Question

我正在使用以下代码初始化卷积DQN：

class ConvDQN(nn.Module):

    def __init__(self, input_dim, output_dim):
        super(ConvDQN, self).__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim
        self.conv = nn.Sequential(
            nn.Conv2d(self.input_dim, 32, kernel_size=8, stride=4),
            nn.ReLU(),
            nn.Conv2d(32, 64, kernel_size=4, stride=2),
            nn.ReLU(),
            nn.Conv2d(64, 64, kernel_size=3, stride=1),
            nn.ReLU()
        )

        self.fc_input_dim = self.feature_size()

        self.fc = nn.Sequential(
            nn.Linear(self.fc_input_dim, 128),
            nn.ReLU(),
            nn.Linear(128, 256),
            nn.ReLU(),
            nn.Linear(256, self.output_dim)
        )

    def forward(self, state):
        features = self.conv(state)
        features = features.view(features.size(0), -1)
        qvals = self.fc(features)
        return qvals

    def feature_size(self):
        return self.conv(autograd.Variable(torch.zeros(1, *self.input_dim))).view(1, -1).size(1)

这给了我错误：

  File "dqn.py", line 86, in __init__
    self.fc_input_dim = self.feature_size()
  File "dqn.py", line 105, in feature_size
    return self.conv(autograd.Variable(torch.zeros(32, *self.input_dim))).view(1, -1).size(1)
  File "C:\Users\ariji\AppData\Local\Programs\Python\Python37\lib\site-packages\torch\nn\modules\module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\Users\ariji\AppData\Local\Programs\Python\Python37\lib\site-packages\torch\nn\modules\container.py", line 92, in forward
    input = module(input)
  File "C:\Users\ariji\AppData\Local\Programs\Python\Python37\lib\site-packages\torch\nn\modules\module.py", line 489, in __call__
    result = self.forward(*input, **kwargs)
  File "C:\Users\ariji\AppData\Local\Programs\Python\Python37\lib\site-packages\torch\nn\modules\conv.py", line 320, in forward
    self.padding, self.dilation, self.groups)
RuntimeError: Expected 4-dimensional input for 4-dimensional weight [32, 4, 8, 8], but got 2-dimensional input of size [1, 4] instead

因此，我得到的事实是，我传递给卷积网络的输入的尺寸不正确。我不明白我应该如何在输入中添加所需的尺寸？还是应该在卷积网络中更改某些内容？

Answer 1

您传递了torch.zeros(1, *self.input_dim)的转换层torch.Size([1, 4])，但是将转换层初始化为，

nn.Sequential(
            nn.Conv2d(self.input_dim, 32, kernel_size=8, stride=4),
            nn.ReLU(),
            nn.Conv2d(32, 64, kernel_size=4, stride=2),
            nn.ReLU(),
            nn.Conv2d(64, 64, kernel_size=3, stride=1),
            nn.ReLU()
        )

因此，self.conv期望具有该大小的张量，但您将其传递给torch.Size([1, 4])

RuntimeError：对于4维权重[32、4、8、8]，需要4维输入，但是却得到了大小为[1、4]的2维输入

1 个答案: