citylearn.multi_community module

class citylearn.multi_community.MultiCommunityEnv(*, communities: Sequence[Mapping[str, Any]], render_directory: str | Path = None, render_session_name: str = None)

Bases: Env

Synchronize multiple independent CityLearnEnv communities.

V1 is an orchestration layer only: every child environment keeps its own physics, demand response state, reward and KPIs. The wrapper coordinates reset/step calls and adds portfolio-level KPI rows from district totals.

PORTFOLIO_COMMUNITY_ID = '__portfolio__'

PORTFOLIO_NAME = 'Portfolio'

property action_space: Mapping[str, Any]

Action spaces keyed by community id.

property central_agent: bool

Common child environment central-agent mode.

close()

Close child environments and flush their render buffers.

property community_ids: List[str]

Community identifiers in deterministic execution order.

property entity_specs: Mapping[str, Any]

Entity interface specifications keyed by community id.

property episode_time_steps: int

Effective episode length in synchronized child environments.

evaluate_v2(control_condition: EvaluationCondition = None, baseline_condition: EvaluationCondition = None, comfort_band: float = None, include_business_as_usual: bool = True) → DataFrame

Return local child KPIs plus portfolio rows aggregated from district KPIs.

export_final_kpis(model: Any = None, filepath: str = 'exported_kpis_multi_community.csv', include_business_as_usual: bool = True, export_business_as_usual_timeseries: bool = True, kpi_round_decimals: int = None, export_community_kpis: bool = True, community_filepath: str = 'exported_kpis.csv')

Export child KPIs to child folders and global multi-community KPIs.

get_metadata() → Mapping[str, Any]

Static wrapper metadata and child metadata.

property interface: str

Common child environment interface mode.

property observation_space: Mapping[str, Any]

Observation spaces keyed by community id.

property observations: Mapping[str, Any]

Current child observations keyed by community id.

render() → Mapping[str, Any]

Render every child environment into its own community subfolder.

reset(seed: int = None, options: Mapping[str, Any] = None) → Tuple[Mapping[str, Any], Mapping[str, Any]]

Reset every child environment and return observations by community.

property rewards: Mapping[str, Any]

Child reward histories keyed by community id.

property seconds_per_time_step: float

Common step duration in seconds.

step(actions: Mapping[str, Any]) → Tuple[Mapping[str, Any], Mapping[str, Any], bool, bool, Mapping[str, Any]]

Apply one action payload per community and advance all child envs once.

property terminated: bool

Whether any child environment has terminated.

property time_step: int

Current synchronized time step.

property time_steps: int

Number of time steps in the synchronized episode.

property truncated: bool

Whether any child environment has truncated.

property unwrapped: MultiCommunityEnv

Returns the base non-wrapped environment.

Returns:

The base non-wrapped gymnasium.Env instance

Return type:

Env

property weights: Dict[str, float]

Portfolio aggregation weights by community id.