def __init__(self, env_init):
check_gym_version()
self.init_env = env_init.copy()
self.action_space = GymActionSpace(self.init_env)
self.observation_space = GymObservationSpace(self.init_env)
self.reward_range = self.init_env.reward_range
self.metadata = self.init_env.metadata
def __init__(self, dict_gym_space, dict_variables=None):
check_gym_version()
spaces.Dict.__init__(self, dict_gym_space)
self._keys_encoding = {}
if dict_variables is not None:
for k, v in dict_variables.items():
self._keys_encoding[k] = v
self.__func = {}
def __init__(self, space=None, gym_to_g2op=None, g2op_to_gym=None):
check_gym_version()
self.__is_init_super = False # is the "super" class initialized, do not modify in child class
self._is_init_space = False # is the instance initialized
self._my_gym_to_g2op = None
self._my_g2op_to_gym = None
self.my_space = None
if self.my_space is not None:
self.base_initialize(space, gym_to_g2op, g2op_to_gym)
def __init__(self,
grid2op_observation_space,
attr_to_keep=ALL_ATTR_OBS,
subtract=None,
divide=None,
functs=None):
check_gym_version()
if not isinstance(grid2op_observation_space, ObservationSpace):
raise RuntimeError(
f"Impossible to create a BoxGymObsSpace without providing a "
f"grid2op observation. You provided {type(grid2op_observation_space)}"
f"as the \"grid2op_observation_space\" attribute.")
self._attr_to_keep = sorted(attr_to_keep)
ob_sp = grid2op_observation_space
tol_redisp = ob_sp.obs_env._tol_poly # add to gen_p otherwise ... well it can crash
extra_for_losses = _compute_extra_power_for_losses(ob_sp)
self.dict_properties = {
"year": (np.zeros(1, dtype=dt_int),
np.zeros(1, dtype=dt_int) + 2200, (1, ), dt_int),
"month": (np.zeros(1, dtype=dt_int),
np.zeros(1, dtype=dt_int) + 12, (1, ), dt_int),
"day": (np.zeros(1, dtype=dt_int), np.zeros(1, dtype=dt_int) + 31,
(1, ), dt_int),
"hour_of_day": (np.zeros(1, dtype=dt_int),
np.zeros(1, dtype=dt_int) + 24, (1, ), dt_int),
"minute_of_hour": (np.zeros(1, dtype=dt_int),
np.zeros(1, dtype=dt_int) + 60, (1, ), dt_int),
"day_of_week": (np.zeros(1, dtype=dt_int),
np.zeros(1, dtype=dt_int) + 7, (1, ), dt_int),
"gen_p":
(np.full(shape=(ob_sp.n_gen, ), fill_value=0., dtype=dt_float) -
tol_redisp - extra_for_losses,
ob_sp.gen_pmax + tol_redisp + extra_for_losses, (ob_sp.n_gen, ),
dt_float),
"gen_q": (np.full(shape=(ob_sp.n_gen, ),
fill_value=-np.inf,
dtype=dt_float),
np.full(shape=(ob_sp.n_gen, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_gen, ), dt_float),
"gen_v": (np.full(shape=(ob_sp.n_gen, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_gen, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_gen, ), dt_float),
"load_p": (np.full(shape=(ob_sp.n_load, ),
fill_value=-np.inf,
dtype=dt_float),
np.full(shape=(ob_sp.n_load, ),
fill_value=+np.inf,
dtype=dt_float), (ob_sp.n_load, ), dt_float),
"load_q": (np.full(shape=(ob_sp.n_load, ),
fill_value=-np.inf,
dtype=dt_float),
np.full(shape=(ob_sp.n_load, ),
fill_value=+np.inf,
dtype=dt_float), (ob_sp.n_load, ), dt_float),
"load_v": (np.full(shape=(ob_sp.n_load, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_load, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_load, ), dt_float),
