What is the best way to delete the elements from one numpy array in another? Essentially I'm after np.delete() where the order of the arrays doesn't matter.
import numpy as np
a = np.array([2,1,3])
print a
b = np.array([4,1,2,5,2,3])
b = np.delete(b, a) # doesn't work as desired
print b # want [4,5,2]
Iterating over the elements of a is very slow for large arrays.
Here's an approach using sorting -
def remove_first_match(a,b):
sidx = b.argsort(kind='mergesort')
unqb, idx = np.unique(b[sidx],return_index=1)
return np.delete(b,sidx[idx[np.in1d(unqb,a)]])
Sample runs -
In [177]: a = np.array([2,1,3])
...: b = np.array([4,1,2,5,2,3,2,3])
...:
In [178]: remove_first_match(a,b)
Out[178]: array([4, 5, 2, 2, 3])
In [179]: a = np.array([2,2,1,3])
...: b = np.array([4,5])
...:
In [180]: remove_first_match(a,b)
Out[180]: array([4, 5])
You can use np.argmax to find the first True element along a set of rows or columns. So, for example, you can do a broadcasted version of this operation this way:
>>> a = np.array([2,1,3])
>>> b = np.array([4,1,2,5,2,3])
>>> np.delete(b, np.argmax(b == a[:, np.newaxis], axis=1))
array([4, 5, 2])
Of course, as with many numpy vectorized operations, the speed comes at the cost of allocating an array of size len(a) * len(b), so depending on your inputs this may not be appropriate.
[4,5,2]?2is also ina, so it should be removed andbwould be[4,5].a. As writtenahas two2's.a = [2,2,1,3],b == [4,5]? What ifa = [2,2,2,1,3]? Your case is too unspecific as it currently stands.b == [4,5]. In the second case I already know thataonly contains elements withinb.ahave duplicate values too like :a = [2,2,1,3]?