SmartBuildSim

Quickstart

This guide walks through the typical SmartBuildSim workflow using the bundled office-small scenario.

1. Install

pip install -e .[dev]

2. Inspect the Scenario

List available presets and export the selected BIM schema:

smartbuildsim bim init outputs/schema.yaml --scenario office-small

The generated YAML includes building zones, areas, and sensor metadata.

3. Generate Synthetic Data

smartbuildsim data generate examples/configs/default.yaml

The dataset is written to outputs/dataset.csv. All generation steps are seeded to ensure determinism. Add an optional determinism section to your YAML to control the global seed used by the CLI:

determinism:
  seed: 2024

See Deterministic execution for component specific overrides.

4. Train Models

smartbuildsim model forecast examples/configs/default.yaml
smartbuildsim model anomalies examples/configs/default.yaml
smartbuildsim cluster run examples/configs/default.yaml
smartbuildsim rl train examples/configs/default.yaml

Artifacts are produced under outputs/: forecasting model (*.joblib), forecast predictions (forecast_predictions.csv), anomaly annotations (anomalies.csv), cluster assignments (clusters.csv), and RL Q-table (rl_q_table.npy).

5. Plot Results

smartbuildsim viz plot examples/configs/default.yaml \
  --anomalies-path outputs/anomalies.csv --clusters-path outputs/clusters.csv

An annotated plot is saved as outputs/plot_<sensor>.png.

6. Extend via Overrides

Customise runs without editing YAML files:

smartbuildsim data generate examples/configs/default.yaml \
  --override data.days=5 --override data.seed=2024

Overrides apply to nested keys using dotted notation.

For a quick regression test of deterministic behaviour run ./reproduce.sh, which executes the CLI pipeline twice and verifies that the generated datasets match byte-for-byte.

Next steps

Consult the reference documentation for module-level deep dives or open examples/scripts/run_example.py to explore the entire workflow as executable Python code.

This site is open source. Improve this page.