Question

idtopick is an array of ids

     idtopick=array([50,48,12,125,3458,155,299,6,7,84,58,63,0,8,-1])

idtolook is another array containing the ids I’m interested in

     idtolook=array([0,8,12,50])

I would like to store in another array the positions of idtopick that corresponds to idtolook.

This is my solution

    positions=array([where(idtopick==dummy)[0][0] for dummy in idtolook])

Resulting in

    array([12, 13,  2,  0])

It works but in reality the arrays I’m working with store millions of point so the above script is rather slow. I would like to know if there’s a way to make it faster. Also, I want to keep the order of idtolook so any algorithm that would sort it wouldn’t work for my case.

Answer 1

You can use sorting:

 sorter = np.argsort(idtopick, kind='mergesort') # you need stable sorting
 sorted_ids = idtopick[sorter]
 positions = np.searchsorted(sorted_ids, idtolook)
 positions = sorter[positions]

Note that it won’t throw an error though if there is and idtolook missing in idtopick. You could actually sort idtolook into the results array too, which should be faster:

 c = np.concatenate((idtopick, idtolook))
 sorter = np.argsort(c, kind='mergesort')
 #reverse = np.argsort(sorter) # The next two lines are this, but faster:
 reverse = np.empty_like(sorter)
 reverse[sorter] = np.arange(len(sorter))
 positions = sorter[reverse[-len(idtolook):]-1]

Which has similarity to the set operations.