Python 中的多层感知器深度学习使用 Pytorch

Question

I am having errors in executing the train function of my code in MLP.

This is the error:

mat1 and mat2 shapes cannot be multiplied (128x10 and 48x10)

My code for the train function is this:

class net(nn.Module):
def __init__(self, input_dim2, hidden_dim2, output_dim2):
    super(net, self).__init__()
    self.input_dim2 = input_dim2
    self.fc1 = nn.Linear(input_dim2, hidden_dim2)
    self.relu = nn.ReLU()
    self.fc2 = nn.Linear(hidden_dim2, hidden_dim2)
    self.fc3 = nn.Linear(hidden_dim2, output_dim2) 
def forward(self, x):
  x = self.fc1(x)
  x = self.relu(x)
  x = self.fc2(x)
  x = self.relu(x)
  x = self.fc3(x) 
  x = F.softmax(self.fc3(x)) 
  
  return x



model = net(input_dim2, hidden_dim2, output_dim2) #create the network
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.RMSprop(model.parameters(), lr = learning_rate2)


def train(num_epochs2):
for i in range(num_epochs2):
  tmp_loss = []
  for (x,y) in train_loader:
    print(y.shape)
    print(x.shape)
    outputs = model(x) #forward pass
    print(outputs.shape)
    loss = criterion(outputs, y) #loss computation
    tmp_loss.append(loss.item()) #recording the loss
    optimizer.zero_grad() #all the accumulated gradient
    loss.backward()  #auto-differentiaton - accumulation of gradient
    optimizer.step() # a gradient step

  print("Loss at {}th epoch: {}".format(i, np.mean(tmp_loss)))  

I don't know where I'm wrong. My code seems to work okay.

Answer 1

From the limited message, I guess the place you are wrong are the following snippets:

x = self.fc3(x) 
x = F.softmax(self.fc3(x))

Try to replace with:

x = self.fc3(x) 
x = F.softmax(x)

---

A good question should include: error backtrace information and complete toy example which could repeat the errors!

Answer 2

Here an relu activation seems to be missing in the ' init ' function. Or there is an extra relu activation in the forward function. Look at the code below and try to figure out what is extra or missing.

def __init__(self, input_dim2, hidden_dim2, output_dim2):
    super(net, self).__init__()
    self.input_dim2 = input_dim2
    self.fc1 = nn.Linear(input_dim2, hidden_dim2)
    self.relu = nn.ReLU()
    self.fc2 = nn.Linear(hidden_dim2, hidden_dim2)
    self.fc3 = nn.Linear(hidden_dim2, output_dim2) 
def forward(self, x):
    x = self.fc1(x)
    x = self.relu(x)
    x = self.fc2(x)
    x = self.relu(x)
    x = self.fc3(x) 
    x = F.softmax(self.fc3(x)) 

return x