How to insert "multiple" numpy rows/ columns at one index-CodePudding

I have 2 numpy arrays:

a
array([[1, 1, 1],
       [2, 2, 2],
       [3, 3, 3],
       [4, 4, 4]])

and

aa
array([[11, 11, 11],
       [22, 22, 22],
       [33, 33, 33],
       [44, 44, 44],
       [55, 55, 55],
       [66, 66, 66]])

I want to insert fragments of aa into a such that a would look like:

a
array([[1, 1, 1],
       [22, 22, 22],
       [33, 33, 33],
       [2, 2, 2],
       [55, 55, 55],
       [66, 66, 66]
       [3, 3, 3],
       [4, 4, 4]]).

I tried:

np.insert(a,[1,3],[aa[1:3], aa[4:]], axis=0)
np.insert(a,[1,3],[[aa[1],aa[2]], [aa[4],aa[5]]], axis=0),

but got: ValueError: shape mismatch: value array of shape (2,2,3) could not be broadcast to indexing result of shape (2,3).

However for a single index it worked:

np.insert(a,1,[aa[1],aa[2]], axis=0)
array([[ 1,  1,  1],
       [22, 22, 22],
       [33, 33, 33],
       [ 2,  2,  2],
       [ 3,  3,  3],
       [ 4,  4,  4]])

but did not work for for slice:

np.insert(a,1,[aa[1:3]], axis=0)

Why np.insert() does not accept slices as values and is there a way to insert fragments of aa into a in one liner?

CodePudding user response：

Suppose we have this:

idx_list = [1,2]
slices = aa[np.r_[1:3,4:6]]
np.insert(a,idx_list,slices, 0)

We get a ValueError:

ValueError: shape mismatch: value array of shape (4,3)  could not be broadcast to indexing result of shape (2,3)

So, here np.insert is expecting to insert one row at each index in idx_list, but we are supplying only 2 index values, while we have 4 rows to insert. Solution is just to double the index values:

idx_list = [1,2]

idx_list_ext = []
for i in idx_list:
    idx_list_ext.extend([i, i])

# [1, 1, 2, 2]

slices = aa[np.r_[1:3,4:6]]
out = np.insert(a,idx_list_ext,slices, 0)

out

array([[ 1,  1,  1],
       [22, 22, 22],
       [33, 33, 33],
       [ 2,  2,  2],
       [55, 55, 55],
       [66, 66, 66],
       [ 3,  3,  3],
       [ 4,  4,  4]])

CodePudding user response：

First create an empty array that is of the same size of you output array:

out = np.empty((8, 3))
>>> out
    array([[0., 0., 0.],
           [0., 0., 0.],
           [0., 0., 0.],
           [0., 0., 0.],
           [0., 0., 0.],
           [0., 0., 0.],
           [0., 0., 0.],
           [0., 0., 0.]])

Then insert inside the out array at the respective indexes, a and aa:

out[[0, 3, 6, 7]] = a
out[[1, 2, 4, 5]] = aa[1:5]
>>> out
    array([[ 1.  1.  1.]
           [22. 22. 22.]
           [33. 33. 33.]
           [ 2.  2.  2.]
           [44. 44. 44.]
           [55. 55. 55.]
           [ 3.  3.  3.]
           [ 4.  4.  4.]])

CodePudding user response：

np.concatenate([
    a[:1],
    aa[1:3],
    a[1:2],
    aa[4:6],
    a[2:]
])

CodePudding user response：

This code works for longer arrays too :

j = 0
for i in range(0, len(aa), 3):
    aa[i] = a[j]
    j  = 1
aa = np.append(aa, a[j:], axis=0)