深度学习：隐含层的反向传播

import torch import torchvision import torch.nn.functional as F from torch.utils import data as torchData import torchvision.transforms as transforms from Accumulat import Accumulator from util import * def init_weights(m): if type(m) == torch.nn.Linear: torch.nn.init.normal_(m.weight, std=0.01) # 定义网络模型 class LinearNet(torch.nn.Module): def __init__(self,num_inputs, hidden_num, class_num): super().__init__() self.linear1 = torch.nn.Linear(num_inputs, hidden_num) self.linear2 = torch.nn.Linear(hidden_num, class_num) self.linear1.apply(init_weights) self.linear2.apply(init_weights) self.flatten = torch.nn.Flatten() self.relu = torch.nn.ReLU() def forward(self, x): x = self.flatten(x) y = self.linear1(x) y = self.relu(y) y = self.linear2(y) return y def load_FashionMNIST(batch_size, resize=None): trans = transforms.ToTensor() # if resize: # trans.insert(0,transforms.Resize(resize)) # trans = transforms.Compose(trans) mnist_train = torchvision.datasets.FashionMNIST( root="../data/", train=True, transform=trans, download=True) mnist_test = torchvision.datasets.FashionMNIST( root="../data/", train=False, transform=trans, download=True) train_loader = torchData.DataLoader(mnist_train, batch_size, shuffle=True, num_workers=4) test_loader = torchData.DataLoader(mnist_test, 1500, shuffle=False, num_workers=4) return train_loader, test_loader def train_epoch_ch3(net, train_iter,loss_fn, optimizer): """训练模型一个迭代周期（定义见第3章）。""" # 训练损失总和、训练准确度总和、样本数 metric = Accumulator(3) for x,y in train_iter: # 计算梯度并更新参数 optimizer.zero_grad() y_hat = net(x) softmax_y = F.softmax(y_hat, dim=1) loss = loss_fn(y_hat, y) loss.backward() optimizer.step() train_acc = accuracy(y_hat, y) metric.add(float(loss), train_acc, y.numel()) # 返回训练损失和训练准确率 return metric[0] / metric[2], metric[1] / metric[2] def train_ch3(net, train_iter, test_iter, num_epochs, loss_fn, optimizer): """训练模型（定义见第3章）。""" for epoch in range(num_epochs): train_metrics = train_epoch_ch3(net, train_iter,loss_fn, optimizer) test_acc = evaluate_accuracy(net, test_iter) print(f'epoch:{epoch}, train loss is:{train_metrics[0]:.5f}, train acc is:{train_metrics[1]:.3f}, test acc is:{test_acc:f}') train_loss, train_acc = train_metrics assert train_loss < 0.5, train_loss assert train_acc <= 1 and train_acc > 0.7, train_acc assert test_acc <= 1 and test_acc > 0.7, test_acc if __name__ == '__main__': batch_size = 10 train_iter, test_iter = load_FashionMNIST(batch_size=batch_size) # 定义损失函数 和 优化器 num_inputs = 784 hidden_num = 256 class_num = 10 net = LinearNet(num_inputs, hidden_num, class_num) loss_fn = torch.nn.CrossEntropyLoss() # pytorch的交叉熵损失计算里已经包含了 softmax，所以在forward里面不应该加上 softmax optimizer = torch.optim.SGD(net.parameters(), lr=0.1) num_epochs = 30 lr = 0.1 train_ch3(net, train_iter, test_iter, num_epochs, loss_fn, optimizer) print('ok')