[FIXED] How to do PyTorch F.cross_entropy?

Issue

I have output with dimension batch_size x 14 x 100 (14 object x 100 classes). I want to do cross-entropy loss with ground truth indices provided with dimension batch_size x 14. However, when I use torch.functional.cross_entropy, I get the error message that says Expected target size [15, 100], got [15, 14]. Does anyone know what the reason is? Thank you in advance

Solution

See the cross entropy documentation

For higher dim inputs, the inputs and targets are expected to be of size (N, C, d_1, ... d_k) and (N, d_1, ... d_k) where N is the batch size and C is the number of classes.

Your output should be of shape batch_size x 100 x 14 rather than batch_size x 14 x 100

Answered By - Karl