Questions about the evaluation~

[jiaweiz716]

Thank you for your great work!

I noticed that you save the evaluation results every 50 epochs, as follows

eval_epochs = list(range(0, 5000, 50))
if verbose and it in eval_epochs:

    res = []
    runs = 1 if args.dataset in ['ogbn-arxiv'] else 3
    for i in range(runs):
        if args.dataset in ['ogbn-arxiv']:
            res.append(self.test_with_val())
        else:
            res.append(self.test_with_val())

    res = np.array(res)
    print('Train/Test Mean Accuracy:',
            repr([res.mean(0), res.std(0)]))

However, I am not sure which epoch result you will use as the final report result in the paper.

Can you give me some hints (maybe early stopping, but I didn't find the setting of early stopping strategy in the code)?

[Suchun-sv]

Thanks for your attention to our work and sorry for the late response.

It depends on the maximum epochs setting (it is a hyper-parameter) we use the last report as the final results. We fix some bugs these days and soon release the new version with the evaluation script. Hope that could help you~

[jiaweiz716]

Thanks for your attention to our work and sorry for the late response.

It depends on the maximum epochs setting (it is a hyper-parameter) we use the last report as the final results. We fix some bugs these days and soon release the new version with the evaluation script. Hope that could help you~

Thank you for your reply! Looking forward to your new version!