动手学深度学习（Pytorch）1预备知识代码

第1章 预备知识

# -*- coding: utf-8 -*-
"""
Created on Sun Aug 15 17:41:37 2021

@author: 23820
"""

import torch
print("版本号为：{}".format(torch.__version__))


tensor_a = torch.arange(0, 12)
print(tensor_a)

print("张量的存储位置：{}".format(tensor_a.device))
print(tensor_a.shape)

print(tensor_a.numel())

X = tensor_a.view(3, 4)
print(X)

torch.zeros((2, 3, 4))
torch.ones((3, 4))
torch.randn((3, 4))

Y = torch.arange(3, 15).view(3, 4)

X + Y
X - Y
X * Y
X / Y

torch.exp(Y.float())
X @ Y.t()

torch.cat((X, Y), 0)
torch.cat((X, Y), 1)

X == Y
torch.sum(X)
torch.norm(X.float()).item()

A = torch.arange(0, 3).view(3, 1)
B = torch.arange(0, 2).view(1, 2)
A + B

X[1:3]
X[1, 2] = 9
X[1:2, :] = 12#前面代表第2行，后面代表所有列


Z = torch.zeros_like(Y)
before = id(Z)
Z[:] = X + Y
id(Z) == before

torch.add(X, Y, out=Z)
id(Z) == before


before = id(X)
X += Y
id(X) == before


import numpy as np
P = np.ones((2, 3))
D = torch.from_numpy(P)
D

D.numpy()


from torch import autograd

#初始化变量
x = torch.arange(4).float()
#将其属性 requires_grad_ 设置为 True ，它将开始追踪（track）在其上的所有操作（这样就可以利用链式法则进行梯度传播了）。
x.requires_grad_(True)
#构建表达式
y = 2*torch.dot(x, x.t())
#调用 backward() 函数自动求梯度。
y.backward()
#验证梯度
print(x.grad)


print(dir(torch.random))
help(torch.ones_like)