[Dataset] Chameleon #5477
[Dataset] Chameleon #5477
Conversation
mufeili
requested review from
rudongyu,
frozenbugs,
jermainewang and
Rhett-Ying
|
To trigger regression tests:
|
python/dgl/data/wiki_network.py
Outdated
| ) | ||
| self._g.ndata["test_mask"] = F.astype( | ||
| F.stack(test_masks, dim=1), F.data_type_dict["bool"] | ||
| ) |
There was a problem hiding this comment.
The general question is whether the logic of data processing / graph construction should belong to DGL core codebase or not.
There was a problem hiding this comment.
Upon the offline discussion, the general principle will be including the logic of data processing in the core DGL codebase.
| class ChameleonDataset(WikiNetworkDataset): | ||
| """Wikipedia page-page network on chameleons from `Multi-scale Attributed | ||
| Node Embedding <https://arxiv.org/abs/1909.13021>`__ and later processed by | ||
| `Geom-GCN: Geometric Graph Convolutional Networks |
There was a problem hiding this comment.
Why Geom-GCN is related? Is this dataset designated for Geom-GCN?
There was a problem hiding this comment.
Geom-GCN introduced this variant of the dataset, including turning the task from node regression into node classification, modifying node features, and introducing these dataset splits.
There was a problem hiding this comment.
Maybe reword it. Currently working sounds like this dataset is just used by Geom-GCN.
e.g. later processed by --> introduced by.
Wikipedia page-page network on chameleons from `Multi-scale Attributed
Node Embedding <https://arxiv.org/abs/1909.13021>`__, introduced by
`Geom-GCN: Geometric Graph Convolutional Networks
There was a problem hiding this comment.
did "later processed by" -> "later modified by"
* update * update * update * lint * update * CI * lint * update doc --------- Co-authored-by: Ubuntu <ubuntu@ip-172-31-36-188.ap-northeast-1.compute.internal>
* update * update * update * lint * update * CI * lint * update doc --------- Co-authored-by: Ubuntu <ubuntu@ip-172-31-36-188.ap-northeast-1.compute.internal>
Description
This PR adds the variant of Chameleon as introduced in Geom-GCN: Geometric Graph Convolutional Networks.
A few thoughts and suggested practice:
Checklist
Please feel free to remove inapplicable items for your PR.
Changes