Policies

smart_control.reinforcement_learning.policies.schedule_policy

Reinforcement learning schedule policies.

ScheduleEvent dataclass

ScheduleEvent(
    start_time: Timedelta,
    device: DeviceType,
    setpoint_name: SetpointName,
    setpoint_value: SetpointValue,
)

An event that sets a specific value at a specific time.

SchedulePolicy

SchedulePolicy(
    time_step_spec,
    action_spec: NestedTensorSpec,
    action_sequence: ActionSequence,
    weekday_schedule: Schedule,
    weekend_schedule: Schedule,
    dow_sin_index: int,
    dow_cos_index: int,
    hod_sin_index: int,
    hod_cos_index: int,
    action_normalizers: dict,
    local_start_time: Timestamp,
    name: Optional[str] = None,
)

Bases: TFPolicy

Policy that selects actions based on time-dependent schedules.

create_baseline_schedule_policy

create_baseline_schedule_policy(tf_env: TFPyEnvironment) -> SchedulePolicy

Create baseline schedule policy.

This is the baseline default policy that we use for benchmarking / initial data collection.

Parameters:

Name	Type	Description	Default
tf_env	TFPyEnvironment	The TFPyEnvironment to interact with.	required

Returns:

Type	Description
SchedulePolicy	The schedule policy.

get_active_setpoint

get_active_setpoint(
    schedule: Schedule,
    device: DeviceType,
    setpoint_name: SetpointName,
    timestamp: Timedelta,
) -> SetpointValue

Find the active setpoint value at a given time.