How does numpy.transpose work for this example?

问题:

I have difficulty understanding how numpy.transpose actually works. For example

a_value = array([[[0, 1],
                  [2, 3]],

                 [[4, 5],
                  [6, 7]]])

and when I do

np.transpose(a_value, (2, 1, 0))

I get

array([[[0, 4],
        [2, 6]],

       [[1, 5],
        [3, 7]]])

How can I derive this transpose manually? I need to understand the formula or the steps intuitively in the above case so I can generalize it for higher dimensions.

回答1:

As given in the documentation -

numpy.transpose(a, axes=None)

axes : list of ints, optional By default, reverse the dimensions, otherwise permute the axes according to the values given.

The second argument is the axes using which the values are permuted. That is for example if the index of initial element is (x,y,z) (where x is 0th axes, y is 1st axes, and z is 2nd axes) , the position of that element in the resulting array becomes (z,y,x) (that is 2nd axes first, then 1st axes, and last 0th axes) , based on the argument you provided for axes .

Since you are transposing an array of shape (2,2,2) , the transposed shape is also (2,2,2) , and the positions would change as -

(0,0,0) -> (0,0,0)
(1,0,0) -> (0,0,1)
(0,1,0) -> (0,1,0)
(1,1,0) -> (0,1,1)
...

---

Since the axes you choose are trivial, lets explain this for another axes. Example -

In [54]: A = np.arange(30).reshape((2, 3, 5))
In [55]: A
Out[55]:
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]]])

In [56]: np.transpose(A,(1,2,0))
Out[56]:
array([[[ 0, 15],
        [ 1, 16],
        [ 2, 17],
        [ 3, 18],
        [ 4, 19]],

       [[ 5, 20],
        [ 6, 21],
        [ 7, 22],
        [ 8, 23],
        [ 9, 24]],

       [[10, 25],
        [11, 26],
        [12, 27],
        [13, 28],
        [14, 29]]])

Here, the first element (0,0,0) becomes the (0,0,0) element in the result.

The second element (0,0,1) becomes the (0,1,0) element in the result. And so on -

(0,0,0) -> (0,0,0)
(0,0,1) -> (0,1,0)
(0,0,2) -> (0,2,0)
...
(2,3,4) -> (3,4,2)
...

回答2:

Here's a little more clarification:

Don't confuse the parameters of np.reshape(z, y, x) with those of np.transpose(0, 1, 2).

np.reshape() uses the dimensions of our matrix, think (sheets, rows, columns), to specify its layout.

np.transpose() uses the integers 0, 1, and 2 to represent the axes we want to swap, and correspond to z, y, and x, respectively.

For example, if we have data in a matrix of 2 sheets, 3 rows, and 5 columns...

We can take the next step and think in terms of lists. So, the z, y, x or sheets, rows, columns representation of a 2x3x5 matrix is...

 [[[ 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]]]

...but the module we're feeding this data into requires a layout such that sheet 1 contains the first row of each of our sheets and sheet 2 contains the second row and so on. Then, we'll need to transpose our data with np.transpose(1, 0, 2). This swaps the z and the y axes and transposes the data.

 [[[ 0,  1,  2,  3,  4],
   [15, 16, 17, 18, 19]],

  [[ 5,  6,  7,  8,  9],
   [20, 21, 22, 23, 24]],

  [[10, 11, 12, 13, 14],
   [25, 26, 27, 28, 29]]]

Note the difference from using np.reshape(3, 2, 5) as this doesn't transpose the data--only re-arranges it.

[[[ 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]]]