Backend: Try out new loss functions #33
Description
Evaluate and compare the model on L1, BCEwithLogits, and macro F1 score for train, val and test accuracies. My suspicion is that the current macro F1 is too coarse for a detailed accuracy evaluation.