How dys2108/sbl works
This repository implements unsupervised spectral-basis learning for graph matrices: a user supplies dataset and result paths, src/main.py loads mesh features and graph/Laplacian matrices, invokes the TensorFlow basis model, and either trains it or produces test embeddings. Training losses and checkpointing are handled in the entry point; testing writes basis outputs. The matlab/ numerical-analysis files are included as a separate capability, but the excerpts do not establish their call relationships. The imported utils module is absent from the tree, so its behavior is omitted.
Main entry points: main.py Training and testing · model.py Basis model
Components 12
Data workflow
- Training and testing
src/main.py - Model checkpoints
- Embedding results
Spectral learning
- Basis model
src/model.py - Spectral objectives
src/main.py
MATLAB analysis
- Base analysis
matlab/base_analysis.m - ACD16 calculation
matlab/cal_ACD16.m - Cosine calculation
matlab/cal_cos.m - Diagonal calculation
matlab/cal_diag.m - Orthogonality calculation
matlab/cal_orth.m
Other components
- Research user
- Mesh datasets
Connections 7
- Research user → Training and testing: configures paths
- Mesh datasets → Training and testing: provides inputs
- Training and testing → Basis model: builds and runs
- Basis model → Training and testing: returns bases
- Training and testing → Spectral objectives: computes objectives
- Training and testing → Model checkpoints: saves checkpoints
- Training and testing → Embedding results: writes embeddings