运行实现：
import torch
from torch.utils.data import DataLoader
from torchvision import transforms
from torchvision.datasets import MNIST
import matplotlib.pyplot as pltclass Net(torch.nn.Module):#net类神经网络主体def __init__(self):#4个全链接层super().__init__()self.fc1 = torch.nn.Linear(28*28, 64)#输入为28*28尺寸图像self.fc2 = torch.nn.Linear(64, 64)#中间三层都是64个节点self.fc3 = torch.nn.Linear(64, 64)self.fc4 = torch.nn.Linear(64, 10)#输出为10个数字类别def forward(self, x):#前向传播x = torch.nn.functional.relu(self.fc1(x))#先全连接线性计算，再套上激活函数x = torch.nn.functional.relu(self.fc2(x))x = torch.nn.functional.relu(self.fc3(x))x = torch.nn.functional.log_softmax(self.fc4(x), dim=1)#输出层用softmax做归一化，log_softmax是为了提高计算稳定性，套上了一个对数函数return xdef get_data_loader(is_train):#导入数据to_tensor = transforms.Compose([transforms.ToTensor()])#导入张量data_set = MNIST("", is_train, transform=to_tensor, download=True)#下载文件，”“里面对应的是下载目录，is_train指定导入训练集还是测试集return DataLoader(data_set, batch_size=15, shuffle=True)#一个批次15张图片，shuffle=true说明数据是随机打乱的，返回数据加载器def evaluate(test_data, net):#评估正确率n_correct = 0n_total = 0with torch.no_grad():for (x, y) in test_data:#取出数据outputs = net.forward(x.view(-1, 28*28))#计算神经网络预测值for i, output in enumerate(outputs):#作比较if torch.argmax(output) == y[i]:#argmax取最大预测概率的序号n_correct += 1#累加正确的n_total += 1return n_correct / n_totaldef main():train_data = get_data_loader(is_train=True)#训练集test_data = get_data_loader(is_train=False)#测试集net = Net()print("initial accuracy:", evaluate(test_data, net))#打印初始网络的正确率，接近0.1optimizer = torch.optim.Adam(net.parameters(), lr=0.001)#以下为pytorch固定写法for epoch in range(3):#epoch是轮次for (x, y) in train_data:net.zero_grad()#初始化output = net.forward(x.view(-1, 28*28))#正向传播loss = torch.nn.functional.nll_loss(output, y)#计算差值，null_loss对数损失函数，为了匹配前面log_softmax的对数运算loss.backward()#反向误差传播optimizer.step()#优化网络参数print("epoch", epoch, "accuracy:", evaluate(test_data, net))for (n, (x, _)) in enumerate(test_data):#抽取4张图像，显示预测结果if n > 3:breakpredict = torch.argmax(net.forward(x[0].view(-1, 28*28)))plt.figure(n)plt.imshow(x[0].view(28, 28),cmap='gray')plt.title("prediction: " + str(int(predict)))plt.show()if __name__ == "__main__":main()中间可能会报错误：（libiomp5md.dll问题）
OMP: Error #15: Initializing libiomp5md.dll, but found libiomp5md.dll already initialized.这个处理就是在anaconda文件夹下面搜索libiomp5md.dll，那bin下面的 libiomp5md.dll文件全部修改命名，就像我这样，两个bin文件夹下面的都改了。
运行结果：
两轮精确度如下：
4个数字预测图片如下：
