玫琳凯网站建设与推广方案 天堂在线搜索

1.Tensorboard

 

 

2.Tensorboard的使用

导入tensorboard并创建SummaryWriter

 

添加标量数据 

 

 

添加图片数据

 

 

 直方图

模型可视化

 

 

3.代码使用

模型可视化

# 记录模型结构
dummy_input = torch.randn(1, 3, 224, 224).to(device)  # 根据你的模型输入尺寸调整
writer.add_graph(model, dummy_input)

训练loss 

for epoch in range(num_epochs):# 训练train_loss = train(model, train_loader, criterion, optimizer, device)# 验证val_loss = validate(model, val_loader, criterion, device)# 记录训练和验证的损失值到TensorBoardwriter.add_scalar('Training Loss', train_loss, epoch)writer.add_scalar('Validation Loss', val_loss, epoch)print(f'Epoch [{epoch+1}/{num_epochs}], Training Loss: {train_loss:.4f}, Validation Loss: {val_loss:.4f}')print('Finished Training')# 关闭SummaryWriter
writer.close()

 模型数据分布

for name, param in model.named_parameters():writer.add_histogram(name, param.clone().cpu().data.numpy(), epoch)writer.add_histogram(f'{name}.grad', param.grad.clone().cpu().data.numpy(), epoch)