gym_acnportal.gym_acnsim.envs.action_spaces¶
Module containing definition of a gym_acnsim action and factory functions for different builtin actions.
See the SimAction docstring for more information on the SimAction class.
Each factory function takes no arguments and returns an instance of type SimAction. Each factory function defines a space_function and and an to_schedule function with the following signatures:
- space_function:
Callable[[GymInterface], Space]
- to_schedule:
Callable[[GymInterface, np.ndarray], Dict[str, List[float]]]
The space_function gives a gym space for a given action type. to_schedule gives an ACN-Sim schedule for a given action. The to_schedule method does not enforce action space constraints, as some learning algorithms treat action space constraints as loose rather than strict.
Module Contents¶
Functions¶
|
Generates a SimAction instance that wraps functions to handle |
|
Generates a SimAction instance that wraps functions to handle |
-
class
gym_acnportal.gym_acnsim.envs.action_spaces.SimAction(space_function: Callable[[GymTrainedInterface], Space], to_schedule: Callable[[GymTrainedInterface, np.ndarray], Dict[str, List[float]]], name: str)¶ Class representing an OpenAI Gym action for an ACN-Sim Simulation.
An instance of SimAction contains a space_function, which generates a gym space from an input interface using attributes and functions of the input Interface, and a to_schedule function, which generates a ACN-Sim schedule from an input interface and action. Gym observations can have many types, including numpy ndarray, but ACN-Sim schedules must be of type Dict[str, List[float]], namely a list of floats for each station id in the network. Each instance also requires name (given as a string).
This class enables Simulation environments with customizable actions, as a SimAction object with user-defined or built in space and to_schedule functions can be input to a BaseSimEnv-like object to enable a new action type without creating a new environment.
Each type of action is the same type of object, but the details of the space and to_schedule functions are different. This was done because space and to_schedule functions are static, as actions of a specific type do not have any attributes. However, each action type requires both a space and schedule generating function, so a wrapping data structure is required.
- Attributes:
- _space_function (Callable[[GymInterface], Space]):
Function that accepts a GymInterface and generates a gym space in which all actions for this instance exist.
- _to_schedule (Callable[[GymInterface, np.ndarray],
Dict[str, List[float]]]):
Function that accepts an interface to a simulation and an action and generates a schedule that can be submitted to an ACN-Sim Simulation.
- name (str): Name of this action. This attribute allows an
environment to distinguish between different types of actions.
-
_space_function:Callable[[GymTrainedInterface], Space]¶
-
_to_schedule:Callable[[GymTrainedInterface, np.ndarray], Dict[str, List[float]]]¶
-
name:str¶
-
get_space(self, interface: gym_acnportal.gym_acnsim.interfaces.GymTrainedInterface) → gym.Space¶ Returns the gym space in which all actions for this action type exist. The characteristics of the interface (for example, number of EVSEs if charging schedules are given) may change the dimensions of the returned space, so this method requires a GymInterface as input.
- Args:
- interface (GymTrainedInterface): Interface to an ACN-Sim
Simulation that contains details of and functions to generate details about the current Simulation.
- Returns:
- Space: A gym space in which all actions for this
action type exist.
-
get_schedule(self, interface: gym_acnportal.gym_acnsim.interfaces.GymTrainedInterface, action: numpy.ndarray) → Dict[str, List[float]]¶ Returns an ACN-Sim schedule given an input action.
- Args:
interface (GymTrainedInterface): Interface to a simulation. action (np.ndarray): Action to be converted into an ACN-Sim
schedule.
- Returns:
- Dict[str, List[float]]: A charging schedule that can be
submitted to an ACN-Sim Simulator.
-
gym_acnportal.gym_acnsim.envs.action_spaces.single_charging_schedule() → gym_acnportal.gym_acnsim.envs.action_spaces.SimAction¶ Generates a SimAction instance that wraps functions to handle actions taking the form of a vector of pilot signals. For this action type, a single entry represents the pilot signal sent to single EVSE at the current timestep. The space bounds pilot signals above the maximum allowable rate over all EVSEs and below the minimum allowable rate over all EVSEs.
As a 0 min rate is assumed to be allowed, the action space lower bound is set to 0 if the station min rates are all greater than 0.
-
gym_acnportal.gym_acnsim.envs.action_spaces.zero_centered_single_charging_schedule() → gym_acnportal.gym_acnsim.envs.action_spaces.SimAction¶ Generates a SimAction instance that wraps functions to handle actions taking the form of a vector of pilot signals. For this action type, actions are assumed to be centered about 0, in that an action of 0 corresponds to a pilot signal of max_rate/2. So, to convert to a schedule, actions need to be shifted by a certain amount and converted to a dictionary.
As a 0 min rate is assumed to be allowed, the action space lower bound is set to -rate_offset_array if the station min rates are all greater than 0.