"p_or": (np.full(shape=(ob_sp.n_line, ),
fill_value=-np.inf,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"q_or": (np.full(shape=(ob_sp.n_line, ),
fill_value=-np.inf,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"a_or": (np.full(shape=(ob_sp.n_line, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"v_or": (np.full(shape=(ob_sp.n_line, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"p_ex": (np.full(shape=(ob_sp.n_line, ),
fill_value=-np.inf,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"q_ex": (np.full(shape=(ob_sp.n_line, ),
fill_value=-np.inf,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"a_ex": (np.full(shape=(ob_sp.n_line, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"v_ex": (np.full(shape=(ob_sp.n_line, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"rho": (np.full(shape=(ob_sp.n_line, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ), dt_float),
"line_status": (np.full(shape=(ob_sp.n_line, ),
fill_value=0,
dtype=dt_int),
np.full(shape=(ob_sp.n_line, ),
fill_value=1,
dtype=dt_int), (ob_sp.n_line, ), dt_int),
"timestep_overflow":
(np.full(shape=(ob_sp.n_line, ),
fill_value=np.iinfo(dt_int).min,
dtype=dt_int),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.iinfo(dt_int).max,
dtype=dt_int), (ob_sp.n_line, ), dt_int),
"topo_vect": (np.full(shape=(ob_sp.dim_topo, ),
fill_value=-1,
dtype=dt_int),
np.full(shape=(ob_sp.dim_topo, ),
fill_value=2,
dtype=dt_int), (ob_sp.dim_topo, ), dt_int),
"time_before_cooldown_line":
(np.full(shape=(ob_sp.n_line, ), fill_value=0, dtype=dt_int),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.iinfo(dt_int).max,
dtype=dt_int), (ob_sp.n_line, ), dt_int),
"time_before_cooldown_sub":
(np.full(shape=(ob_sp.n_sub, ), fill_value=0, dtype=dt_int),
np.full(shape=(ob_sp.n_sub, ),
fill_value=np.iinfo(dt_int).max,
dtype=dt_int), (ob_sp.n_sub, ), dt_int),
"time_next_maintenance": (np.full(shape=(ob_sp.n_line, ),
fill_value=-1,
dtype=dt_int),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.iinfo(dt_int).max,
dtype=dt_int), (ob_sp.n_line, ),
dt_int),
"duration_next_maintenance":
(np.full(shape=(ob_sp.n_line, ), fill_value=0, dtype=dt_int),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.iinfo(dt_int).max,
dtype=dt_int), (ob_sp.n_line, ), dt_int),
"target_dispatch": (np.minimum(ob_sp.gen_pmin, -ob_sp.gen_pmax),
np.maximum(-ob_sp.gen_pmin, +ob_sp.gen_pmax),
(ob_sp.n_gen, ), dt_float),
"actual_dispatch": (np.minimum(ob_sp.gen_pmin, -ob_sp.gen_pmax),
np.maximum(-ob_sp.gen_pmin, +ob_sp.gen_pmax),
(ob_sp.n_gen, ), dt_float),
"storage_charge": (np.full(shape=(ob_sp.n_storage, ),
fill_value=0,
dtype=dt_float),
1.0 * ob_sp.storage_Emax, (ob_sp.n_storage, ),
dt_float),
"storage_power_target": (-1.0 * ob_sp.storage_max_p_prod,
1.0 * ob_sp.storage_max_p_absorb,
(ob_sp.n_storage, ), dt_float),
"storage_power": (-1.0 * ob_sp.storage_max_p_prod,
1.0 * ob_sp.storage_max_p_absorb,
(ob_sp.n_storage, ), dt_float),
"curtailment": (np.full(shape=(ob_sp.n_gen, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_gen, ),
fill_value=1.0,
dtype=dt_float), (ob_sp.n_gen, ),
dt_float),
