tf.transpose()函数介绍和示例

tf.transpose()函数介绍和示例

tf.transpose(X, perm=None, name=‘transpose’, conjugate=False)

释义：交换维度

• X：需要变换的张量
• perm：新的维度序列
• name：（可选）操作名称
• conjugate：（可选），设置为True，等价于 tf.conj(tf.transpose())，为维度倒序交换，且共轭

示例1：二维

import tensorflow as tf
import numpy as np
 
A = np.array([[1, 2, 3], [4, 5, 6]])
X = tf.transpose(A, [1, 0])             # 相当于转置
 
with tf.Session() as sess:
    print('original:\n',A)
    print('tranpose:\n',sess.run(X))

original:
 [[1 2 3]
 [4 5 6]]
tranpose:
 [[1 4]
 [2 5]
 [3 6]]

示例2：三维

import tensorflow as tf
import numpy as np
 
A = np.arange(12).reshape([2,3,2])
X = tf.transpose(A,[0,2,1])              # 交换维度 1 和维度 2
Y = tf.transpose(A,[1,0,2])              # 交换维度 0 和维度 1 
with tf.Session() as sess:
    print('original:\n', A)
    print('A.shape：', A.shape)
    print('='*30)
    
    print('transpose [0,2,1]:\n', sess.run(X))   
    print('X.shape：', X.shape)
    print('='*30)
    
    print('transpose [1,0,2]:\n', sess.run(Y))
    print('Y.shape：', Y.shape)

original:
 [[[ 0  1]
  [ 2  3]
  [ 4  5]]

 [[ 6  7]
  [ 8  9]
  [10 11]]]
A.shape： (2, 3, 2)
==============================
transpose [0,2,1]:
 [[[ 0  2  4]
  [ 1  3  5]]

 [[ 6  8 10]
  [ 7  9 11]]]
X.shape： (2, 2, 3)
==============================
transpose [1,0,2]:
 [[[ 0  1]
  [ 6  7]]

 [[ 2  3]
  [ 8  9]]

 [[ 4  5]
  [10 11]]]
Y.shape： (3, 2, 2)

示例3：conjugate

• 实数情况下，维度倒序交换

  import tensorflow as tf
  import numpy as np
   
  A = tf.constant(np.arange(120).reshape([2,3,4,5]))
  
  X = tf.transpose(A, conjugate=True)      # conjugate=True，维度倒序交换
  X1 = tf.conj(tf.transpose(A))            # 同理，tf.conj(tf.transpose())
  X2 = tf.transpose(A, [3,2,1,0])          # 同理，维度 [3,2,1,0]
  with tf.Session() as sess:
      print('A.shape：', A.shape)
      print('X.shape：', X.shape)
      print('X1.shape：', X1.shape)
      print('X2.shape：', X2.shape)
  

  A.shape： (2, 3, 4, 5)
  X.shape： (5, 4, 3, 2)
  X1.shape： (5, 4, 3, 2)
  X2.shape： (5, 4, 3, 2)
  

• 复数情况下，复数共轭，且维度倒序交换

  import tensorflow as tf
   
  A = tf.constant([[1 + 1j, 2 + 2j, 3 + 3j],
                   [4 + 4j, 5 + 5j, 6 + 6j]])
  X = tf.transpose(A, conjugate=True)             # conjugate=True，复数共轭，维度倒序
  X_ = tf.conj(tf.transpose(A))                   # 同理，tf.conj(tf.transpose())
  with tf.Session() as sess:
      print('A.shape：', A.shape)
      print(sess.run(A))
      print('='*30)
      
      print('X.shape：', X.shape)
      print(sess.run(X))
      print('='*30)
      
      print('X_.shape：', X_.shape)
      print(sess.run(X_))
  

  A.shape： (2, 3)
  [[1.+1.j 2.+2.j 3.+3.j]
   [4.+4.j 5.+5.j 6.+6.j]]
  ==============================
  X.shape： (3, 2)
  [[1.-1.j 4.-4.j]
   [2.-2.j 5.-5.j]
   [3.-3.j 6.-6.j]]
  ==============================
  X_.shape： (3, 2)
  [[1.-1.j 4.-4.j]
   [2.-2.j 5.-5.j]
   [3.-3.j 6.-6.j]]