The tensorboard tutorial - PR curve

1

I have read the contents of tensorboard tutorial ([https://pytorch.org/tutorials/intermediate/tensorboard_tutorial.html]

And noted that for PR curve, the tutorial adopted the prediction (i.e. tensorboard_preds) instead of label, which seems to be contradicting with the parameter setting of PR curve as set out in documentation.
[https://pytorch.org/docs/stable/_modules/torch/utils/tensorboard/writer.html#SummaryWriter.add_pr_curve].

If based on the tutorial, the curve seems strange as most of them even accuracy less than 80% are 90 degree at the right upper corner.

Any help on this matter?

Thanks.

2

Thanks for reporting this issue! It seems it’s already tracked here.