Why my model is empty？

1

I am trying to build Logistic matrix factorization.When I train my model, I found that my model is empty.I am unfamiliar to PyTorch and I don’t know what causes the error.It really confuse me.Can you give me some suggestions?Thank you!
this is my code

class LMF(nn.Module):
    def __init__(self, C, R_test , n_factor):
       '''
       :param C:
       :param R_test:
       :param n_factor:
       '''
       super(LMF, self).__init__() 
       self.n_factor = n_factor
       self.C = torch.tensor(C , dtype=torch.double)
       self.R = torch.tensor(R_test , dtype=torch.double)
       self.n_user = C.shape[0]
       self.n_item = C.shape[1]
       self.X = Parameter(torch.randn([self.n_user, self.n_factor], dtype=torch.double))
       self.Y = Parameter(torch.randn([self.n_item, self.n_factor], dtype=torch.double))
       self.user_biases = Parameter(torch.randn((self.n_user, 1), dtype=torch.double))
       self.item_biases = Parameter(torch.randn((self.n_item, 1), dtype=torch.double))

    def forward(self):

        A=torch.mm(self.X,self.Y.T)
        A=A+self.user_biases
        A=A+self.item_biases.T
        loss=self.C*(A) - (1+self.C)*(torch.log((1+torch.exp(A))))
        loss=loss.sum()
        return -loss

    def MPR(self):
        rank = 0.0
        R_sum = torch.sum(self.R)
        R_hat=torch.mm(self.X , self.Y.T)
        R_hat_rank = torch.argsort(torch.argsort(-R_hat, dim=1))
        print(self.R.size())
        print(R_hat_rank .size())
        A = self.R*(R_hat_rank / torch.tensor(self.n_item,dtype=torch.double))
        rank = torch.sum(A)/ R_sum
        return rank.item()

image
image1086×127 1.08 KB

2

print(model) will print the registered modules, which are not present in your current implementation.
You could create the extra_repr method, which would print additional information about your custom module, similar to this one.

To print the parameters, you could use:

for name, param in model.named_parameters():
    print(name, param)

Note that tensors will not be registered inside the module, and won’t be returned by model.parameters(), model.buffers(), or model.state_dict().
To register tensors as buffers, use self.register_buffer(name, tensor).

3

thank you very much!