PyTorch中给的view()和resize_()方法的使用

 view()函数

view()函数是调整矩阵的形状

a1 = torch.arange(0,12).view(3,4)
print(a1)
# tensor([[ 0,  1,  2,  3],
#         [ 4,  5,  6,  7],
#         [ 8,  9, 10, 11]])

第一个参数是有多少行，第二个参数是每一行有多少个参数，但是必须保证range()函数与view()函数的数量是一致的，否则会报下图错误

import torch

a1 = torch.arange(0,12).view(3,2)
print(a1)


Traceback (most recent call last):
  File "C:/Users/yiwan/Desktop/Hornet/test/__init__.py", line 3, in <module>
    a1 = torch.arange(0,12).view(3,2)
RuntimeError: shape '[3, 2]' is invalid for input of size 12

view()函数变形的时候不修改里面的数据，只改变大小。

resize_函数

resize_函数也是修改矩阵的尺寸的，但是有一定区别，如果超过了矩阵的大小，会自动补全。补全的是矩阵的内存，resize_()第一个参数是有多少行，第二个参数是每行有多少个。

a2 =a1.resize_(3,5)

print(a2)

#tensor([[0,1,2, 3,4],
# [5, 6,7, 8,9],
# [10,1, 32651599815639144,25896191785697385, 27303570963497060]])


a3 = a1.resize_(5,3)
print(a3)

# tensor([[                0,                 1,                 2],
#         [                3,                 4,                 5],
#         [                6,                 7,                 8],
#         [                9,                10,                11],
#         [25896191785238631, 27866512327901300, 31525674138992744]])

该函数是截取某一个列或者某一行

a4 = a1 .resize_(1,4)
print(a3)
# tensor([[0, 1, 2, 3]])

a5  = a1.resize_(3,1)
print(a5)
# tensor([[0],
#         [1],
#         [2]])


a6  = a1.resize_(2,4)
print(a6)
#tensor([[0, 1, 2, 3],
#       [4, 5, 6, 7]])

但是不能取负数,否则会报超出内存的错误

a7 = a1.resize_(3,-1)
print(a7)
# Traceback (most recent call last):
#   File "C:/Users/yiwan/Desktop/Hornet/test/__init__.py", line 43, in <module>
#     a7 = a1.resize_(3,-1)
# RuntimeError: [enforce fail at ..\c10\core\CPUAllocator.cpp:53] ((ptrdiff_t)nbytes) >= 0. alloc_cpu() seems to have been called with negative number: 18446744073709551592
#