In fact, it is a relatively simple knowledge point, but because I often forget it, record it here
Take [:, -1, :] as an example
The colon represents all elements of the dimension
-1 represents the last element of the dimension
Such as
output = np.array([i for i in range(8)]).reshape(2, 2, 2)
print(output)
The output is
[[[0 1]
[2 3]]
[[4 5]
[6 7]]]
And after processing [:, -1, :]
print(output[:, -1, :])
become
[[2 3]
[6 7]]
[2 3] is the entire content of the last element in the second dimension
Note that this result actually reduces one dimension, and the result is a two-dimensional matrix
Similarly, we can look at the results of [:, :, -1] and [-1, :, :]
print(output[:, :, -1])
print(output[-1, :, :])
Respectively are
[[1 3]
[5 7]]
[[4 5]
[6 7]]
The result is actually reduced by one dimension, a two-dimensional matrix