Replicating the indices result of Matlab's ISMEMBER function in NumPy?

I have been racking my brain for a solution that is along the lines of this older question. I have been trying to find a Python code pattern that replicates the indices result. For example:

A = [3;4;4;3;6]
B = [2;5;2;6;3;2;2;5]
[tf ix] = ismember(A,B)
>> A(tf)

ans =

     3
     3
     6
>> B(ix(tf))

ans =

     3
     3
     6

What this allows me to do is if there is an array C ordered the same way as B I can now appropriately insert the values of C into a new array D that is ordered the same way as A. I do this mapping of data a lot! I would love for this to work for various data types like strings and datetimes in pa开发者_JS百科rticular. It seems that numpy's in1d get's me half way there. I'm also open to other Pythonic ideas as well!

D(tf) = C(ix(tf))

Thank you!

---

import numpy as np

A = np.array([3,4,4,3,6])
B = np.array([2,5,2,6,3,6,2,2,5])

def ismember(a, b):
    # tf = np.in1d(a,b) # for newer versions of numpy
    tf = np.array([i in b for i in a])
    u = np.unique(a[tf])
    index = np.array([(np.where(b == i))[0][-1] if t else 0 for i,t in zip(a,tf)])
    return tf, index

tf,ix=ismember(A,B)
print(tf)
# [ True False False  True  True]
print(ix)
# [4 0 0 4 5]
print(A[tf])
# [3 3 6]
print(B[ix[tf]])
# [3 3 6]