Observers

smart_control.reinforcement_learning.observers.base_observer

Base observer interface for all RL observers.

This module defines the Observer abstract class that all RL observers should implement.

Observer

Bases: ABC

Abstract base class for all observers.

Observers are objects that monitor the training process, collect metrics, and visualize the agent's behavior. They are called with trajectories during data collection.

reset abstractmethod

reset() -> None

Reset the observer to its initial state.

This method is called when a new episode starts.

smart_control.reinforcement_learning.observers.print_status_observer

Reinforcement learning print status observer.

PrintStatusObserver

PrintStatusObserver(
    status_interval_steps: int = 1,
    environment=None,
    replay_buffer=None,
    time_zone=DEFAULT_TIME_ZONE,
)

Bases: Observer

Observer that prints status information.

This observer prints information about the training progress, including rewards, execution time, and replay buffer size.

reset

reset() -> None

Reset the observer to its initial state.

smart_control.reinforcement_learning.observers.rendering_observer

Observer for rendering and visualizing environments.

This module provides an observer for rendering RL environments and visualizing agent behavior through plots.

RenderingObserver

RenderingObserver(
    render_interval_steps: int = 10,
    env=None,
    render_fn: Optional[Callable] = None,
    plot_fn: Optional[Callable] = None,
    clear_output_before_render: bool = True,
    time_zone: str = DEFAULT_TIME_ZONE,
    save_path: str = RENDERS_PATH,
)

Bases: Observer

Observer that renders the environment and plots metrics.

This observer renders the environment at specified intervals and can also show plots of metrics.

Initialize the observer.

Parameters:

Name	Type	Description	Default
render_interval_steps	int	Number of steps between renders.	10
env		The environment to render. This must support the current_simulation_timestamp property if plot_fn is specified.	None
render_fn	Optional[Callable]	Optional function to use for rendering. If not provided, environment.render() will be used.	None
plot_fn	Optional[Callable]	Optional function to use for plotting. If not provided, no plotting will be done.	None
clear_output_before_render	bool	Whether to clear output before rendering.	True
time_zone	str	Time zone for plotting timestamps.	DEFAULT_TIME_ZONE
save_path	str	Directory path to save rendered visualizations.	RENDERS_PATH

reset

reset() -> None

Reset the observer to its initial state.

smart_control.reinforcement_learning.observers.composite_observer

Reinforcement learning composite observer.

CompositeObserver

CompositeObserver(observers: Sequence[Observer])

Bases: Observer

Observer that combines multiple observers.

This observer calls all of its constituent observers whenever it is called. It provides a convenient way to use multiple observers together.

Initialize the observer.

Parameters:

Name	Type	Description	Default
observers	Sequence[Observer]	A sequence of observers to combine.	required

add_observer

add_observer(observer: Observer) -> None

Add an observer to the composite.

Parameters:

Name	Type	Description	Default
observer	Observer	The observer to add.	required

close

close() -> None

Close all observers.

remove_observer

remove_observer(observer: Observer) -> None

Remove an observer from the composite.

Parameters:

Name	Type	Description	Default
observer	Observer	The observer to remove.	required

reset

reset() -> None

Reset all observers.