how to transform element then insert into numpy array

The desired array has a different shape than the original array, a. Therefore, you can not generate the desired array by using simple slice assignment, or a single call to assignment functions like numpy.place, numpy.put, numpy.putmask or numpy.copyto.

So instead of regarding this as an assignment-and-insertion operation, I think it might be better to regard this as construction of a new array. There are many ways to do this:

• You could use numpy.concatenate:

  np.concatenate([a[:,:1], b, a[:,2:]], axis=1)
  

• or numpy.column_stack:

  np.column_stack([a[:,:1], b, a[:,2:]])
  

• or numpy.c_:

  np.c_[a[:,:1], b, a[:,2:]]
  

• or numpy.bmat:

  np.array(np.bmat([a[:,:1], b, a[:,2:]]))
  

• or allocate space for an empty array and assign values to slices of c:

  c = np.empty((a.shape[0], a.shape[1]+b.shape[1]-1), dtype=a.dtype)
  c[:, :1] = a[:, :1]
  c[:, 1:4] = b
  c[:, 4:] = a[:, 2:]
  

  This is not necessary here, but it is good to keep in mind that allocate-and-assign is a viable way to construct arrays. Sometimes, though not here, it is even your fastest option.

• or numpy.delete and numpy.insert:

  np.insert(np.delete(a, [1], 1), [1], b, axis=1)
  

  I don't recommend using insert and delete however. The np.delete creates a new array with values that are copied from a. In contrast, the slices a[:,:1], and a[:,2:] are views of a so they do not require any copying. The extra allocation and copying makes this method slower than the other methods shown above.

For example, if we define:

import numpy as np
a = np.array([[  0.,   2.,   0.,   0.,   0.,   0.],
              [  0.,  14.,   0.,   0.,   0.,   0.]])
b = np.array([(1, 0, 0), 
              (0, 0, 1)])

then

In [69]: np.concatenate([a[:,:1],b,a[:,2:]], axis=1)
Out[69]: 
array([[ 0.,  1.,  0.,  0.,  0.,  0.,  0.,  0.],
       [ 0.,  0.,  0.,  1.,  0.,  0.,  0.,  0.]])

(The other options produce the same result.)