这是应用 softmax 的正确方法吗？

Question

self.classifier = nn.Sequential(
        nn.Flatten(),
        nn.Linear(in_features = 32*8*8, out_features = 26),
        nn.ReLU(),
        nn.Linear(in_features = 26, out_features = output_shape),
        nn.Softmax(dim=1)
    )

and my loss fn is

loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(params = model_0.parameters(),
                            lr = 0.07)

Is that the right way to use softmax? output_shape is equal to num of class (this is multi class classification)

If my implementation isn't wrong, then why do all of my data in 1 batch output the same class (even each data has very similar output probability)

Answer 1

No, CrossEntropyLoss doesn't require Softmax as it already includes it (or actually LogSoftmax ): https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html?highlight=crossentropy#torch.nn.CrossEntropyLoss .