动手学深度学习d2l——2预备知识

2.1.1. torch.view()和torch.reshape()的区别

代码是用jupyter notebook写的

import torch

a = torch.arange(6).reshape(2,3)
a
'''输出
tensor([[0, 1, 2],
        [3, 4, 5]])
'''

a = a.T
a
'''输出
tensor([[0, 3],
        [1, 4],
        [2, 5]])
'''

a.reshape(-1)
'''输出
tensor([0, 3, 1, 4, 2, 5])
'''

a.view(-1)
'''输出
RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.
'''

转置应该只改变读取顺序，不改变内存顺序（不确定是不是）
view需要保证读取顺序是内存连续的
总结：用reshape

2.1.3 广播机制

import torch
a = torch.arange(6).reshape(6,1)
b = torch.arange(3).reshape(1,3)
a,b,a+b
'''输出
(tensor([[0],
         [1],
         [2],
         [3],
         [4],
         [5]]),
 tensor([[0, 1, 2]]),
 tensor([[0, 1, 2],
         [1, 2, 3],
         [2, 3, 4],
         [3, 4, 5],
         [4, 5, 6],
         [5, 6, 7]]))
'''

m×1的矩阵 + 1×n的矩阵 = m×n的矩阵
可以理解为坐标轴，练习题里的三维张量相加可以看作三个坐标轴
实际上张量相加时，值为1的维度都会尝试进行广播，比如

# 有一个维度值为1
a2 = torch.arange(24).reshape(2,3,4)
b2 = torch.ones(1,3,4)
a2,b2,a2+b2
'''输出
(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([[[1., 1., 1., 1.],
          [1., 1., 1., 1.],
          [1., 1., 1., 1.]]]),
 tensor([[[ 1.,  2.,  3.,  4.],
          [ 5.,  6.,  7.,  8.],
          [ 9., 10., 11., 12.]],
 
         [[13., 14., 15., 16.],
          [17., 18., 19., 20.],
          [21., 22., 23., 24.]]]))
'''

# 有两个维度值为1
a3 = torch.arange(24).reshape(2,3,4)
b3 = torch.ones(1,1,4)
a3,b3,a3+b3
'''输出
(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([[[1., 1., 1., 1.]]]),
 tensor([[[ 1.,  2.,  3.,  4.],
          [ 5.,  6.,  7.,  8.],
          [ 9., 10., 11., 12.]],
 
         [[13., 14., 15., 16.],
          [17., 18., 19., 20.],
          [21., 22., 23., 24.]]]))
'''