Reshaping arrays using Numpy.Reshape

Be careful when reshaping. Even if it works, the arrangement of elements may not be what you want.

Start with a simple array that we can visualize:

In [805]: x = np.arange(24).reshape(3,2,4)
In [806]: x
Out[806]: 
array([[[ 0,  1,  2,  3],
        [ 4,  5,  6,  7]],

       [[ 8,  9, 10, 11],
        [12, 13, 14, 15]],

       [[16, 17, 18, 19],
        [20, 21, 22, 23]]])

reshape to (-1,2,1) - but lets drop the last 1 for a more compact display:

In [807]: x.reshape(-1,2)
Out[807]: 
array([[ 0,  1],
       [ 2,  3],
       [ 4,  5],
       [ 6,  7],
       [ 8,  9],
       [10, 11],
       [12, 13],
       [14, 15],
       [16, 17],
       [18, 19],
       [20, 21],
       [22, 23]])

Notice how the original [0,1,2,3] line gets split into 2 lines.

Another way of redistributing the last dimension of size 4 is:

In [808]: np.vstack([x[...,i] for i in range(4)])
Out[808]: 
array([[ 0,  4],
       [ 8, 12],
       [16, 20],
       [ 1,  5],
       [ 9, 13],
       [17, 21],
       [ 2,  6],
       [10, 14],
       [18, 22],
       [ 3,  7],
       [11, 15],
       [19, 23]])

That may be clearer if we used np.stack and got (4,3,2) shape

array([[[ 0,  4],
        [ 8, 12],
        [16, 20]],
  ....

x.transpose(2,0,1) produces the same thing.

reshape preserves the ravelled/flattened order of elements. Transpose changes it.

In [812]: x.transpose(2,0,1).ravel()
Out[812]: 
array([ 0,  4,  8, 12, 16, 20,  1,  5,  9, 13, 17, 21,  2,  6, 10, 14,...])
In [813]: x.reshape(-2,2).ravel()
Out[813]: 
array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, ...])

Reshaping an array of shape (7231,80,60,4) -> (-1,80,60,1) does "just work":

train = np.arange(np.prod((7231,80,60,4))).reshape(7231,80,60,4)
print(train.shape)

X = train.reshape(-1,80,60,1)
print(X.shape)

Output:

(7231, 80, 60, 4)
(28924, 80, 60, 1)

So the issues you're having must not directly stem from the reshape that you're trying to do. My guess is that your problem might relate to the form/contents of your train_data array (or the array you try to create from it with np.array([i[0] for i in train])). Of course, the problem could also be in a section of your code that you didn't post in your question. It would probably be helpful if you posted a bit more of your actual code.

In particular, when you got the error message:

IndexError: index 15484 is out of bounds for axis 0 with size 7231

it should have included a stack trace that pointed directly back to the problematic line in your code. Did the stack trace say that the error was raised from the line in which you create X:

X = np.array([i[0] for i in train]).reshape(-1,80,60,1)

or did it point to a different line in your code?

The code you have doesn't do what you think it does. i[0] gets the 0th element in the first axis, which is your 80, that's not what you want.

Anyway, what you really what is just to select the first slice in the fastest direction, so just do:

X = train[:,:,:,0:1]

If your data is not actually the size you say it is, then try:

X = np.array([i.reshape(80, 60, 4)[:,:,0:1] for i in train])