TensorBoard在pytorch训练过程中如何使用,及数据读取问题解决方法

模块导入

首先从torch中导入tensorboard的SummaryWriter日志记录模块

from torch.utils.tensorboard import SummaryWriter

然后导入要用到的os库，当然你们也要导入自己模型训练需要用到的库

import os

日志记录文件的创建

import os

log_dir = 'runs/EfficientNet_B3_experiment2'

# 检查目录是否存在
if os.path.exists(log_dir):
    # 如果目录存在，获取目录下的所有文件和子目录列表
    files = os.listdir(log_dir)

    # 遍历目录下的文件和子目录
    for file in files:
        # 拼接文件的完整路径
        file_path = os.path.join(log_dir, file)

        # 判断是否为文件
        if os.path.isfile(file_path):
            # 如果是文件，删除该文件
            os.remove(file_path)
        elif os.path.isdir(file_path):
            # 如果是目录，递归地删除目录及其下的所有文件和子目录
            for root, dirs, files in os.walk(file_path, topdown=False):
                for name in files:
                    os.remove(os.path.join(root, name))
                for name in dirs:
                    os.rmdir(os.path.join(root, name))
            os.rmdir(file_path)

# 创建新的SummaryWriter
writer = SummaryWriter(log_dir)

这个代码会自动创建并更新日志文件目录，请谨慎使用，记得改
log_dir = 'runs/EfficientNet_B3_experiment2'路径名字小心把之前保存好的数据删除了
之后模型训练的数据将会写入到log_dir这个路径文件中，在由TensorBoard张量板调用显示数据

训练中如何写入数据

for epoch in range(num_epochs):
    model.train()
    running_loss = 0.0
    correct = 0
    total = 0
    start_time = time.time()

    for images, labels in train_loader:
        images, labels = images.to(device), labels.to(device)
        optimizer.zero_grad()
        outputs = model(images)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

        # 记录学习率
        current_lr = optimizer.param_groups[0]['lr']
        writer.add_scalar('Learning Rate', current_lr, epoch)

        # 记录梯度范数
        total_norm = 0
        for p in model.parameters():
            param_norm = p.grad.data.norm(2)
            total_norm += param_norm.item() ** 2
        total_norm = total_norm ** 0.5
        writer.add_scalar('Gradient Norm', total_norm, epoch)

        running_loss += loss.item()
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

    train_loss = running_loss / len(train_loader)
    train_accuracy = 100 * correct / total

    # 记录训练损失和准确率
    writer.add_scalar('Training Loss', train_loss, epoch)
    writer.add_scalar('Training Accuracy', train_accuracy, epoch)

    # 记录模型参数的直方图
    for name, param in model.named_parameters():
        writer.add_histogram(name, param, epoch)

    # 记录网络结构（通常只需要记录一次）
    if epoch == 0:
        writer.add_graph(model, images.to(device))

    # 记录输入图片
    img_grid = torchvision.utils.make_grid(images)
    writer.add_image('train_images', img_grid, epoch)

    # 使用matplotlib记录渲染的图片
    fig, ax = plt.subplots()
    ax.plot(np.arange(len(labels)), labels.cpu().numpy(), 'b', label='True')
    ax.plot(np.arange(len(predicted)), predicted.cpu().numpy(), 'r', label='Predicted')
    ax.legend()
    writer.add_figure('predictions vs. actuals', fig, epoch)
        # 验证模型
    model.eval()
    val_loss = 0.0
    correct = 0
    total = 0
    
    all_preds = []
    all_labels = []
    
    with torch.no_grad():
        for images, labels in test_loader:
            images, labels = images.to(device), labels.to(device)
            outputs = model(images)
            loss = criterion(outputs, labels)
            val_loss += loss.item()
            _, predicted = torch.max(outputs.data, 1)
            all_preds.extend(predicted.cpu().numpy())
            all_labels.extend(labels.cpu().numpy())
            total += labels.size(0)
            correct += (predicted == labels).sum().item()

    val_loss /= len(test_loader)
    val_accuracy = 100 * correct / total

    if val_accuracy > best_val_accuracy:
        # 当新的最佳验证准确率出现时，保存模型状态字典
        best_val_accuracy = val_accuracy
        best_model_state_dict = model.state_dict()

    # 记录验证损失和准确率
    writer.add_scalar('Validation Loss', val_loss, epoch)
    writer.add_scalar('Validation Accuracy', val_accuracy, epoch)

    # 记录多条曲线
    writer.add_scalars('Loss', {'train': train_loss, 'val': val_loss}, epoch)
    writer.add_scalars('Accuracy', {'train': train_accuracy, 'val': val_accuracy}, epoch)

    # 打印每个epoch的训练和验证结果
    print(f'Epoch [{epoch+1}/{num_epochs}], '
          f'Train Loss: {train_loss:.4f}, Train Accuracy: {train_accuracy:.2f}%, '
          f'Validation Loss: {val_loss:.4f}, Validation Accuracy: {val_accuracy:.2f}%, '
          f'Time: {time.time() - start_time:.2f}s')

以上代码分别记录了

如何提取保存的数据调用TensorBoard面板

在终端输入以下代码

tensorboard --logdir='修改为自己的log_dir路径'

然后点击 http://localhost:6006/就可以成功加载面板了

可能会遇到的问题

如果数据读取失败那么请检查数据路径是否正确
注意数据文件中不能有任何中文