如何解决以下CNN架构的错误"mat1 and mat2 shapes cannot be multiplied"?

How to resolve the error "mat1 and mat2 shapes cannot be multiplied" for the following CNN architecture?

我正在尝试使用批量归一化实现 Conv1d 模型,但出现错误:

RuntimeError                              Traceback (most recent call last)
<ipython-input-117-ef6e122ea50c> in <module>()
----> 1 test()
      2 for epoch in range(1, n_epochs + 1):
      3   train(epoch)
      4   test()

7 frames
/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py in linear(input, weight, bias)
   1751     if has_torch_function_variadic(input, weight):
   1752         return handle_torch_function(linear, (input, weight), input, weight, bias=bias)
-> 1753     return torch._C._nn.linear(input, weight, bias)
   1754 
   1755 

RuntimeError: mat1 and mat2 shapes cannot be multiplied (32x140 and 100x10)

我使用的批量大小为 32,数据的特征数量为 40。我一直在尝试计算 32 x 140 来自哪里,但我无法做到这一点.这是我尝试使用的 CNN 的架构:

class MyModel(nn.Module):
    def __init__(self):
        super(MyModel, self).__init__()
        #self.flatten=nn.Flatten()
        self.net_stack=nn.Sequential(
            nn.Conv1d(in_channels=1, out_channels=25, kernel_size=5, stride=2), #applying batch norm
            nn.ReLU(),
            nn.BatchNorm1d(25, affine=True),
            nn.Conv1d(in_channels=25, out_channels=20, kernel_size=5, stride=2), #applying batch norm
            nn.ReLU(),
            nn.BatchNorm1d(20, affine=True),
            nn.Flatten()
            nn.Linear(20*5, 10),
            nn.Softmax(dim=1))

    def forward(self,x):
        # result=self.net_stack(x[None])
        result=self.net_stack(x[:, None, :])
        return result

这个完全连接的应该从:

nn.Linear(20*5, 10)

至:

nn.Linear(20*7, 10)

为什么?

如果你的输入数据长度是40,那么(B是batch size):

  • 第一次转换 (K=25) 后的输出:B x 25 x 18
  • 第二次转换 (K=20) 后的输出:B x 20 x 7
  • nn.Flatten()后的输出:B x 140,即如果B=32,则32x140