Graph Relational Attention #4031
Conversation
There was a problem hiding this comment.
This is great. Left a few comments.
Note that we also need to add RGATConv to nn/conv/__init__.py. Furthermore, it would be great if you can add some tests for RGATConv in /test/nn/conv/test_rgat_conv.py (similar to other tests in that folder). Let me know if you need any help for this!
Changed the model class name from Net to RGAT.
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
|
@rusty1s All the issues raised have been addressed. Two issues above are still open with comments made by me for each of them. Please read them and let me know if any further modification somewhere is required. Furthermore, have also added |
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
There was a problem hiding this comment.
Thank you for the updates. This looks great!
Looking at documentation (https://pytorch-geometric--4031.org.readthedocs.build/en/4031/modules/nn.html#torch_geometric.nn.conv.RGATConv), can we make the text of formulas non-italic? For example via \text{(additive attention logits)}.
Furthermore, once a few tests lands I think this is ready to merge. Let me know if you need any guidance in doing so.
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
for more information, see https://pre-commit.ci
|
Have made the text of formulas non-italic in the documentation (https://pytorch-geometric--4031.org.readthedocs.build/en/4031/modules/nn.html#torch_geometric.nn.conv.RGATConv). Furthermore, have also added few tests, as can be seen in |
|
Thanks you for your updates and being so responsive! I will have some last review today or tomorrow and will try to quickly merge afterwards. |
Added Relational Graph Attention Networks, which computes attention coefficients for relational graphs for each relation type for any number of relations and attention heads. Moreover, also provides functionality for edge attributes to handle.