SmartBuildSim

Modelling Pipelines

smartbuildsim.models offers four focused modelling workflows built on top of scikit-learn and NumPy utilities:

1. Forecasting — deterministic regression for predicting future sensor values.
2. Anomaly detection — IsolationForest-based labelling of unexpected readings.
3. Clustering — KMeans grouping of zones based on selected sensors.
4. Reinforcement learning — tabular Q-learning for thermostat control.

Each workflow consumes the feature-engineered datasets described in Feature Engineering and many are orchestrated together in examples/scripts/run_example.py.

The sections below summarise the shared workflow stages before diving into each specialised page:

• Load deterministic data via smartbuildsim data generate or the Python helpers in smartbuildsim.data.generator. W razie potrzeby ustaw parametry generatora (np. anomaly_chance), aby wymusić sygnał do benchmarków.
• Prepare features as needed using FeatureConfig from smartbuildsim.features.engineering.
• Train, evaluate, and persist artefacts using the relevant model module.
• Compare models za pomocą smartbuildsim.evaluation.benchmark, który udostępnia wielokrotne losowania, testy istotności oraz analizę wrażliwości na skalowanie jednostek (szczegóły w Benchmarkach).
• Visualise outcomes with smartbuildsim viz plot to confirm the effect of the trained models.

Continue to the dedicated pages for configuration details and runnable examples.

This site is open source. Improve this page.