REF/PERF: deduplicate kth_smallest #40559
Conversation
pandas/core/algorithms.py
Outdated
| @@ -1305,6 +1305,8 @@ def compute(self, method: str) -> Series: | |||
| narr = len(arr) | |||
| n = min(n, narr) | |||
|
|
|||
| # arr passed into kth_smallest must be contiguous, which | |||
There was a problem hiding this comment.
use ascontiguousarray instead (or specify the ordering flag in copy)
pandas/_libs/algos.pxd
Outdated
| @@ -12,6 +12,48 @@ cdef inline Py_ssize_t swap(numeric *a, numeric *b) nogil: | |||
| return 0 | |||
|
|
|||
|
|
|||
| cdef inline numeric kth_smallest_c(numeric* arr, Py_ssize_t k, Py_ssize_t n) nogil: | |||
There was a problem hiding this comment.
why is the implementation in the pxd file?
There was a problem hiding this comment.
It's cimported from groupby.pyx
There was a problem hiding this comment.
that means that it needs to be declared in the pxd file, the actual implementation can be in the pyx file
There was a problem hiding this comment.
Oh got it I see what you mean will change
pandas/_libs/algos.pyx
Outdated
|
|
||
| Parameters | ||
| ---------- | ||
| arr: numeric* arr |
There was a problem hiding this comment.
trailing "arr" not needed?
space before the colon in Parameters section
| @@ -64,6 +64,14 @@ cdef: | |||
| float64_t NaN = <float64_t>np.NaN | |||
| int64_t NPY_NAT = get_nat() | |||
|
|
|||
| cdef enum TiebreakEnumType: | |||
There was a problem hiding this comment.
it was in the pxd but didnt need to be
There was a problem hiding this comment.
Yep sorry it was an unrelated change
jreback
added
Algos
|
@mzeitlin11 any top-level user visible perf worth mentioning in the what's new? (e.g. nlargest / nsmallest) |
Based on the benchmark should speed up nlargest/nsmallest about 5-10%. Let me know if that's worth adding a whatsnew |
|
thanks @mzeitlin11 ok this was more of a refactor then. |
Ensuring a contiguous input seems to give about a 5-10% performance improvement on the existing benchmark
gil.ParallelKth