关于#神经网络#的问题：RuntimeError: mat1 and mat2 shapes cannot be multiplied (64x64 and 128x64)

Question

RuntimeError: mat1 and mat2 shapes cannot be multiplied (64x64 and 128x64)
近期在调试一个四层的BP神经网络，报出了这样的错误：

大概知道问题出在这里：

    def forward(self, x):
        x = F.relu(self.hidden_1(x))
        x = self.dropout(self.bn1(x))
        x = F.relu(self.hidden_2(x))
        x = self.dropout(self.bn2(x))
        x = F.relu(self.hidden_3(x))
        x = self.dropout(self.bn3(x))
        x = F.relu(self.hidden_4(x))%问题所在
        x = self.dropout(self.bn4(x))
        x = self.out(x)
        return x

目前已经尝试过添加

x = torch.flatten(x,1)

不能解决/(ㄒoㄒ)/~~，感觉也不是池化层输出形状的问题叭，因为打印在def forward(self, x)中print(x.shape)输出是([64,64])，唉，神经网络小菜请教如何解决🙇‍
下面贴出相关代码：

class Classifier(torch.nn.Module):
    def __init__(self, n_feature, n_hidden, n_output, dropout=0.5):
        super(Classifier, self).__init__()
        self.dropout = torch.nn.Dropout(dropout)

        self.hidden_1 = torch.nn.Linear(n_feature, n_hidden)
        self.bn1 = torch.nn.BatchNorm1d(n_hidden)

        self.hidden_2 = torch.nn.Linear(n_hidden, n_hidden//4)
        self.bn2 = torch.nn.BatchNorm1d(n_hidden//4)

        self.hidden_3 = torch.nn.Linear(n_hidden//4, n_hidden//8)
        self.bn3 = torch.nn.BatchNorm1d(n_hidden//8)
        #三层卷积与四层卷积
        self.hidden_4 = torch.nn.Linear(n_hidden // 4, n_hidden // 8)
        self.bn4 = torch.nn.BatchNorm1d(n_hidden // 8)

        self.out = torch.nn.Linear(n_hidden//8, n_output)

    def forward(self, x):
        x = F.relu(self.hidden_1(x))
        x = self.dropout(self.bn1(x))
        x = F.relu(self.hidden_2(x))
        x = self.dropout(self.bn2(x))
        x = F.relu(self.hidden_3(x))
        x = self.dropout(self.bn3(x))
        x = F.relu(self.hidden_4(x))
        x = self.dropout(self.bn4(x))
        x = self.out(x)
        return x

 for i, data in enumerate(train_loader):
            optimizer.zero_grad() 
            train_pred = model(data[0])
            batch_loss = loss(train_pred, data[1]) 

            batch_loss.backward()  
            optimizer.step()  

            train_acc += np.sum(np.argmax(train_pred.cpu().data.numpy(), axis=1) == data[1].numpy())
            train_loss += batch_loss.item()

Answer 1

问题在后面的 model(data[0])，data 和 x 不匹配，设置错误.

可以打印和追踪矩阵的形状，是开始设置时就不匹配，还是中间被修改了，在哪里发生修改的。