transpose在numpy和torch中的不同
numpy和torch中transpose的功能不同
numpy.ndarray.transpose的官方文档
torch.transpose的官方文档
numpy.transpose需指定一个新的axis的顺序
import numpy as np
array_ = np.arange(24).reshape(1,2,3,4)
print(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]]]]
array_t = array_.transpose([0,3,2,1])
print(array_t)
# [[[[ 0 12]
# [ 4 16]
# [ 8 20]]
#
# [[ 1 13]
# [ 5 17]
# [ 9 21]]
#
# [[ 2 14]
# [ 6 18]
# [10 22]]
#
# [[ 3 15]
# [ 7 19]
# [11 23]]]]
array_tt = array_.transpose([0,3,2,1])
print(array_tt)
# [[[[ 0 1 2 3]
# [ 4 5 6 7]
# [ 8 9 10 11]]
#
# [[12 13 14 15]
# [16 17 18 19]
# [20 21 22 23]]]]
torch.transpose仅交换两个特定的axis
import torch
import numpy as np
tensor_ = torch.from_numpy(np.arange(24).reshape(1,2,3,4))
print(tensor_)
# tensor([[[[ 0, 1, 2, 3],
# [ 4, 5, 6, 7],
# [ 8, 9, 10, 11]],
#
# [[12, 13, 14, 15],
# [16, 17, 18, 19],
# [20, 21, 22, 23]]]])
tensor_t = tensor_.transpose(1, 3)
print(tensor_t)
# tensor([[[[ 0, 12],
# [ 4, 16],
# [ 8, 20]],
#
# [[ 1, 13],
# [ 5, 17],
# [ 9, 21]],
#
# [[ 2, 14],
# [ 6, 18],
# [10, 22]],
#
# [[ 3, 15],
# [ 7, 19],
# [11, 23]]]])
tensor_tt = tensor_t.transpose(1, 3)
print(tensor_tt)
# tensor([[[[ 0, 1, 2, 3],
# [ 4, 5, 6, 7],
# [ 8, 9, 10, 11]],
#
# [[12, 13, 14, 15],
# [16, 17, 18, 19],
# [20, 21, 22, 23]]]])
在numpy中使用torch的语法会导致ValueError
import numpy as np
array_ = np.arange(24).reshape(1,2,3,4)
array_t = array_.transpose(1,3)
# Traceback (most recent call last):
# File "", line 1, in
# ValueError: axes don't match array