实现多层感知机

 首先导入需要的包，定义超参数，加载数据集，参数随机初始化，定义网络模型，选择优化器，定义损失函数，训练。主要是代码量少，容易看懂。

#导入需要的包
import torch
from torch import nn
from d2l import torch as d2l
import torch.nn.functional as F
#定义超参数
batch_size = 256
num_epochs,lr = 10,0.1
num_inputs = 784
num_outputs = 10
num_hiddens = 256
#加载数据集
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
#初始化权重和偏置
W1 = torch.randn(num_inputs, num_hiddens, requires_grad=True)
b1 = torch.zeros(num_hiddens, requires_grad=True)
W2 = torch.randn(num_hiddens, num_outputs, requires_grad=True)
b2 = torch.zeros(num_outputs, requires_grad=True)
params = [W1, b1, W2, b2]

#模型网络结构
def net(X):
    X = X.reshape((-1, num_inputs))
    H = F.relu(X @ W1 + b1)
    return (H @ W2 + b2)
#定义损失函数和优化器
loss = nn.CrossEntropyLoss()
updater = torch.optim.SGD(params,lr=lr)
#训练
d2l.train_ch3(net,train_iter,test_iter,loss,num_epochs,updater)

 可以定义relu函数

def relu(X):
    a = torch.zeros_like(X)
    return torch.max(X,a)