How s-sayan/shearnet works
ShearNet is a JAX-based galaxy-shear estimation product: it simulates or loads labeled galaxy images, trains a model, evaluates observations with ShearNet or ngmix/metacalibration, and merges estimator catalogs. The graph also includes documented pipeline dispatch and the run/recovery layer. Some edges rely on README statements or SOURCE INDEX references; unsampled implementation details are not inferred.
Main entry points: evaluation/service.py Evaluation service · training/service.py Training service · pipeline.py Pipeline CLI · train.py Training CLI
Components 23
Configuration
- Config validation
shearnet/config/schema.py - Config loading
shearnet/config/config_handler.py - Training CLI
shearnet/cli/train.py - Evaluation CLI
shearnet/cli/evaluate.py - Pipeline CLI
shearnet/cli/pipeline.py
Simulation and training
- Galaxy simulation
shearnet/core/dataset.py - Differentiable simulation
shearnet/core/dataset_jax.py - In-loop generation
shearnet/core/inloop.py - Neural models
shearnet/core/models.py - Training service
shearnet/training/service.py
Evaluation and catalogs
- Evaluation service
shearnet/evaluation/service.py - Observation rendering
shearnet/evaluation/rendering.py - Model prediction
shearnet/evaluation/predictor.py - Measurement workers
shearnet/evaluation/workers.py - NGmix metacalibration
shearnet/methods/ngmix.py - Catalog schema
shearnet/io/catalog_schema.py - FITS catalog
shearnet/io/fits_catalog.py - Catalog merge
shearnet/io/merge.py
Runs and orchestration
- Run artifacts
shearnet/artifacts/runs.py - Training checkpoints
shearnet/artifacts/checkpoints.py - Recovery controls
shearnet/artifacts/recovery.py
Other components
- Researcher
- Slurm scheduler
Connections 29
- Researcher → Training CLI: starts training
README.md - Researcher → Evaluation CLI: starts evaluation
README.md - Researcher → Pipeline CLI: submits or resumes
README.md - Training CLI → Config loading: loads config
shearnet/cli/train.py - Config loading → Config validation: validates settings
shearnet/config/config_handler.py - Training CLI → Training service: runs training
shearnet/cli/train.py - Training service → Galaxy simulation: renders upfront data
shearnet/training/service.py - Training service → In-loop generation: selects in-loop
shearnet/training/service.py - In-loop generation → Differentiable simulation: generates batches
shearnet/core/inloop.py - Training service → Neural models: trains model
shearnet/training/service.py - Training service → Run artifacts: writes run
shearnet/training/service.py - Training service → Training checkpoints: saves checkpoints
shearnet/training/service.py - Pipeline CLI → Slurm scheduler: submits stages
README.md - Pipeline CLI → Training CLI: dispatches training
shearnet/cli/pipeline.py - Pipeline CLI → Evaluation service: dispatches evaluation
shearnet/cli/pipeline.py - Pipeline CLI → Catalog merge: dispatches merge
shearnet/cli/pipeline.py - Evaluation CLI → Config loading: loads config
- Evaluation CLI → Evaluation service: runs estimator
README.md - Evaluation service → Observation rendering: renders observations
shearnet/evaluation/service.py - Evaluation service → Model prediction: uses prediction
shearnet/evaluation/service.py - Evaluation service → Measurement workers: dispatches work
shearnet/evaluation/service.py - Measurement workers → NGmix metacalibration: measures metacal
shearnet/evaluation/workers.py - Evaluation service → Catalog schema: builds tables
shearnet/evaluation/service.py - Evaluation service → FITS catalog: writes catalog
shearnet/evaluation/service.py - Evaluation service → Run artifacts: reads run artifacts
shearnet/evaluation/service.py - Evaluation service → Recovery controls: guards and resumes
shearnet/evaluation/service.py - Researcher → Catalog merge: requests catalog merge
README.md - Catalog merge → FITS catalog: reads catalogs
shearnet/io/merge.py - Catalog merge → Catalog schema: checks schema
shearnet/io/merge.py