I have a numpy array padded_train_x of shape (2500,500).
The problem is,
when I try to get the shape of an element of this array like
padded_train_x[0].shape it outputs (500,) but when I run it as padded_train_x[0:1] it outputs (1,500). Why does this happen?
I'm trying to make prediction in an LSTM model using keras but I have to use padded_train_x[0:1] as the input instead of simply padded_train_x[0]
That is because making slice by padded_train_x[0:1] you get 2d array:
a = np.linspace(1024).reshape(64,-1)
b = a[0]
c = a[0:1]
b
array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15])
b[0]
0
c
array([[ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]])
c[0]
array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15])
As to why it happens, let's wait for someone more expert, not sure there really is a reason.
NumPy keeps dimensions when slicing and drops them when indexing. It's actually a Python thing, the same happens with lists.
You can drop single-dimensional axes with np.squeeze
a = np.ones((2500, 500))
a[0].shape
(500,)
a[0:1].shape
(1, 500)
a[0:1].squeeze().shape
(500,)
