我有一段代码（来自这里），用于分类MINST数字。代码运行正常。他们使用了CrossEntropyLoss和Adam优化器。

模型代码如下

class CNN(nn.Module):    def __init__(self):        super(CNN, self).__init__()        self.conv1 = nn.Sequential(                     nn.Conv2d(                in_channels=1,                              out_channels=16,                            kernel_size=5,                              stride=1,                                   padding=2,                              ),                                          nn.ReLU(),                                  nn.MaxPool2d(kernel_size=2),            )        self.conv2 = nn.Sequential(                     nn.Conv2d(16, 32, 5, 1, 2),                 nn.ReLU(),                                  nn.MaxPool2d(2),                        )        # fully connected layer, output 10 classes        self.out = nn.Linear(32 * 7 * 7, 10)        # self.softmax = torch.nn.Softmax(dim=1)    def forward(self, x):        x = self.conv1(x)        x = self.conv2(x)        # flatten the output of conv2 to (batch_size, 32 * 7 * 7)        x = x.view(x.size(0), -1)               output = self.out(x)        # output = self.softmax(output)        return output, x    # return x for visualizationThe shape of the `b_x` and `b_y` is

torch.Size([100, 1, 28, 28]) torch.Size([100])

现在，我希望从输出层获得连续值。比如，我希望输出类似于1.0、0.9、8.6、7.0等。如果输出层的值是1.0且标签是1，那就意味着预测是完美的。否则，就是不完美的。更简单地说，我想把MNIST数字视为一个回归问题。

因此，我将损失函数改成了MSELoss，优化器改成了SGD（代码的其余部分与网站上的保持一致）。但现在，我遇到了一个错误

/home/Opps_0/.local/lib/python3.8/site-packages/torch/nn/modules/loss.py:528: UserWarning: Using a target size (torch.Size([100])) that is different to the input size (torch.Size([100, 10])). This will likely lead to incorrect results due to broadcasting. Please ensure they have the same size.  return F.mse_loss(input, target, reduction=self.reduction)Traceback (most recent call last):  File "/usr/lib/python3.8/runpy.py", line 194, in _run_module_as_main    return _run_code(code, main_globals, None,  File "/usr/lib/python3.8/runpy.py", line 87, in _run_code    exec(code, run_globals)  File "/home/Opps_0/Desktop/MNIST/src/train.py", line 60, in <module>    train(NB_EPOCS, model, loaders)  File "/home/Opps_0/Desktop/MNIST/src/train.py", line 45, in train    loss = criterion(output, b_y)  File "/home/Opps_0/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl    result = self.forward(*input, **kwargs)  File "/home/Opps_0/.local/lib/python3.8/site-packages/torch/nn/modules/loss.py", line 528, in forward    return F.mse_loss(input, target, reduction=self.reduction)  File "/home/Opps_0/.local/lib/python3.8/site-packages/torch/nn/functional.py", line 2925, in mse_loss    expanded_input, expanded_target = torch.broadcast_tensors(input, target)  File "/home/Opps_0/.local/lib/python3.8/site-packages/torch/functional.py", line 74, in broadcast_tensors    return _VF.broadcast_tensors(tensors)  # type: ignoreRuntimeError: The size of tensor a (10) must match the size of tensor b (100) at non-singleton dimension 1

你能告诉我如何修改才能从输出层获得连续值吗？

回答：

假设你的目标形状为(batch_size,)，类似于以下内容：

 >>> model = CNN() >>> criterion = nn.MSELoss() >>> output, _ = model(torch.rand(2, 1, 28, 28)) >>> b_y = torch.randint(0, 10, (2,)) tensor([1, 2, 6, 5, 7])

使用MSELoss计算损失会导致：

 >>> loss = criterion(output, b_y)

RuntimeError: The size of tensor a (10) must match the size of tensor b (2) at non-singleton dimension 1.

这意味着你的目标b_y的形状不正确，它需要匹配output的形状，即它需要是一个二维张量。

由于你正在用回归损失优化这个任务，你可以将你的目标编码为稀疏向量，也称为独热编码。你可以使用内置的torch.nn.functional.one_hot轻松做到这一点：

>>> ohe_target = torch.nn.functional.one_hot(b_y, num_classes=10)tensor([[0, 1, 0, 0, 0, 0, 0, 0, 0, 0],        [0, 0, 1, 0, 0, 0, 0, 0, 0, 0],        [0, 0, 0, 0, 0, 0, 1, 0, 0, 0],        [0, 0, 0, 0, 0, 1, 0, 0, 0, 0],        [0, 0, 0, 0, 0, 0, 0, 1, 0, 0]])

现在你可以正确地计算损失：

>>> criterion(output, ohe_target)tensor(0.1169, grad_fn=<MseLossBackward>)