具有元组的二维numpy数组：扩展为带有原始元组组合的多个数组

Question

Suppose I have a NumPy array like the following:

0      1      (2,3)
(4,5)  6      7
9      (10,)  11

How could I split it a list of the arrays below? Conditions: no tuples in final arrays, and each array contains a unique combination of the tuple elements. That is, there are two tuples in the original array, so there are 4 different combinations of those elements. The first array below has the combination (2) and (4) from the original.

0  1  2
4  6  7
9  10 11

0  1  2
5  6  7
9  10 11

0  1  3
4  6  7
9  10 11

0  1  3
5  6  7
9  10 11

Answer 1

Such a data structure is awkward, as @unutbu suggested. But if you want to do it, it is doable:

In [52]:
A
Out[52]:
array([[0, 1, (2, 3)],
       [(4, 5), 6, 7],
       [9, (10,), 11]], dtype=object)

In [53]:
import itertools
AR=A.ravel()
AT=[i for i, val in enumerate(AAR) if isinstance(val, tuple)]
for item in itertools.product(*AR[AT]):
    B=AR.copy()
    B[AT]=item
    print B.reshape(A.shape)
[[0 1 2]
 [4 6 7]
 [9 10 11]]

[[0 1 2]
 [5 6 7]
 [9 10 11]]

[[0 1 3]
 [4 6 7]
 [9 10 11]]

[[0 1 3]
 [5 6 7]
 [9 10 11]]