参考 http://pytorch123.com/

1. tensor 张量

• empty 不初始化

import torch
x = torch.empty(5,3) # 不初始化
print(x)tensor([[1.0010e-38, 4.2246e-39, 1.0286e-38],[1.0653e-38, 1.0194e-38, 8.4490e-39],[1.0469e-38, 9.3674e-39, 9.9184e-39],[8.7245e-39, 9.2755e-39, 8.9082e-39],[9.9184e-39, 8.4490e-39, 9.6429e-39]])

• rand 随机初始化 0 - 1 之间

x = torch.rand(5,3) # 随机初始化tensor([[0.5931, 0.2422, 0.2738],[0.0949, 0.4755, 0.7422],[0.7418, 0.5980, 0.4837],[0.4228, 0.4489, 0.2633],[0.7277, 0.7254, 0.8932]])

• zeros 初始化为0，dtype指定数据类型

x = torch.zeros(5,3,dtype=torch.long)tensor([[0, 0, 0],[0, 0, 0],[0, 0, 0],[0, 0, 0],[0, 0, 0]])

• 直接赋值

x = torch.tensor([[5.5, 3], [2,4]])tensor([[5.5000, 3.0000],[2.0000, 4.0000]])

• new_* 方法，继承之前张量的属性，也可以覆盖以前的属性

x = x.new_ones(5,3,dtype=torch.double)
tensor([[1., 1., 1.],[1., 1., 1.],[1., 1., 1.],[1., 1., 1.],[1., 1., 1.]], dtype=torch.float64)x = x.new_zeros(2,4)
tensor([[0., 0., 0., 0.],[0., 0., 0., 0.]], dtype=torch.float64) # 可见属性继承了之前的

• rand_like 形状跟之前的一样

x = torch.randn_like(x,dtype=torch.float)
print(x)
print(x.size())tensor([[ 0.2575, -0.3525,  1.2242, -0.0641],[ 0.0307,  0.0433, -0.3609,  2.0844]])
torch.Size([2, 4])

2. 运算

x = torch.eye(3)
y = torch.zeros(3,3)
print(x+y) # +
print(torch.add(x,y)) # addres = torch.empty(2,2)
print(res.size())  # torch.Size([2, 2])
torch.add(x,y,out=res) # out 为输出变量
print(res)
print(res.size()) # torch.Size([3, 3])# in-place 加法
y.add_(x) # y = y+x， y 会变， 注意是 add_ 有下划线
print(y)

3. 切片、形状size()、改变形状view()

切片跟numpy一样

print(x[:,:1].size())  # torch.Size([3, 1])x = torch.randn(4,4)
y = x.view(16)
z = x.view(-1,8) # -1 自动推断
print(x.size(), y.size(), z.size())# torch.Size([4, 4]) torch.Size([16]) torch.Size([2, 8])

4. item() 只能读取一个元素

x = torch.randn(1)
print(x)
print(x.item())
# tensor([-0.3280])
# -0.327981561422348x = torch.randn(2,3)
print(x)
print(x[0,1].item()) # 只能获取一个元素# tensor([[-1.2239,  0.3518,  1.1019],
#         [-0.1341,  1.0625,  0.2442]])
# 0.3518247902393341