"curtailment_limit": (np.full(shape=(ob_sp.n_gen, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_gen, ),
fill_value=1.0,
dtype=dt_float), (ob_sp.n_gen, ),
dt_float),
"curtailment_mw": (np.full(shape=(ob_sp.n_gen, ),
fill_value=0.,
dtype=dt_float), 1.0 * ob_sp.gen_pmax,
(ob_sp.n_gen, ), dt_float),
"curtailment_limit_mw": (np.full(shape=(ob_sp.n_gen, ),
fill_value=0.,
dtype=dt_float),
1.0 * ob_sp.gen_pmax, (ob_sp.n_gen, ),
dt_float),
"thermal_limit": (np.full(shape=(ob_sp.n_line, ),
fill_value=0.,
dtype=dt_float),
np.full(shape=(ob_sp.n_line, ),
fill_value=np.inf,
dtype=dt_float), (ob_sp.n_line, ),
dt_float),
}
self.dict_properties["prod_p"] = self.dict_properties["gen_p"]
self.dict_properties["prod_q"] = self.dict_properties["gen_q"]
self.dict_properties["prod_v"] = self.dict_properties["gen_v"]
self.dict_properties["gen_p_before_curtail"] = self.dict_properties[
"gen_p"]
if functs is None:
functs = {}
for key in functs.keys():
if key not in self._attr_to_keep:
raise RuntimeError(
f"The key {key} is present in the \"functs\" dictionary but not in the "
f"\"attr_to_keep\". This is not consistent: either ignore this function, "
f"in that case remove \"{key}\" from \"functs\" or you want to add "
f"something to your observation, in that case add it to \"attr_to_keep\""
)
if subtract is None:
subtract = {}
self._subtract = subtract
if divide is None:
divide = {}
self._divide = divide
# handle the "functional" part
self._template_obs = ob_sp._template_obj.copy()
self.__func = {}
self._dims = None
low, high, shape, dtype = self._get_info(functs)
# initialize the base container
Box.__init__(self, low=low, high=high, shape=shape, dtype=dtype)
def __init__(self,
grid2op_action_space,
attr_to_keep=ALL_ATTR,
nb_bins=None):
check_gym_version()
if not isinstance(grid2op_action_space, ActionSpace):
raise RuntimeError(
f"Impossible to create a BoxGymActSpace without providing a "
f"grid2op action_space. You provided {type(grid2op_action_space)}"
f"as the \"grid2op_action_space\" attribute.")
if nb_bins is None:
nb_bins = {
"redispatch": 7,
"set_storage": 7,
"curtail": 7,
"curtail_mw": 7
}
if attr_to_keep == ALL_ATTR:
# by default, i remove all the attributes that are not supported by the action type
# i do not do that if the user specified specific attributes to keep. This is his responsibility in
# in this case
attr_to_keep = {
el
for el in attr_to_keep
if grid2op_action_space.supports_type(el)
}
for el in attr_to_keep:
if el not in ATTR_DISCRETE:
warnings.warn(
f"The class \"MultiDiscreteActSpace\" should mainly be used to consider only discrete "
f"actions (eg. set_line_status, set_bus or change_bus). Though it is possible to use "
f"\"{el}\" when building it, be aware that this continuous action will be treated "
f"as discrete by splitting it into bins. "
f"Consider using the \"BoxGymActSpace\" for these attributes."
)
self._attr_to_keep = sorted(attr_to_keep)
act_sp = grid2op_action_space
self._act_space = copy.deepcopy(grid2op_action_space)
low_gen = -1.0 * act_sp.gen_max_ramp_down
high_gen = 1.0 * act_sp.gen_max_ramp_up
low_gen[~act_sp.gen_redispatchable] = 0.
high_gen[~act_sp.gen_redispatchable] = 0.
