a question about loss in TwoLayerNet about cs231n-2017

[Burton2000]

Thanks for the question.

Basically from https://cs231n.github.io/linear-classify/#softmax

There is two formulas for cross entropy loss given:

The one on the left is involves calculating the softmax of the output scores, then the loss is the negative log of the correct classes softmax score.

I implemented the formula on the right which is why correct_class_scores is subtracted from the loss.

Looking at my code again it might be a bit unclear as I calculate the softmax scores for the network outputs but then dont actually use this anywhere. I might go back and change this to make it more clear.

Hope that helps you.