torch学习笔记(5)——autograd

import torch
import torch.nn as nn
import numpy as np
import matplotlib.pyplot as plt

# autograd
# fn1:torch.autograd.backward()自动求取梯度
# 参数：tensors:用于求导的tensor；retain_graph:保存计算图；create_graph:创建导数计算图，用于高阶求导；grad_tensors：多梯度权重
# fn2:torch.autograd.grad()自动求取梯度
# 参数：outputs：用于求导的tensor；input：需要求grad的tensor；create_graph:创建导数计算图，用于高阶求导；retain_graph:保存计算图；grad_outputs：多提取权重

# 案例一：backward()应用
w = torch.tensor([1.], requires_grad=True)
x = torch.tensor([2.], requires_grad=True)
a = torch.add(w, x)
b = torch.add(w, 1)
y0 = torch.mul(a, b)
y1 = torch.add(a, b)
loss = torch.cat([y0, y1], dim=0)
grad_tensors = torch.tensor([1., 1.])  # 设置权重
loss.backward(gradient=grad_tensors)
# y0.backward(retain_graph=True)  # 保留计算图
print(w.grad)

# 案例二：grad()应用
x = torch.tensor([3.], requires_grad=True)
y = torch.pow(x, 2)
grad_1 = torch.autograd.grad(y, x, create_graph=True)  # y对x求梯度，6
print(grad_1)
grad_2 = torch.autograd.grad(grad_1[0], x)  # y对x的梯度求梯度，2
print(grad_2)

# autograd使用注意事项
# 1、grad不会自动释放
# 2、requires_grad默认True
# 3、leaf不可以执行in-place

# 案例三：grad不释放后果
w = torch.tensor([1.], requires_grad=True)
x = torch.tensor([2.], requires_grad=True)
for i in range(4):
    a = torch.add(w, x)
    b = torch.add(w, 1)
    y = torch.mul(a, b)
    y.backward()
    # print(w.grad)# requires_grad默认True
    print(b.requires_grad)
    # w.grad.zero_()  # grad清零，name_表示in-place（原位）操作
# in-place操作
# 常规操作会开辟新的地址
a = torch.ones((1,))
print(id(a), a)
a = a + torch.ones((1,))
print(id(a), a)
# in-place操作不会开辟新的地址
a += torch.ones((1,))
print(id(a), a)
# leaf结点不能执行in-place操作