numpy reshape 2d matrix to 3d matrix by stacking columns

Question

I have a numpy matrix with shape (x,y) . I want to get a tensor which is a vertical stack of y vectors with shape (x,1) . Lets say I have the following matrix:

array([[ 1.,  2.],
       [ 1.,  2.],
       [ 1.,  2.]])

when I do np.reshape(2,3,1) . I will get:

array([[[ 1.],
        [ 2.],
        [ 1.]],
       [[ 2.],
        [ 1.],
        [ 2.]]])

but I want this:

array([[[ 1.],
        [ 1.],
        [ 1.]],
       [[ 2.],
        [ 2.],
        [ 2.]]])

Answer 1

In [135]: arr=np.array([[1,2],[1,2],[1,2]])
In [136]: arr.shape
Out[136]: (3, 2)
In [137]: arr.transpose()
Out[137]: 
array([[1, 1, 1],
       [2, 2, 2]])
In [138]: arr.transpose()[:,:,None]
Out[138]: 
array([[[1],
        [1],
        [1]],

       [[2],
        [2],
        [2]]])

You want a shape (2,3,1). Starting with (3,2), that means you have to switch the 2 axes, and add one. That can be done in either order. Here I choose transpose to do the switch, and [:,:,None] to add the dimension.

arr.reshape(3,2,1).swapaxes(0,1) also works. More obscurely, np.stack(arr[:,:,None],1) .