Difference between np.array[a:b, c:d] and np.array[a:b][c:d] [duplicate]

Question

My confusion arises from the following example:

import numpy as np

a = np.array([['A', 'B', 'C', 'D', 'E'],
       ['F', 'G', 'H', 'I', 'J'],
       ['K', 'L', 'M', 'N', 'O'],
       ['P', 'Q', 'R', 'S', 'T'],
       ['U', 'V', 'W', 'X', 'Y']])

print(a[:2, 1:4])
print()
print(a[:2][1:4])

Output:

[['B', 'C', 'D'],
 ['G', 'H', 'I']]

[['F', 'G', 'H', 'I', 'J']]

---

x = np.arange(10)
x.shape = (2,5)

In this example given by Numpy.org in the Single element indexing section, x[0,2] = x[0][2], so I assumed this would hold for the above.

I've tried the above with other examples such as y = np.arange(35).reshape(5,7) with y[1:5:2,::3] and y[1:5:2][::3]. The output is different: I think this does not hold when slices are used.

Answer 1

First of all, 2D ndarray can be sliced as a[row slice, col slice]. So, x[:2, 1:4] is to slice ndarray x based on both row slice ([:2) and column slice ([1:4]).

However, x[:2][1:4] is slice [:2] first, and then slice [1:4]. Thus, x[:2][1:4] is the same as x[1:2].