Concatenating column vectors using numpy arrays

Question

I'd like to concatenate 'column' vectors using numpy arrays but because numpy sees all arrays as row vectors by default, np.hstack and np.concatenate along any axis don't help (and neither did np.transpose as expected).

a = np.array((0, 1))
b = np.array((2, 1))
c = np.array((-1, -1))

np.hstack((a, b, c))
# array([ 0,  1,  2,  1, -1, -1])  ## Noooooo
np.reshape(np.hstack((a, b, c)), (2, 3))
# array([[ 0,  1,  2], [ 1, -1, -1]]) ## Reshaping won't help

One possibility (but too cumbersome) is

np.hstack((a[:, np.newaxis], b[:, np.newaxis], c[:, np.newaxis]))
# array([[ 0,  2, -1], [ 1,  1, -1]]) ##

Are there better ways?

abudis · Accepted Answer · 2013-02-07 14:18:33Z

61

I believe numpy.column_stack should do what you want. Example:

>>> a = np.array((0, 1))
>>> b = np.array((2, 1))
>>> c = np.array((-1, -1))
>>> numpy.column_stack((a,b,c))
array([[ 0,  2, -1],
       [ 1,  1, -1]])

It is essentially equal to

>>> numpy.vstack((a,b,c)).T

though. As it says in the documentation.

edited Feb 7, 2013 at 14:18

answered Feb 7, 2013 at 14:07

abudis

2,8917 gold badges36 silver badges41 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

green diod Over a year ago

I didn't see that one!! I'm afraid the mathesaurus web site is quite outdated. I wish hstack would do the trick without resorting to column_stack but it's good enough.

Pavan Yalamanchili · Accepted Answer · 2013-02-06 23:49:33Z

3

I tried the following. Hope this is good enough for what you are doing ?

>>> np.vstack((a,b,c))
array([[ 0,  1],
       [ 2,  1],
       [-1, -1]])
>>> np.vstack((a,b,c)).T
array([[ 0,  2, -1],
       [ 1,  1, -1]])

answered Feb 6, 2013 at 23:49

Pavan Yalamanchili

12.1k2 gold badges39 silver badges56 bronze badges

4 Comments

green diod Over a year ago

Yeah, that's good, thanks but I thought that there would be something that could be done along the hstack way (I mean in a more straightforward way. To lower the cognitive load, it would be better to see and concatenate these vectors as column vectors)

Pavan Yalamanchili Over a year ago

a, b and c are row vectors. so it is logical hstack creates a row vector. Having a = np.array(([0], [1])) (and the same with b and c) followed by a hstack would be what you are looking for.

green diod Over a year ago

We could do it like this, but we could also just consider that these are all column vectors and it is supposed to be ok as far as numpy is concerned. The problem with this approach is you're already changing the shape of the vectors. In fact, I stumbled upon this when I played with some simple linear algebra in Matlab and doing the same in Python as done there. However in Matlab everything is a 2D-matrix, so we would get the result with a plain [a' b' c']. Unfortunately, using a direct translation np.hstack((a.T, b.T, c.T)) fails indeed

hwrd Over a year ago

Regarding the comment "To lower the cognitive load, it would be better to see and concatenate these vectors as column vectors)", why not just: np.concatenate([v.reshape((-1,1)) for v in (a, b, c)], axis=1)

Collectives™ on Stack Overflow

Concatenating column vectors using numpy arrays

2 Answers 2

1 Comment

4 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

4 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related