How to efficiently add a column in a multidimensional numpy array with different dimensions?

Question

Say I have these two arrays.

a = np.ones((2, 3, 4))
[[[1. 1. 1. 1.]
[1. 1. 1. 1.]
[1. 1. 1. 1.]]

[[1. 1. 1. 1.]
[1. 1. 1. 1.]
[1. 1. 1. 1.]]]

b = np.ones(6) * 9
[9. 9. 9. 9. 9. 9.]

Question. How do I efficiently get c such that it appears like this.

[[[1. 1. 1. 1. 9.]
[1. 1. 1. 1. 9.]
[1. 1. 1. 1. 9.]]

[[1. 1. 1. 1. 9.]
[1. 1. 1. 1. 9.]
[1. 1. 1. 1. 9.]]]

I've tried using np.c_, np.column_stack and np.insert but I'm struggling trying to figure out how to get the dimensions right. I can do it easily in the 1D and 2D case but anything more than that. :/

not the quickest, but it works for your small example …. np.c_[(a.reshape(6,4),b)].reshape(2,3,-1) — NaN
– NaN, Commented Dec 20, 2018 at 21:19

hpaulj · Accepted Answer · 2018-12-21 00:31:54Z

a is (2,3,4). To concatenate on the last axis, b has to be (2,3,1) (or more generally (2,3,n)). b.reshape(2,3) gets you part way there, add a np.newaxis to get the rest of the way. Or include the 3rd axis in the reshape: b.reshape(2,3,1).

In [21]: np.concatenate((a, b.reshape(2,3)[:,:,None]),-1)
Out[21]: 
array([[[1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.]],

       [[1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.]]])

np.c_ works with the same reshaping, np.c_[a, b.reshape(2,3,1)].

np.c_ (or np.r_) takes a string parameter that tells it how to expand dimensions if needed. np.c_ is the equivalent of np.r_['-1,2,0',a, b.reshape(2,3)[:,:,None]]. That string parameter is a bit hard to follow, but playing around I found this works:

In [27]: np.c_['-1,3,0',a, b.reshape(2,3)]
Out[27]: 
array([[[1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.]],

       [[1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.],
        [1., 1., 1., 1., 9.]]])

Keep in mind that np.c_ uses np.concatenate (as do all the 'stack' functions), so In [21] is the most direct version. Still some people like the convenience of the np.c_ format.

(np.column_stack joins on axis 1; the docs explicitly talk of returning a 2d array. You have 3d case.)

dstack ('depth' stack) works: np.dstack((a, b.reshape(2,3))). It creates the 3d arrays with:

In [49]: np.atleast_3d(b.reshape(2,3)).shape
Out[49]: (2, 3, 1)

which in this case is the same as what the '-1,3,0' string does.

Collectives™ on Stack Overflow

How to efficiently add a column in a multidimensional numpy array with different dimensions?

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related