selecting rows from 2 matrices using an indicator vector in numpy

Question

Suppose I have a vector:

f = np.array([1,1,0,0]) #(4,)

and 2 matrices:

m1 = np.array([[1,2],[3,4],[5,6],[7,8]]) #(4,2)
m2 = np.array([[10,20],[30,40],[50,60],[70,80]]) #(4,2)

How can I create a new matrix m3 that selects rows from m1 where f == 1 and m2 otherwise?

I want m3 to be:

>>> m3
array([[ 1,  2],
       [ 3,  4],
       [50, 60],
       [70, 80]])

How do I achieve this? Would prefer a solution that I can use in theano as well.

`np.vstack((m1[np.where(f==1)], m2[np.where(f!=1)]))`? I am not familiar with `theano`. — Abdou, Feb 01 '17 at 19:59

score 2 · Accepted Answer · answered Feb 01 '17 at 19:59

2

I don't know about Theano but for numpy:

np.where(f[:, None], m1, m2)

answered Feb 01 '17 at 19:59

Paul Panzer

1

or `where(f,m1.T,m2.T).T` – B. M. Feb 01 '17 at 20:05
@B.M. interesting. I would have expected your line to return a non-C-contiguous array, but it doesn't. It seems exactly equivalent to mine. – Paul Panzer Feb 01 '17 at 20:19

1 Answers1