Question

I’ve a problem with numpy’s array constructor. I want to initialize an 2-D array with tuples, but it doesn’t work as with integers :

>>> A = array([[0, 0], [3, 5]])
>>> print(A)
[[0 0]
[3 5]]
>>> A[1, 1] = 7
>>> print(A)
[[0 0]
[3 7]]
>>> A = array([[(0, 0), (0, 1)], [(1, 0), None]], dtype=object)
>>> A[1, 1] = (2, 3)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
IndexError: invalid index
>>> A.shape
(2,)

I really need to fill these matrix “by hand”. Any idea ?

Answer 1

Your code seems to work for me (using the explicit numpy namespace). I’m using numpy v1.6.1:

In [8]: import numpy as np
In [9]: A = np.array([[(0, 0), (0, 1)], [(1, 0), None]], dtype=object)

In [10]: A[1, 1] = (2, 3)

In [11]: A.shape
Out[11]: (2, 2)

In [12]: A
Out[12]: 
array([[(0, 0), (0, 1)],
       [(1, 0), (2, 3)]], dtype=object)

What version of numpy are you using?

Update This seems to be an issue related to the numpy version since I can reproduce the OP’s error using numpy v1.5.1 (the version that comes packaged with the base python install in OSX Lion). I’m not sure if this was a bug in numpy that was fixed or a change in the implementation. I would either update to a newer version of numpy or use this simple workaround:

>>> A = np.array([[(0, 0), (0, 1)], [(1, 0), None]], dtype=object)
>>> A[1][1] = (2,3)
>>> A
array([[(0, 0), (0, 1)], [(1, 0), (2, 3)]], dtype=object)

Update #2 Here’s a general fix that hopefully you can adapt:

>>> C = np.empty((2,2),object)
>>> B = [[(0, 0), (0, 1)], [(1, 0), None]]
>>> C[:] = B
>>> C
array([[(0, 0), (0, 1)],
       [(1, 0), None]], dtype=object)
>>> C.shape
(2, 2)
>>> C[1,1] = (2,3)
>>> C
array([[(0, 0), (0, 1)],
       [(1, 0), (2, 3)]], dtype=object)