How astrovall02/tess-vet-cnn works
TESS-Vet is a research pipeline for classifying TESS exoplanet candidates as planets or false positives. Its main workflow prepares candidate labels and engineered catalog features, pairs them with phase-folded light curves, trains and evaluates a dual-branch CNN under a train/validation/test protocol, then summarizes saved predictions. Separate branches cover catalog-only feature studies and centroid vetting. The README documents some data-acquisition and model details not visible in the sampled source, so those links are cited to the README or left unverified.
Main entry points: processed_labels_v2.csv Processed labels · run_protocol.py Training protocol · data_setup.py Dataset and splits · fetch_data.py Catalog ingestion
Components 20
Data preparation
- Catalog ingestion
src/fetch_data.py - Processed labels
data/processed_labels_v2.csv - Light-curve ETL
notebooks/02_NASA_ETL_Pipeline.ipynb - Light-curve arrays
data - Catalog features
src/build_aux_features.py
CNN classification
- Dataset and splits
src/data_setup.py - Dual-branch CNN
src/models.py - Training protocol
src/run_protocol.py
Evaluation studies
- Saved probabilities
results/probs - Metrics and predictions
experiments/make_results.py - Catalog baseline
experiments/catalog_baseline.py - Catalog ablation
experiments/catalog_ablation.py
Centroid vetting
- SPOC coverage
centroid/check_spoc_coverage.py - Centroid features
centroid/extract_centroid_features.py - Difference-image checks
centroid/centroid_checks.py
Other components
- ExoFOP catalog
- MAST / TESS
- SPOC light curves
- TIC catalog
- transit-diffImage
Connections 19
- ExoFOP catalog → Catalog ingestion: provides table
src/fetch_data.py - Catalog ingestion → Processed labels: writes labels
src/fetch_data.py - MAST / TESS → Light-curve ETL: provides curves
README.md - Light-curve ETL → Light-curve arrays: builds arrays
README.md - Processed labels → Catalog features: engineers features
src/build_aux_features.py - Processed labels → Dataset and splits: loads labels
- Light-curve arrays → Dataset and splits: loads arrays
src/data_setup.py - Dataset and splits → Training protocol: feeds batches
src/run_protocol.py - Catalog features → Training protocol: supplies features
- Training protocol → Dual-branch CNN: trains and predicts
- Training protocol → Saved probabilities: saves predictions
src/run_protocol.py - Saved probabilities → Metrics and predictions: summarizes
experiments/make_results.py - Processed labels → Catalog baseline: evaluates features
experiments/catalog_baseline.py - Processed labels → Catalog ablation: tests ablations
experiments/catalog_ablation.py - Processed labels → SPOC coverage: checks targets
centroid/check_spoc_coverage.py - SPOC light curves → SPOC coverage: searches sectors
centroid/check_spoc_coverage.py - SPOC light curves → Centroid features: provides centroids
centroid/extract_centroid_features.py - TIC catalog → Difference-image checks: provides positions
centroid/centroid_checks.py - transit-diffImage → Difference-image checks: supports fits
README.md