Loss function for multi-class classification

fastai
1

Hello,

I currently finished up on the digit recognition tutorial and decided to work with a 28x28 image set for classifying 10 different items of clothing.

My code was working fine until I reached the Learner stage and encountered the following error:

image
image1581×839 65.9 KB

How do I proceed to perform multi-class classification for the same, should I convert the labels into one-hot vectors and proceed to do so?

Thank you

2

You need to change your metrics and losses. Use CrossEntropyLossFlat and xxx_multi. For example, accuracy_multi instead of accuracy