# Copyright (c) 2019-2020, RTE (https://www.rte-france.com)
# See AUTHORS.txt
# This Source Code Form is subject to the terms of the Mozilla Public License, version 2.0.
# If a copy of the Mozilla Public License, version 2.0 was not distributed with this file,
# you can obtain one at http://mozilla.org/MPL/2.0/.
# SPDX-License-Identifier: MPL-2.0
# This file is part of Grid2Op, Grid2Op a testbed platform to model sequential decision making in power systems.
import copy
import os
from abc import ABC, abstractmethod
from grid2op.Space import RandomObject
from grid2op.Observation import BaseObservation
from grid2op.Action import BaseAction, ActionSpace
[docs]class BaseAgent(RandomObject, ABC):
"""
This class represents the base class of an BaseAgent. All bot / controller / agent used in the Grid2Op simulator
should derived from this class.
To work properly, it is advise to create BaseAgent after the :class:`grid2op.Environment` has been created and reuse
the :attr:`grid2op.Environment.Environment.action_space` to build the BaseAgent.
Attributes
-----------
action_space: :class:`grid2op.Action.ActionSpace`
It represent the action space ie a tool that can serve to create valid action. Note that a valid action can
be illegal or ambiguous, and so lead to a "game over" or to a error. But at least it will have a proper size.
"""
def __init__(self, action_space: ActionSpace):
RandomObject.__init__(self)
self.action_space = copy.deepcopy(action_space)
[docs] def reset(self, obs: BaseObservation):
"""
This method is called at the beginning of a new episode.
It is implemented by agents to reset their internal state if needed.
Attributes
-----------
obs: :class:`grid2op.Observation.BaseObservation`
The first observation corresponding to the initial state of the environment.
"""
pass
[docs] def seed(self, seed: int) -> None:
"""
This function is used to guarantee that the "pseudo random numbers" generated and used by the agent instance
will be deterministic.
This guarantee, if the recommendation in :func:`BaseAgent.act` are followed that the agent will produce the same
set of actions if it faces the same observations in the same order. This is particularly important for
random agent.
You can override this function with the method of your choosing, but if you do so, don't forget to call
`super().seed(seed)`.
Parameters
----------
seed: ``int``
The seed used
Returns
-------
seed: ``tuple``
a tuple of seed used
"""
return super().seed(seed), self.action_space.seed(seed)
[docs] @abstractmethod
def act(self, observation: BaseObservation, reward: float, done : bool=False) -> BaseAction:
"""
This is the main method of an BaseAgent. Given the current observation and the current reward (ie the reward
that the environment send to the agent after the previous action has been implemented).
Notes
-----
In order to be reproducible, and to make proper use of the
:func:`BaseAgent.seed` capabilities, you must absolutely NOT use the `random` python module (which will not
be seeded) nor the `np.random` module and avoid any other "sources" of pseudo random numbers.
You can adapt your code the following way. Instead of using `np.random` use `self.space_prng`.
For example, if you wanted to write
`np.random.randint(1,5)` replace it by `self.space_prng.randint(1,5)`. It is the same for `np.random.normal()`
that is
replaced by `self.space_prng.normal()`.
You have an example of such usage in :func:`RandomAgent.my_act`.
If you really need other sources of randomness (for example if you use tensorflow or torch) we strongly
recommend you to overload the :func:`BaseAgent.seed` accordingly. In that
Parameters
----------
observation: :class:`grid2op.Observation.BaseObservation`
The current observation of the :class:`grid2op.Environment.Environment`
reward: ``float``
The current reward. This is the reward obtained by the previous action
done: ``bool``
Whether the episode has ended or not. Used to maintain gym compatibility
Returns
-------
res: :class:`grid2op.Action.PlaybleAction`
The action chosen by the bot / controler / agent.
"""
return self.action_space()
[docs] def save_state(self, savestate_path :os.PathLike):
"""
An optional method to save the internal state of your agent.
The saved state can later be re-loaded with `self.load_state`, e.g. to repeat
a Grid2Op time step with exactly the same internal parameterization. This
can be useful to repeat Grid2Op experiments and analyze why your agent performed
certain actions in past time steps. Concept developed by Fraunhofer IEE KES.
Notes
-----
First, the internal state your agent consists of attributes that are contained in
the :class:`grid2op.Agent.BaseAgent` and :class:`grid2op.Agent.BaseAgent.action_space`.
Examples are the parameterization and seeds of the random number generator that your
agent uses. Such attributes can easily be obtained with the :func:`getattr` and stored
in a common file format, such as `.npy`.
Second, your agent may contain custom attributes, such as e.g. a vector of line indices
from a Grid2Op observation. You could obtain and save them in the same way as explained
before.
Third, your agent may contain very specific modules such as `Tensorflow` that
do not support the simple :func:`getattr`. However, these modules normally have
their own methods to save an internal state. Examples of such methods are
:func:`save_weights` that you can integrate in your implementation of `self.save_state`.
Parameters
----------
savestate_path: ``string``
The path to which your agent state variables should be saved
"""
pass
[docs] def load_state(self, loadstate_path :os.PathLike):
"""
An optional method to re-load the internal agent state that was saved with `self.save_state`.
This can be useful to re-set your agent to an earlier simulation time step and reproduce
past experiments with Grid2Op. Concept developed by Fraunhofer IEE KES.
Notes
-----
First, the internal state your agent consists of attributes that are contained in
the :class:`grid2op.Agent.BaseAgent` and :class:`grid2op.Agent.BaseAgent.action_space`.
Such attributes can easily be re-set with :func:`setattr`.
Second, your agent may contain custom attributes, such as e.g. a vector of line indices
from a Grid2Op observation. You can re-set them with :func:`setattr` as well.
Third, your agent may contain very specific modules such as `Tensorflow` that
do not support the simple :func:`setattr`. However, these modules normally have
their own methods to re-load an internal state. Examples of such methods are
:func:`load_weights` that you can integrate in your implementation of `self.load_state`.
Parameters
----------
savestate_path: ``string``
The path from which your agent state variables should be loaded
"""
pass