generate long list (with repetition) of permutations of an array with numpy

Question

I want to be able to generate a list (of arbitrary length) of arrays, each array being a random permutation of one array A.

There's this python option :

for x in range(n):
    random.shuffle(A)
    routes.append(A)

But I suspect there's a less procedural way to achieve this with numpy. Any suggestion?

Answer 1

Consider this example:

In [59]: crr
Out[59]: 
array([[0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938]])

In [60]: np.random.shuffle(crr[0])

In [61]: crr
Out[61]: 
array([[0.42109715, 0.50779425, 0.93753455, 0.11773652, 0.08751938,
        0.83704624],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938],
       [0.83704624, 0.42109715, 0.50779425, 0.93753455, 0.11773652,
        0.08751938]])

As random shuffle is in-place, here we can specify only the first row to be shuffled. Therefore if you use np.repeat or np.tile to repeat your A for N times and reshape the resultant into a 2D array like crr here, you can do this to reach final goal:

In [69]: for v in crr:
    ...:     np.random.shuffle(v)