MAINT Remove get_memview_* helpers in KDTree & BallTree #15097
Conversation
| # flatten X, and save original shape information | ||
| np_Xarr = X.reshape((-1, self.data.shape[1])) | ||
| cdef DTYPE_t[:, ::1] Xarr = get_memview_DTYPE_2D(np_Xarr) | ||
| cdef const DTYPE_t[:, ::1] Xarr = np_Xarr |
There was a problem hiding this comment.
Input array can be read-only, in which case setting it to a non const memoryview would fail.
|
Do we get some unintended speed degradation here? Could you maybe make sure there are no odd python interactions after the change? |
I did benchmarks at the time, that showed no significant performance cost, but can't find them anymore. I'm waiting for official ASV benchmarks to be added #16723 so we have an official reference point. |
|
@rth : are you still interested in pursuing this PR? 🙂 |
|
@jjerphan Feel free to continue it in a separate PR and ping me for review :) I think the only thing remaining was to resolve conflicts and run some benchmarks to confirm. |
|
Thanks for the quick reply! |
4 participants
The
get_memview_*Cython helpers in KDTree and BallTree were used because at the time numpy/Cython didn't fully support memoryviews and now can be removed.This was adapted from larsmans's refactoring PR in #4217