在 pytorch 中绘制训练和验证损失图

Question

我正在使用 pytorch 来训练我的 CNN 网络。我想绘制训练和验证损失曲线以可视化模型性能。如何绘制两条曲线？

我有下面的代码

# create a function (this my favorite choice)
def RMSELoss(predicted,target):
    return torch.sqrt(torch.mean((predicted-target)**2))

criterion = RMSELoss

# loss = torch.sqrt(criterion(x, y))
optimizer = torch.optim.Adam(model.parameters(), lr=0.0001)
epochs = 300

n_total_steps = len(train_dataset)

trainingEpoch_loss = []
validationEpoch_loss = []

for epoch in range(epochs):
    step_loss = []
    model.train()
    for i, data in enumerate(train_dataset):
        feature,target = data['data'].type(torch.FloatTensor),torch.tensor(data['target']).type(torch.FloatTensor)
         
        # Clear the gradients
        optimizer.zero_grad()
        # Forward Pass
        outputs = model(feature)
        # Find the Loss
        training_loss = criterion(outputs, target)
        # Calculate gradients
        training_loss.backward()
        # Update Weights
        optimizer.step()
        # Calculate Loss
        step_loss.append(training_loss.item())
        if (i+1) % 1 == 0:
            print (f'Epoch [{epoch+1}/{epochs}], Step [{i+1}/{n_total_steps}], Loss: {training_loss.item():.4f}')
    trainingEpoch_loss.append(np.array(step_loss).mean())
 
    model.eval()     # Optional when not using Model Specific layer
    for i, data in enumerate(val_dataset):
        validationStep_loss = []
        feature,target = data['data'].type(torch.FloatTensor),torch.tensor(data['target']).type(torch.FloatTensor)
        
        # Forward Pass
        outputs = model(feature)
        # Find the Loss
        validation_loss = criterion(outputs, target)
        # Calculate Loss
        validationStep_loss.append(validation_loss.item())
    validationEpoch_loss.append(np.array(validationStep_loss).mean())

你能让我知道我做得对吗？另请让我知道如何绘制训练和验证损失？

Answer 1

您将 epoch 损失收集到trainingEpoch_loss和validationEpoch_loss列表中是正确的。现在，训练结束后，添加代码来绘制损失：

from matplotlib import pyplot as plt
plt.plot(trainingEpoch_loss, label='train_loss')
plt.plot(validationEpoch_loss,label='val_loss')
plt.legend()
plt.show

阅读 matplotlib 文档以获取更多奇特的绘图功能。