[Example] JKNet

[xnuohz]

Description

Pytorch Implementation of JKNet in paper Representation Learning on Graphs with Jumping Knowledge Networks @mufeili

Checklist

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [$CATEGORY] (such as [NN], [Model], [Doc], [Feature]])
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented
• To the my best knowledge, examples are either not affected by this change,
  or have been fixed to be compatible with this change
• Related issue is referred in this PR
• If the PR is for a new model/paper, I've updated the example index here.

Changes

[mufeili]

Have you tried tuning the hyperparameters? Did you use the ones from the paper?

[xnuohz]

Have you tried tuning the hyperparameters? Did you use the ones from the paper?

I just follow the model architectures from the original paper.

[mufeili]

Have you tried tuning the hyperparameters? Did you use the ones from the paper?

I just follow the model architectures from the original paper.

How about the hyperparameters like learning rate?

[xnuohz]

Have you tried tuning the hyperparameters? Did you use the ones from the paper?

I just follow the model architectures from the original paper.

How about the hyperparameters like learning rate?

I tried lr, dropout, lamb and hid-dim, it's useless.

[mufeili]

Have you tried tuning the number of GNN layers? The paper said that they tried 1-6. Also 500 epochs might be too much for these two datasets. Early stopping based on validation accuracy does not work well with these two datasets.

There's also a comment left to address.

[xnuohz]

Have you tried tuning the number of GNN layers? The paper said that they tried 1-6. Also 500 epochs might be too much for these two datasets. Early stopping based on validation accuracy does not work well with these two datasets.

There's also a comment left to address.

Yes, I've already tried. I set epochs smaller with different GNN layer and mode, it still doesn't work well.