How mohaaa216/gnn-flowchart-comparator works
This is a pedagogical Siamese-GNN flowchart comparator: it generates synthetic flowchart pairs, encodes them, optionally trains and evaluates a similarity model, then aligns node embeddings to visualize differences. The README also documents training-history plots and prediction examples. The demo is the runtime orchestrator; training is opt-in. Relationships below follow the sampled source and README. Other capabilities described as future work are not current subsystems.
Main entry points: demo.py Demo Entry · train.py Trainer · generator.py Flowchart Generator · encoder.py Siamese Comparator
Components 16
Synthetic Data
- Flowchart Generator
data/generator.py - Pair Variants
data/generator.py - Graph Converter
data/dataset.py - Family Split
data/dataset.py - Pair Batcher
data/dataset.py
Similarity Model
- Siamese Comparator
models/encoder.py - GINE Encoder
models/encoder.py - Trainer
models/train.py - Evaluator
models/train.py - Model Checkpoint
Difference Analysis
- Node Aligner
explainability/node_diff.py - Difference Visualizer
explainability/node_diff.py
Demo Orchestration
- Demo Entry
demo.py - Training Curves
visualizations.py - Prediction Results
Other components
- User
Connections 18
- User → Demo Entry: runs
README.md - Demo Entry → Flowchart Generator: generates graphs
demo.py - Flowchart Generator → Pair Variants: creates variants
README.md - Demo Entry → Graph Converter: encodes pairs
demo.py - Graph Converter → Flowchart Generator: validates graphs
data/dataset.py - Demo Entry → Family Split: splits families
demo.py - Demo Entry → Trainer: trains optionally
demo.py - Trainer → Pair Batcher: batches pairs
models/train.py - Trainer → Siamese Comparator: optimizes
models/train.py - Siamese Comparator → GINE Encoder: encodes both graphs
models/encoder.py - Trainer → Model Checkpoint: saves best model
models/train.py - Demo Entry → Model Checkpoint: loads model
demo.py - Demo Entry → Evaluator: evaluates test set
demo.py - Evaluator → Siamese Comparator: scores pairs
models/train.py - Demo Entry → Node Aligner: aligns embeddings
demo.py - Demo Entry → Difference Visualizer: renders differences
demo.py - Demo Entry → Training Curves: plots training history
demo.py - Demo Entry → Prediction Results: writes predictions
demo.py