Question

在尝试建立pytorch模型时，我收到以下错误：尝试进行Pytorch autograd时，损失对象不可调用。（相关代码如下所示）

optimizer = torch.optim.Adam(model.parameters(), lr=lr, 
  betas(0.0,0.9))

def train(epoch, shuffle, wisdom_model, optim, loss):
    print('train')
    accuracy = 0
    batch_num = 0
    wisdom_model.train()
    for batch in data.train_dl:

        optim.zero_grad()

        result = model(batch[0])
        loss = nn.CrossEntropyLoss()(result, batch[1].long())

        loss.backward()

        accuracy += accuracy(result, batch[1])
        print(accuracy)
        pdb.set_trace()
        batch_num += 1

    return accuracy / batch_num

TypeError                                 Traceback (most recent call last)
<ipython-input-28-5b9c9fe3b320> in <module>
----> 1 run(1, False)

<ipython-input-27-d0d67dbf6eb2> in run(num_models, dropout)
     71 
     72     for epoch in range(10):
---> 73         train_accuracy = train(epoch, False, model, optimizer, loss)
     74         accuracy.append(validate(epoch, model))
     75 

<ipython-input-27-d0d67dbf6eb2> in train(epoch, shuffle, model, optim, loss)
     24         pdb.set_trace()
     25 
---> 26         loss.backward()
     27         optim.step()
     28 

TypeError: 'int' object is not callable

Answer 1

问题出在这一行：

loss = nn.CrossEntropyLoss()(result, batch[1].long())

签出nn.CrossEntropyLoss。

看起来不应该这样：

nn.CrossEntropyLoss()()

应如下所示：

nn.CrossEntropyLoss()

Answer 2

问题可能出在您目标的数据类型，即batch [1]。检查它是否为Tensor（）类型。一个简单的torch.tensor（batch [1]）就可以完成这项工作。

TypeError：“ int”对象在loss.backward（）

2 个答案: