Pytorch基础之——（二）设立计算图并自动计算

1.numpy和pytorch实现梯度下降法

2.设定初始值

import numpy as np
import torch

device = torch.device('cpu')
# 参数
batch_size, in_dim, hidden_dim, out_dim = 64, 1000, 100, 10

# 生成输入输出数据、初始化权重
## numpy
x_np = np.random.randn(batch_size, in_dim)
y_np = np.random.randn(batch_size, out_dim)
w1_np = np.random.randn(in_dim, hidden_dim)
w2_np = np.random.randn(hidden_dim, out_dim)
## pytorch
x_tc = torch.randn(batch_size, in_dim, device=device)
y_tc = torch.randn(batch_size, out_dim, device=device)
w1_tc = torch.randn(in_dim, hidden_dim, device=device, requires_grad=True)
w2_tc = torch.randn(out_dim, hidden_dim, device=device, requires_grad=True)

3.求取梯度

# 学习率
learning_rate = 1e-6

# 更新轮数
for i in range(500):
    # 前向网络
    ## numpy
    h = x_np.dot(w1_np)
    h_relu = np.maximum(h