从零实现多层感知机

import torch
from torch import nn
from d2l import torch as d2l

# 批量大小
batch_size = 256
# 训练集和测试集
train_iter, test_iter = d2l.load_data_fashion_mnist(batch_size)
# 输入特征维度、输出特征维度、隐藏层维度
num_inputs, num_outputs, num_hiddens = 784, 10, 256

# W1矩阵为输入特征维度×隐藏层维度
W1 = nn.Parameter(torch.randn(num_inputs, num_hiddens, requires_grad=True) * 0.01)
# b1偏置向量为隐藏层维度
b1 = nn.Parameter(torch.zeros(num_hiddens, requires_grad=True))
# W2矩阵为隐藏层维度×输出特征维度
W2 = nn.Parameter(torch.randn(num_hiddens, num_outputs, requires_grad=True) * 0.01)
# b2偏置向量为输出特征维度
b2 = nn.Parameter(torch.zeros(num_outputs, requires_grad=True))
# 网络中用到的所有参数打包
params = [W1, b1, W2, b2]

# 定义relu函数
def relu(X):
    print(X)
    a = torch.zeros_like(X)
    return torch.max(X, a)
# 定义网络模型
def net(X):
    X = X.reshape((-1, num_inputs))
    H = relu(X@W1 + b1)  # 这里“@”代表矩阵乘法
    return (H@W2 + b2)
# 定义交叉熵损失函数
loss = nn.CrossEntropyLoss(reduction='none')
# 定义迭代次数、学习率
num_epochs, lr = 10, 0.1
# 定义小批量梯度下降算法
updater = torch.optim.SGD(params, lr=lr)

# 训练过程
d2l.train_ch3(net, train_iter, test_iter, loss, num_epochs, updater)
# 测试过程
d2l.predict_ch3(net, test_iter)