AUC & ROC for MultiClass Classification CNN

[yrouphail]

I’ve seen a few topics about this, but none of them have came to a working conclusion. I’m trying to get an AUC for each class as simply inserting AUROC() as a metric did not work. I know that scikit learn has a workingmulticlass/multilabel roc, but I do not know how to implement that into a fastai learner. I was wondering and hoping if anyone knew of an efficient way to calculate the AUC, even if its not as a metric. Thanks in advance!

[muellerzr (Zachary Mueller)]

Right here:

AUC for multiclass classification

Adapted Using AUC as metric in fastai for multiclass from sklearn.metrics import roc_auc_score def auroc_score(input, target): input, target = input.cpu().numpy(), target.cpu().numpy() return roc_auc_score(target, input) @dataclass class AUC(Callback): "Computes the auc for multi-class classification" clas:int=1 def on_epoch_begin(self, **kwargs): self.output, self.target = [], [] def on_batch_end(self, last_output, last_target, train, **kwargs): if …

[yrouphail]

Gave it a shot, and got the following error: