How pred-javeriana/pred-engine works
pred-engine is a Python forecasting and model-evaluation library with no UI. Its main workflow ingests and classifies demand data, routes each SKU to model families for selection and fitting, evaluates candidates with causal walk-forward evidence, then evaluates forecasts on a reserved period and publishes results through the platform database. The README documents these stages and optional semantic ingestion, synthetic-data augmentation, model families, statistical verdicts, and artifacts. The supplied excerpts show orchestration and several internal boundaries; unsampled subsystem links are included only where the README explicitly describes them.
Main entry points: pred_engine/pipeline.py Pipeline · ingesta/pipeline.py Ingestion pipeline · enrutador.py Selection router · generacion.py Reserved forecasts
Components 25
Ingestion and data
- Ingestion pipeline
src/pred_engine/ingesta/pipeline.py - Semantic header probe
src/pred_engine/ingesta/sonda/sonda.py - Series categorization
src/pred_engine/ingesta/categorizacion/enrutador.py - Raw data root
src/pred_engine/ingesta/data/deposito.py - Synthetic augmentation
src/pred_engine/aumentacion/fase0.py
Model selection
- Selection router
src/pred_engine/optimizacion/router/enrutador.py - Hyperparameter search
src/pred_engine/optimizacion/optimizadores/HPO/estudio.py - Classical models
src/pred_engine/comun/modelos/modelos_clasicos/sarima.py - Machine learning models
src/pred_engine/comun/modelos/modelos_machine_learning/lgbm.py - Deep learning models
src/pred_engine/comun/modelos/modelos_deep_learning/mlp.py - Foundation model
src/pred_engine/comun/modelos/modelos_fundacionales/pipeline.py - Candidate manifest
src/pred_engine/comun/modelos/manifiesto_candidatos.py
Forecast evaluation
- Walk-forward evaluation
src/pred_engine/comun/walkforward/walk_forward.py - Reserved forecasts
src/pred_engine/forecasting/generador_pronosticos/generacion.py - Candidate validation
src/pred_engine/forecasting/adaptador_candidatos/validacion.py - Forecast error metrics
src/pred_engine/forecasting/evaluaciones/calculo_errores/metricas.py - Selection verdicts
src/pred_engine/forecasting/evaluaciones/veredictos/veredictos.py - HLN-DM testing
src/pred_engine/forecasting/evaluaciones/diebold_mariano/prueba.py - Evaluation confirmation
src/pred_engine/forecasting/evaluaciones/confirmacion/confirmacion.py
Evidence and delivery
- Evidence repository
src/pred_engine/forecasting/persistencia_evidencia/repositorio.py - Platform SQLite and Parquet
- Run artifacts
src/pred_engine/run_artifacts.py
Shared runtime
- Pipeline
src/pred_engine/pipeline.py - Parallel units
src/pred_engine/comun/ejecucion_paralela.py
Other components
- Data operator
Connections 28
- Data operator → Ingestion pipeline: provides CSV
src/pred_engine/ingesta/pipeline.py - Data operator → Pipeline: starts run
README.md - Ingestion pipeline → Raw data root: deposits raw
src/pred_engine/ingesta/pipeline.py - Ingestion pipeline → Semantic header probe: probes headers
src/pred_engine/ingesta/pipeline.py - Semantic header probe → Data operator: uses provider
README.md - Ingestion pipeline → Series categorization: classifies panel
src/pred_engine/ingesta/pipeline.py - Synthetic augmentation → Ingestion pipeline: supplies seed CSV
README.md - Pipeline → Ingestion pipeline: runs L1
README.md - Pipeline → Selection router: routes candidates
src/pred_engine/pipeline.py - Pipeline → Parallel units: dispatches units
src/pred_engine/pipeline.py - Hyperparameter search → Walk-forward evaluation: scores trials
src/pred_engine/optimizacion/optimizadores/HPO/estudio.py - Selection router → Classical models: selects family
README.md - Selection router → Machine learning models: selects family
README.md - Selection router → Deep learning models: selects family
README.md - Selection router → Foundation model: selects family
README.md - Pipeline → Walk-forward evaluation: evaluates fits
README.md - Pipeline → Candidate manifest: emits candidates
README.md - Pipeline → Reserved forecasts: runs reserved forecasts
README.md - Reserved forecasts → Candidate validation: validates candidates
src/pred_engine/forecasting/generador_pronosticos/generacion.py - Reserved forecasts → Forecast error metrics: evaluates forecasts
src/pred_engine/forecasting/generador_pronosticos/generacion.py - Pipeline → Selection verdicts: selects verdicts
README.md - Pipeline → HLN-DM testing: runs complementary test
README.md - Pipeline → Evaluation confirmation: confirms evaluation
- Pipeline → Evidence repository: persists evidence
README.md - Evidence repository → Platform SQLite and Parquet: writes platform DB
README.md - Pipeline → Run artifacts: records run
README.md - Walk-forward evaluation → Run artifacts: produces evidence
README.md - Pipeline → Platform SQLite and Parquet: publishes outputs
README.md