Add attract-repel link prediction example #10107
Merged
+219
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
for more information, see https://pre-commit.ci
This was referenced Mar 10, 2025
puririshi98
requested changes
Mar 12, 2025
Contributor
puririshi98
left a comment
puririshi98
left a comment
There was a problem hiding this comment.
before i review the code,
@tommyly201 if you see where it says link_pred.py in https://github.com/tommyly201/pytorch_geometric/blob/feature/ar-link-pred-example/examples/README.md
please add something below that explaining how you can get better accuracy (share the range of improvement across your benchmarks as an estimate in parenthesis). then link the example and the corresponding paper.
Update README.md to include ar_link_pred.py example
Update README.md to include ar_link_pred.py example
for more information, see https://pre-commit.ci
For a flexible implementation that can enhance any GNN model with attract-repel capabilities, see [`attract_repel_wrapper.py`](./attract_repel_wrapper.py).
puririshi98
requested changes
Apr 1, 2025
puririshi98
requested changes
Apr 1, 2025
Contributor
puririshi98
left a comment
puririshi98
left a comment
There was a problem hiding this comment.
otherwise LGTM i can merge once we get #10105 (review) merged
for more information, see https://pre-commit.ci
…myly201/pytorch_geometric into feature/ar-link-pred-example
for more information, see https://pre-commit.ci
Contributor
|
LGTM, will approve and merge once we get #10105 in |
puririshi98
added a commit
that referenced
this pull request
Apr 3, 2025
# Add Attract-Repel Embeddings for Link Prediction This PR adds an implementation of Attract-Repel embeddings from the paper "Pseudo-Euclidean Attract-Repel Embeddings for Undirected Graphs" by Peysakhovich et al. ## Overview Attract-Repel embeddings address a fundamental limitation in traditional graph embeddings: their inability to effectively represent non-transitive relationships. The method splits node representations into two components: - Attract component: nodes with similar attract vectors are likely to connect - Repel component: nodes with similar repel vectors are unlikely to connect Link prediction scores are calculated as: attract_dot_product - repel_dot_product ## Implementation - Added `ARLinkPredictor` class that implements the core functionality - Added thorough tests to validate the implementation - The implementation is minimal and focused on the core concept ## Benefits - Improves link prediction performance on graphs with non-transitive relationships - Requires minimal architectural changes to existing models - Provides useful information about graph structure through R-fraction analysis - Example PR [#10107 ](#10107) --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Rishi Puri <[email protected]> Co-authored-by: Rishi Puri <[email protected]>
chrisn-pik
pushed a commit
to chrisn-pik/pytorch_geometric
that referenced
this pull request
Jun 30, 2025
# Add Attract-Repel Embeddings for Link Prediction This PR adds an implementation of Attract-Repel embeddings from the paper "Pseudo-Euclidean Attract-Repel Embeddings for Undirected Graphs" by Peysakhovich et al. ## Overview Attract-Repel embeddings address a fundamental limitation in traditional graph embeddings: their inability to effectively represent non-transitive relationships. The method splits node representations into two components: - Attract component: nodes with similar attract vectors are likely to connect - Repel component: nodes with similar repel vectors are unlikely to connect Link prediction scores are calculated as: attract_dot_product - repel_dot_product ## Implementation - Added `ARLinkPredictor` class that implements the core functionality - Added thorough tests to validate the implementation - The implementation is minimal and focused on the core concept ## Benefits - Improves link prediction performance on graphs with non-transitive relationships - Requires minimal architectural changes to existing models - Provides useful information about graph structure through R-fraction analysis - Example PR [pyg-team#10107 ](pyg-team#10107) --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Rishi Puri <[email protected]> Co-authored-by: Rishi Puri <[email protected]>
chrisn-pik
pushed a commit
to chrisn-pik/pytorch_geometric
that referenced
this pull request
Jun 30, 2025
# Attract-Repel Link Prediction Example This PR adds an example implementation showcasing the attract-repel embeddings (from PR pyg-team#10105) on link prediction tasks. The example demonstrates how attract-repel embeddings can significantly improve performance on citation networks. ## Overview The example: - Compares traditional vs attract-repel embeddings on link prediction - Works with standard Planetoid datasets (Cora, CiteSeer, PubMed) - Shows performance improvements across different networks - Calculates and reports the R-fraction for analysis ## Results | Dataset | Method | Test AUC | |---------|--------|----------| | Cora | Traditional | 0.6624 | | Cora | Attract-Repel | 0.8945 | | PubMed | Traditional | 0.8977 | | PubMed | Attract-Repel | 0.9607 | | CitiSeer | Traditional | 0.8067 | | CitiSeer | Attract-Repel | 0.8206 | The attract-repel approach shows consistent improvements across datasets: - **Cora**: +0.2321 AUC (R-fraction: 0.4075) - **PubMed**: +0.0630 AUC (R-fraction: 0.5062) - **CitiSeer**: +0.0139 AUC (R-fraction: 0.4869) The varying R-fractions indicate that the model adapts to different graph structures, with PubMed using more of the repel component than Cora. ## Usage Traditional approach: python examples/ar_link_pred.py --dataset Cora Attract-Repel approach python examples/ar_link_pred.py --dataset Cora --use_ar This example complements the attract-repel embeddings implementation in PR pyg-team#10105. --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Rishi Puri <[email protected]> Co-authored-by: Rishi Puri <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Attract-Repel Link Prediction Example
This PR adds an example implementation showcasing the attract-repel embeddings (from PR #10105) on link prediction tasks. The example demonstrates how attract-repel embeddings can significantly improve performance on citation networks.
Overview
The example:
Results
The attract-repel approach shows consistent improvements across datasets:
The varying R-fractions indicate that the model adapts to different graph structures, with PubMed using more of the repel component than Cora.
Usage
Traditional approach:
python examples/ar_link_pred.py --dataset Cora
Attract-Repel approach
python examples/ar_link_pred.py --dataset Cora --use_ar
This example complements the attract-repel embeddings implementation in PR #10105.