How jeonjas25/talus works
Talus is a Washington State landslide-susceptibility modeling project: geospatial covariates and landslide labels become a cleaned sample matrix, which is trained and evaluated with watershed-based spatial folds and a separate Oso holdout. The final model also predicts statewide raster risk, which feeds SHAP interpretation and map products. The README describes upstream geospatial preparation and sampling, but the tree contains no source for those steps; those inputs are shown without invented internal processing links.
Main entry points: save_model.py Model and metrics · full_map.py Statewide prediction · shap_analysis.py SHAP analysis · final_train.py Oso evaluation
Components 22
Geospatial inputs
- Geospatial sources
- Sample matrix
- Matrix cleaning
src/cleaner.py - Clean matrix
- Spatial folds
- Oso holdout
- Aligned feature rasters
Modeling and evaluation
- Spatial CV
src/train.py - Tune weighting
src/tuning_phase_one.py - Tune depth and rate
src/tuning_phase_two.py - Tune estimators
src/tuning_phase_three.py - Oso evaluation
src/final_train.py - Model and metrics
src/save_model.py - Saved XGBoost model
outputs/models/xgb_final.json - Evaluation metrics
outputs/models
Risk products
- Statewide prediction
src/full_map.py - Susceptibility raster
- Interactive map
src/make_site.py - Susceptibility map
outputs/maps/susceptibility_map.html - Static map figures
src/figure_maker.py
Interpretation
- SHAP analysis
src/shap_analysis.py - Risk and SHAP figures
outputs/figures
Connections 30
- Geospatial sources → Sample matrix: inform inputs
- Sample matrix → Matrix cleaning: loads
- Matrix cleaning → Clean matrix: writes
- Clean matrix → Spatial CV: loads
- Spatial folds → Spatial CV: loads
- Clean matrix → Tune weighting: loads
- Spatial folds → Tune weighting: loads
- Clean matrix → Tune depth and rate: loads
- Spatial folds → Tune depth and rate: loads
- Clean matrix → Tune estimators: loads
- Spatial folds → Tune estimators: loads
- Clean matrix → Oso evaluation: loads
- Spatial folds → Oso evaluation: loads
- Oso holdout → Oso evaluation: evaluates
- Clean matrix → Model and metrics: loads
- Spatial folds → Model and metrics: loads
- Oso holdout → Model and metrics: evaluates
- Model and metrics → Saved XGBoost model: writes
- Model and metrics → Evaluation metrics: writes
- Clean matrix → Statewide prediction: loads
- Spatial folds → Statewide prediction: loads
- Aligned feature rasters → Statewide prediction: reads
- Statewide prediction → Susceptibility raster: writes
- Clean matrix → SHAP analysis: loads
- Spatial folds → SHAP analysis: loads
- SHAP analysis → Risk and SHAP figures: writes
- Susceptibility raster → Interactive map: reads
- Interactive map → Susceptibility map: writes
- Susceptibility raster → Static map figures: reads
- Static map figures → Risk and SHAP figures: writes