ENH: Let searchsorted work on arrays with descending values

Quite often arrays are sorted, either ascending or descending, and we want to find where the startpoint of a certain value is at. This is very easy for ascending arrays using searchsorted, but requires a bit more understanding for descending arrays.

It seems relatively straightward to it make it easy for descending arrays also by adding a ascending parameter to searchsorted.

Would this be outside of scope for numpy, or is such an ascending parameter relevant to include in numpy’s searchsorted?

A demonstration:

>>> import numpy as np

>>> def searchsorted(a, v, side='left', sorter=None, ascending=True):
>>>     if not ascending:
>>>         a = a[::-1]
>>>         side = {'left': 'right', 'right': 'left'}[side]
>>>         sorter = sorter[::-1] if sorter is not None else sorter
>>>         return len(a) - np.searchsorted(a, v, side=side, sorter=sorter)
>>>     return np.searchsorted(a, v, side=side, sorter=sorter)

>>> a = np.array([2, 1, 1])
>>> searchsorted(a, 1, ascending=False)
1
>>> searchsorted(a, 2, side='right', ascending=False)
1
``

[eric-wieser]

Perhaps reverse=True to match the builtin sorted.

I’m +1 on adding an argument to specify explicitly, but -1 on trying to auto-detect like np.digitize does.

[seberg]

That probably depends on the solution. If just reversing the array and using the normal searchsorted code is enough, you can do that inside array_searchsorted. If it is not a good solution, you would need to modify PyArray_SearchSorted, but rename the new function so the function signature of PyArray_SearchSorted is not modified (the original would become just 1 line of code calling the new function).