numpy 中的两个以下重塑 function 有什么区别？

Question

I am building a neural network. where I have to flatten my training dataset.

I have two options. 1 is:

train_x_flatten = train_x_orig.reshape(train_x_orig.shape[0], -1).T

and 2nd one is:

 train_x_flatten = train_x_orig.reshape(train_x_orig.shape[1]*train_x_orig.shape[2]*train_x_orig.shape[3], 209)

both gave the same shape but I found difference while computing cost? why is that? thank you

Answer 1

Your original tensor is of at least rank 4 based on the second example. The first example pulls each element, ordered by increasing the right-most index, and inserts the elements into rows the length of the zeroth shape. Then transposes.

The second example again pull elements from by incrementing from the right-most index, ie:

element = train_x_orig[0, 0, 0, 0]
new_row.append(element)
element = train_x_orig[0, 0, 0, 1]
new_row.append(element)

but the size of the row is different. It is now the dimension of everything else in the tensor.

Here is an example to illustrate.

First we create an ordered array and reshape it to rank 4.

import numpy as np

x = np.arange(36).reshape(3,2,3,2)
x
# returns:
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]]],


       [[[24, 25],
         [26, 27],
         [28, 29]],

        [[30, 31],
         [32, 33],
         [34, 35]]]])

Here is the output of the first example

x.reshape(x.shape[0], -1).T
# returns:
array([[ 0, 12, 24],
       [ 1, 13, 25],
       [ 2, 14, 26],
       [ 3, 15, 27],
       [ 4, 16, 28],
       [ 5, 17, 29],
       [ 6, 18, 30],
       [ 7, 19, 31],
       [ 8, 20, 32],
       [ 9, 21, 33],
       [10, 22, 34],
       [11, 23, 35]])

And here is the second example

x.reshape(x.shape[1]*x.shape[2]*x.shape[3], -1)
# returns:
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],
       [24, 25, 26],
       [27, 28, 29],
       [30, 31, 32],
       [33, 34, 35]])

How the elements get reordered is fundamentally different.