How nandannmenon/local-optimization-policy-link-prediction works
This repository implements a notebook-based link-prediction experiment: graph data is prepared and split, node features and virtual-node connectivity are constructed, and a DQN chooses the GCN aggregation depth used to train and evaluate a link predictor. The README also documents heuristic edge features, validation, and saved results. All implementation components point to the single notebook because no source excerpts were provided; the architecture and relationships below are based on README statements, not inspected notebook code.
Main entry points: LOP_Link_Prediction_RL.ipynb Graph Preparation · C-elegans-frontal.txt.gz C. elegans Data · cora.zip CORA Data · lop_results.csv Experiment Results
Components 14
Graph Data
- C. elegans Data
C-elegans-frontal.txt.gz - CORA Data
cora.zip - Synthetic Graphs
LOP_Link_Prediction_RL.ipynb - Graph Preparation
LOP_Link_Prediction_RL.ipynb - Node Features
LOP_Link_Prediction_RL.ipynb
LOP Model
- Virtual Node
LOP_Link_Prediction_RL.ipynb - Edge Heuristics
LOP_Link_Prediction_RL.ipynb - DQN Policy
LOP_Link_Prediction_RL.ipynb - GCN Encoders
LOP_Link_Prediction_RL.ipynb - Link Predictor
LOP_Link_Prediction_RL.ipynb
Training and Results
- RL Training
LOP_Link_Prediction_RL.ipynb - Validation and Test
LOP_Link_Prediction_RL.ipynb - Experiment Results
lop_results.csv
Other components
- Researcher
Connections 14
- Researcher → Graph Preparation: runs experiment
README.md - C. elegans Data → Graph Preparation: loads graph
README.md - CORA Data → Graph Preparation: loads graph
README.md - Graph Preparation → Synthetic Graphs: generates fallback
README.md - Graph Preparation → Node Features: splits edges
README.md - Node Features → Virtual Node: augments graph
README.md - RL Training → DQN Policy: selects hop
README.md - DQN Policy → GCN Encoders: chooses depth
README.md - GCN Encoders → Link Predictor: produces embeddings
README.md - Edge Heuristics → Link Predictor: supplies heuristics
README.md - RL Training → GCN Encoders: trains encoder
README.md - RL Training → Link Predictor: trains predictor
README.md - RL Training → Validation and Test: evaluates model
README.md - Validation and Test → Experiment Results: saves metrics
README.md