errors
|
Sure, I'll add it and link to this PR |
* Implement np.intersect1d * Add jitable helper to function * Fix argsort failing tests * Fix linter errors
* Implement np.intersect1d * Add jitable helper to function * Fix argsort failing tests * Fix linter errors
* Implement np.intersect1d * Add jitable helper to function * Fix argsort failing tests * Fix linter errors
|
It seems that this function returns arrays of dynamic shapes, and is thus not compatible with jit? |
|
Correct, the semantics of numpy's #3614 tracks the effort to add size hints to functions that share this issue – many are done but Many of these kinds of features are implemented on an as-needed basis. If this is important to your use-case, you might chime-in on that issue with ideas of what kind of shape-hint behavior would be useful to you. |
This is an implementation of np.intersect1d from #70 (and #2078 )