import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torchvision
import matplotlib.pyplot as plt
import numpy as np

input_size = 28
num_classes = 10
num_epochs = 40
batch_size = 64

# 确保 CUDA 可用并且使用它
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

train_dataset = torchvision.datasets.MNIST(
    root='./data', train=True, transform=torchvision.transforms.ToTensor(), download=True
)
test_dataset = torchvision.datasets.MNIST(root='./data', train=False, transform=torchvision.transforms.ToTensor())

train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                            batch_size=batch_size,
                                            shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
                                           batch_size=batch_size,
                                           shuffle=True)

#定义了一个CNN类作为神经网络模型
# in_channels：输入数据的通道数。在第一层卷积中，输入的MNIST图像是灰度图像，因此 in_channels=1。
# out_channels：输出的特征图通道数，即卷积核的数量。每个卷积核生成一个特征图。
# kernel_size：卷积核的大小。例如，kernel_size=5 表示卷积核大小为5x5。
# stride：卷积核的步幅。在代码中设置为1，表示卷积核每次移动1个像素。
# padding：填充的像素数。在代码中设置为2，表示在输入数据周围填充2个像素。
class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.conv1 = nn.Sequential(
            nn.Conv2d(1, 16, 5, 1, 2),
            nn.ReLU(),
            nn.MaxPool2d(2)
        )
        self.conv2 = nn.Sequential(
            nn.Conv2d(16, 32, 5, 1, 2),
            nn.ReLU(),
            nn.MaxPool2d(2)
        )
        self.conv3 = nn.Sequential(
            nn.Conv2d(32, 64, 5, 1, 2),
            nn.ReLU(),
            nn.MaxPool2d(2)
        )
        self.conv4 = nn.Sequential(
            nn.Conv2d(64, 32, 5, 1, 2),
            nn.ReLU(),
            nn.MaxPool2d(2)
        )
        self.out = nn.Linear(32, num_classes)  # 确保输出类别数正确

    def forward(self, x):
        x = self.conv1(x).to(device)  # 将数据迁移到GPU
        x = self.conv2(x).to(device)
        x=self.conv3(x).to(device)
        x = self.conv4(x).to(device)
        x = x.view(x.size(0), -1).to(device)
        x = self.out(x).to(device)
        return x

def accuracy(prediction, label):
    pred = torch.max(prediction.data, 1)[1]
    rights = pred.eq(label.data.view_as(pred)).sum()
    return rights, len(label)

net = CNN().to(device)  # 将模型迁移到GPU
criterion = nn.CrossEntropyLoss().to(device)  # 将损失函数迁移到GPU
optimizer = optim.Adam(net.parameters(), lr=0.001)

tr = []
for epoch in range(40):
    train_rights = []
    for batch_id, (data, target) in enumerate(train_loader):
        data, target = data.to(device), target.to(device)  # 将数据迁移到GPU
        optimizer.zero_grad()
        output = net(data)
        loss = criterion(output, target)
        loss.backward()
        optimizer.step()
        right = accuracy(output, target)
        train_rights.append(right)
        tr.append(loss)
        if batch_id % 100 == 0:
            val_right = []
            for (data, target) in test_loader:
                data, target = data.to(device), target.to(device)  # 将数据迁移到GPU
                output = net(data)
                right = accuracy(output, target)
                val_right.append(right)
            train_r = (sum(t[0] for t in train_rights), sum(t[1] for t in train_rights))
            val_r = (sum(t[0] for t in val_right), sum(t[1] for t in val_right))
            print("train_acc {}, test_acc {}".format(train_r[0] / train_r[1], val_r[0] / val_r[1]))

# 注意，绘图时不需要将数据迁移到GPU
plt.plot(list(range(len([tensor.detach().cpu().numpy() for tensor in tr]))), [tensor.detach().cpu().numpy() for tensor in tr])
plt.show()
torch.save(net.state_dict(), 'model.pth')