I used boolean masking.

Script:

import numpy as np

a = 0.5 
b = 0.6

M = np.zeros((16, 16))  # empty matrix

np.fill_diagonal(M, 0.9)  # diagonal elements
 
M[0, [1, 3]] = a   
M[3, [0, 2]] = b  
M[5, [4, 6]] = a   
 

print(M)