How ravvah/ecoshift works
EcoShift is an experimental predict-then-optimize decision-science system for scheduling flexible industrial loads against electricity-price and carbon forecasts. The forecast service exposes predictions and has a separate model-development path for feature engineering, training, evaluation, and MLflow tracking. The optimizer service defines load constraints and solver implementations for producing schedules. The README describes decision-regret analysis as a goal, but no regret-analysis implementation appears in the supplied tree. The optimizer’s runtime wiring and any connection between the two services are not established by the excerpts.
Main entry points: predict.py Predict endpoint · scipy_lp_solver.py SciPy LP solver · predictor.py Prediction service · forecaster.py Energy forecaster
Components 27
Forecast API
- Forecast app
forecast_service/src/forecast_service/app/main.py - Forecast routes
forecast_service/src/forecast_service/app/api/v1/router.py - Predict endpoint
forecast_service/src/forecast_service/app/api/v1/endpoints/predict.py - Health endpoint
forecast_service/src/forecast_service/app/api/v1/endpoints/health.py - API dependencies
forecast_service/src/forecast_service/app/api/dependencies.py - API security
forecast_service/src/forecast_service/app/core/security.py - Prediction service
forecast_service/src/forecast_service/app/services/predictor.py
Forecast Models
- Data loading
forecast_service/src/forecast_service/data/data_loader.py - Feature engineering
forecast_service/src/forecast_service/features/features.py - Energy forecaster
forecast_service/src/forecast_service/model/forecaster.py - Model training
forecast_service/src/forecast_service/model/trainer.py - Model evaluation
forecast_service/src/forecast_service/model/evaluator.py - Hyperparameter tuning
forecast_service/src/forecast_service/model/tuner.py - MLflow tracking
forecast_service/src/forecast_service/tracking/mlflow_tracker.py
Optimization
- Optimize endpoint
optimizer_service/src/optimizer_service/app/api/v1/endpoints/optimize.py - Optimization service
optimizer_service/src/optimizer_service/app/services/optimizer.py - Load constraints
optimizer_service/src/optimizer_service/domain/constraints.py - Solver factory
optimizer_service/src/optimizer_service/solver/factory.py - Solver interface
optimizer_service/src/optimizer_service/solver/base.py - SciPy LP solver
optimizer_service/src/optimizer_service/solver/scipy_lp_solver.py - Schedule result
optimizer_service/src/optimizer_service/domain/result.py
Contracts
- Forecast request
forecast_service/src/forecast_service/app/schemas/request.py - Forecast response
forecast_service/src/forecast_service/app/schemas/response.py - Optimization contracts
optimizer_service/src/optimizer_service/app/schemas
Other components
- API caller
- MLflow registry
- SciPy optimizer
Connections 28
- API caller → Predict endpoint: requests forecast
- Forecast app → Prediction service: loads and warms
forecast_service/src/forecast_service/app/main.py - Forecast app → Forecast routes: mounts routes
forecast_service/src/forecast_service/app/main.py - Forecast routes → Predict endpoint: registers route
forecast_service/src/forecast_service/app/api/v1/router.py - Forecast routes → Health endpoint: registers route
forecast_service/src/forecast_service/app/api/v1/router.py - Predict endpoint → API dependencies: resolves service
forecast_service/src/forecast_service/app/api/v1/endpoints/predict.py - Predict endpoint → API security: checks API key
forecast_service/src/forecast_service/app/api/v1/endpoints/predict.py - Predict endpoint → Prediction service: calls prediction
forecast_service/src/forecast_service/app/api/v1/endpoints/predict.py - Predict endpoint → Forecast request: accepts request
forecast_service/src/forecast_service/app/api/v1/endpoints/predict.py - Predict endpoint → Forecast response: returns response
forecast_service/src/forecast_service/app/api/v1/endpoints/predict.py - Prediction service → Forecast request: reads history
forecast_service/src/forecast_service/app/services/predictor.py - Prediction service → Forecast response: builds forecast
forecast_service/src/forecast_service/app/services/predictor.py - Prediction service → MLflow tracking: loads models
forecast_service/src/forecast_service/app/services/predictor.py - Energy forecaster → Feature engineering: transforms features
forecast_service/src/forecast_service/model/forecaster.py - Model training → Energy forecaster: fits and predicts
forecast_service/src/forecast_service/model/trainer.py - Model training → Model evaluation: evaluates folds
forecast_service/src/forecast_service/model/trainer.py - Hyperparameter tuning → Energy forecaster: uses models
forecast_service/src/forecast_service/model/tuner.py - Hyperparameter tuning → Model training: runs training
forecast_service/src/forecast_service/model/tuner.py - MLflow tracking → Energy forecaster: logs model
forecast_service/src/forecast_service/tracking/mlflow_tracker.py - MLflow tracking → MLflow registry: loads and registers
forecast_service/src/forecast_service/tracking/mlflow_tracker.py - Optimize endpoint → Optimization service: delegates optimization
- Optimization service → Optimization contracts: uses API schemas
- Optimization service → Solver factory: selects solver
- Solver factory → SciPy LP solver: creates solver
optimizer_service/src/optimizer_service/solver/factory.py - SciPy LP solver → Solver interface: implements interface
optimizer_service/src/optimizer_service/solver/scipy_lp_solver.py - SciPy LP solver → Load constraints: uses constraints
optimizer_service/src/optimizer_service/solver/scipy_lp_solver.py - SciPy LP solver → Schedule result: builds schedule
optimizer_service/src/optimizer_service/solver/scipy_lp_solver.py - SciPy LP solver → SciPy optimizer: solves LP
optimizer_service/src/optimizer_service/solver/scipy_lp_solver.py