numpy.pad、numpy.reshape、numpy.transpose

在卷积神经网络中，在convolutional layer需要对tensor进行填充，在数据处理时需要改变tensor的形状，对tensor进行转置。

---

 numpy.pad(array, pad_width, mode, **kwargs)，array为需要填充的array对象，pad_width为各个维度需要填充的宽度，mode为填充模式，常用的为constant。

import numpy as np

# np.pad
vector = np.array([1, 2])
vector_pad_constant = np.pad(vector, (1, 2), 'constant', constant_values=(3, 4))
vector_pad_edge = np.pad(vector, (1, 2), 'edge')
print(vector_pad_constant)
print(vector_pad_edge)

matrix = np.array([[1, 2], [3, 4]])
matrix_pad_constant = np.pad(matrix, [(1, 1), (2, 2)], 'constant', constant_values=[(1,2), (3, 4)]) # 二维数组pad_width传入数组,表示先在上下方向填充多少行，再在左右填充多少列
matrix_pad_edge = np.pad(matrix, [(1, 1), (2, 2)], 'edge')
print('---------------------------------------')
print(matrix_pad_constant)
print(matrix_pad_edge)

[3 1 2 4 4]
[1 1 2 2 2]
---------------------------------------
[[3 3 1 1 4 4]
 [3 3 1 2 4 4]
 [3 3 3 4 4 4]
 [3 3 2 2 4 4]]
[[1 1 1 2 2 2]
 [1 1 1 2 2 2]
 [3 3 3 4 4 4]
 [3 3 3 4 4 4]]

 

---

  numpy.reshape(a, newshape, order='C')，a为需要reshape对象，newshape为新的形状。

# np.reshape
matrix_reshape = matrix.reshape(4, -1) # reshape为一个4*？的矩阵，-1表示列数自动计算
print(matrix_reshape.shape)
print(matrix_reshape)

_matrix_reshape = matrix.reshape(-1, 4) # 行数自动计算
print(_matrix_reshape.shape)
print(_matrix_reshape)

(4, 1)
[[1]
 [2]
 [3]
 [4]]
(1, 4)
[[1 2 3 4]]

 

---

 

numpy.transpose(a, axes=None)，a为需要转置对象，axes为None时，默认反转矩阵的维度，否则按照axes的值重新排列矩阵。

将一个矩阵的各个维度重新排列，可以理解为对矩阵进行旋转从而改变了观察矩阵的视角。

By default, reverse the dimensions, otherwise permute the axes according to the values given.。

# np.transpose
tensor = np.array([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12])
tensor_reshape = tensor.reshape((2, 2, 3))
print(tensor_reshape.shape)
print(tensor_reshape)

tensor_transpose = tensor_reshape.transpose(2, 0, 1)
print(tensor_transpose.shape)
print(tensor_transpose)

(2, 2, 3)
[[[ 1  2  3]
  [ 4  5  6]]

 [[ 7  8  9]
  [10 11 12]]]
(3, 2, 2)
[[[ 1  4]
  [ 7 10]]

 [[ 2  5]
  [ 8 11]]

 [[ 3  6]
  [ 9 12]]]