使用Numpy.Reshape重塑数组

Question

I am trying to reshape an array using numpy.reshape but always come across the index error

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

I then printed out the shape of the array which was

(7231,80,60,4)

My code is

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

(im trying to reshape all of my image to (-1,80,60,1))

I thought -1 autocompleted the dimensions, so i am confused as to why I am getting this error?

train is:

    train = train_data[:-500]

and train_data is an array with tuples of image pixels and labels

Can someone help me?

Answer 1

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, ...])

Answer 2

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])

Answer 3

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?