Graph Cross Network #3650
Graph Cross Network #3650
Conversation
Codecov Report
@@ Coverage Diff @@
## master #3650 +/- ##
==========================================
- Coverage 81.58% 80.82% -0.77%
==========================================
Files 295 299 +4
Lines 14861 15279 +418
==========================================
+ Hits 12125 12349 +224
- Misses 2736 2930 +194
Continue to review full report at Codecov.
|
| self.augment_self_loops() | ||
| self.degree_as_feature() | ||
|
|
||
| def degree_as_feature(self): |
There was a problem hiding this comment.
We can replace that with torch_geometric.transforms.OneHotDegree:
dataset = TUDataset(root, name, transform=T.OneHotDegree(...))| else: | ||
| g.x = node_feat | ||
|
|
||
| def augment_self_loops(self): |
There was a problem hiding this comment.
Not sure why we need this. Can you clarify?
| def forward(self, graph): | ||
| embed, logits1, logits2 = self.gxn(graph) | ||
| logits = F.relu(self.lin1(embed)) | ||
| if self.dropout > 0: |
There was a problem hiding this comment.
This can be removed. F.dropout is a no-op for p=0.0.
| # classification loss | ||
| classify_loss = F.nll_loss(cls_logits, labels.to(device)) | ||
|
|
||
| # loss for vertex infomax pooling |
There was a problem hiding this comment.
Can we integrate the computation of the VIX loss into GraphCrossNet model, e.g., GraphCrossNet.pooling_loss(...)?
| from typing import List, Optional, Union | ||
|
|
||
|
|
||
| class GCNConvWithDropout(torch.nn.Module): |
There was a problem hiding this comment.
Let's not create two much modules here :)
For example, GCNConvWithDropout, DenseLayer can be safely dropped, IMO.
|
|
||
|
|
||
| class GraphCrossModule(torch.nn.Module): | ||
| r""" |
There was a problem hiding this comment.
Please follow the docstring conventions of our other modules to ensure that documentation is rendered correctly, see here.
| num_cross_layers: int = 2): | ||
| super(GraphCrossModule, self).__init__() | ||
| if isinstance(pool_ratios, float): | ||
| pool_ratios = (pool_ratios, pool_ratios) |
There was a problem hiding this comment.
| pool_ratios = (pool_ratios, pool_ratios) | |
| pool_ratios = [pool_ratios] * num_cross_layers |
| Default: :obj:`1.0` | ||
| fuse_weight : float, optional | ||
| The weight parameter used at the end of GXN for channel fusion. | ||
| Default: :obj:`1.0` |
Co-authored-by: Matthias Fey <[email protected]>
Co-authored-by: Matthias Fey <[email protected]>
for more information, see https://pre-commit.ci
This PR contains the PyG implementation of the paper Graph Cross Networks with Vertex Infomax Pooling (NeurIPS 2020). It has been adapted from DGL implementation here, and the author's code is here. I have also included an example to show end-to-end training on graph classification task.
Submitted as part of Stanford CS224W: Machine Learning with Graphs, Autumn 2021-2022.