Tensorboard tutorial add_pr_curve( )

[kanru-wang]

In https://pytorch.org/tutorials/intermediate/tensorboard_tutorial.html#, section 6, Assessing trained models with TensorBoard, there are these two lines.

    class_probs.append(class_probs_batch)
    class_preds.append(class_preds_batch)

Shouldn't the second line be

    class_preds.append(labels)

instead?

because later add_pr_curve takes ground truth labels instead of predicted labels.

[WYBupup]

I have found this problem, too. And I change the second parameter of writer.add_pr_curve() from “test_preds == class_index” to “testset.targets == class_index”. And I got terrible percision-Recall curve for each class( one of the pr_curve was showed in the following picture). I just wonder if you have corrected the error in the tutorial and run the code you corrected. If you have, was the pr curve you got same as mine?
I am looking forward to your "response!"

[WYBupup]

As for the terrible result of pr_curve I got after I changing “test_preds==class_index” to “test_preds == testset.targets”, the reason is the order of testset.targets doesn’t match tensorboard_probs’ , since the later one was got from testloader which was created by shuffling the testset. Instead, I changed the code “class_preds.append(class_preds_batch)” to “class_preds.append(labels)” like what you said. I got pretty reasonable pr_curve.(showed in the picture)

[dwang55]

Guys, I found the same thing. You crroect the code the same way just as me. I didn't check the github, so I opend a topic on Pytorch forum before.
It should be a code error in the tutorial, thank you for make me sure for that.

[holly1238]

This issue is fixed and merged in #1469