some question about the code

[Shu-HowTing]

class OrdinalLogisticModel(nn.Module):
"""
 Parameters
    ----------
    predictor : nn.Module
        When called, must return a torch.FloatTensor with shape [batch_size, 1]
"""

hello , thank u for sharing! I have a question here. why return a torch.FloatTensor with shape [batch_size, 1]. Shouldn't it be return a torch.FloatTensor with shape [batch_size, num_class - 1]?

[Shu-HowTing]

oh，sorry，maybe you are right, no matter what the value of $num\_class$, the predictor output the logit value of the sample. it should be a scalar. just like the linear regression!

[EthanRosenthal]

Yup, exactly!