# nb, dim, []
self.dict_properties = {
"set_line_status":
([3 for _ in range(act_sp.n_line)], act_sp.n_line, self.ATTR_SET),
"change_line_status": ([2 for _ in range(act_sp.n_line)],
act_sp.n_line, self.ATTR_CHANGE),
"set_bus": ([4 for _ in range(act_sp.dim_topo)], act_sp.dim_topo,
self.ATTR_SET),
"change_bus": ([2 for _ in range(act_sp.dim_topo)],
act_sp.dim_topo, self.ATTR_CHANGE),
"sub_set_bus":
(None, act_sp.n_sub, self.ATTR_NEEDBUILD
), # dimension will be computed on the fly, if the stuff is used
"sub_change_bus":
(None, act_sp.n_sub, self.ATTR_NEEDBUILD
), # dimension will be computed on the fly, if the stuff is used
"one_sub_set":
(None, 1, self.ATTR_NEEDBUILD
), # dimension will be computed on the fly, if the stuff is used
"one_sub_change":
(None, 1, self.ATTR_NEEDBUILD
), # dimension will be computed on the fly, if the stuff is used
}
self._nb_bins = nb_bins
for el in ["redispatch", "set_storage", "curtail", "curtail_mw"]:
if el in attr_to_keep:
if el not in nb_bins:
raise RuntimeError(
f"The attribute you want to keep \"{el}\" is not present in the "
f"\"nb_bins\". This attribute is continuous, you have to specify in how "
f"how to convert it to a discrete space. See the documentation "
f"for more information.")
if el == "redispatch":
self.dict_properties[el] = ([
nb_bins[el] if redisp else 1
for redisp in act_sp.gen_redispatchable
], act_sp.n_gen, self.ATTR_NEEDBINARIZED)
elif el == "curtail" or el == "curtail_mw":
self.dict_properties[el] = ([
nb_bins[el] if renew else 1
for renew in act_sp.gen_renewable
], act_sp.n_gen, self.ATTR_NEEDBINARIZED)
elif el == "set_storage":
self.dict_properties[el] = ([
nb_bins[el] for _ in range(act_sp.n_storage)
], act_sp.n_gen, self.ATTR_NEEDBINARIZED)
else:
raise RuntimeError(f"Unknown attribute \"{el}\"")
self._dims = None
self._functs = None # final functions that is applied to the gym action to map it to a grid2Op action
self._binarizers = None # contains all the stuff to binarize the data
self._types = None
nvec = self._get_info()
# initialize the base container
MultiDiscrete.__init__(self, nvec=nvec)
def __init__(self,
grid2op_action_space,
attr_to_keep=ALL_ATTR,
add=None,
multiply=None,
functs=None):
if not isinstance(grid2op_action_space, ActionSpace):
raise RuntimeError(
f"Impossible to create a BoxGymActSpace without providing a "
f"grid2op action_space. You provided {type(grid2op_action_space)}"
f"as the \"grid2op_action_space\" attribute.")
check_gym_version()
if attr_to_keep == ALL_ATTR:
# by default, i remove all the attributes that are not supported by the action type
# i do not do that if the user specified specific attributes to keep. This is his responsibility in
# in this case
attr_to_keep = {
el
for el in attr_to_keep
if grid2op_action_space.supports_type(el)
}
for el in ATTR_DISCRETE:
if el in attr_to_keep:
warnings.warn(
f"The class \"BoxGymActSpace\" should mainly be used to consider only continuous "
f"actions (eg. redispatch, set_storage or curtail). Though it is possible to use "
f"\"{el}\" when building it, be aware that this discrete action will be treated "
f"as continuous. Consider using the \"MultiDiscreteActSpace\" for these attributes."
)
self._attr_to_keep = sorted(attr_to_keep)
act_sp = grid2op_action_space
self._act_space = copy.deepcopy(grid2op_action_space)
low_gen = -1.0 * act_sp.gen_max_ramp_down
high_gen = 1.0 * act_sp.gen_max_ramp_up
low_gen[~act_sp.gen_redispatchable] = 0.
high_gen[~act_sp.gen_redispatchable] = 0.
curtail = np.full(shape=(act_sp.n_gen, ),
fill_value=0.,
dtype=dt_float)
curtail[~act_sp.gen_renewable] = 1.0
curtail_mw = np.full(shape=(act_sp.n_gen, ),
fill_value=0.,
dtype=dt_float)
curtail_mw[~act_sp.gen_renewable] = act_sp.gen_pmax[~act_sp.
gen_renewable]
self.dict_properties = {
"set_line_status":
(np.full(shape=(act_sp.n_line, ), fill_value=-1, dtype=dt_int),
np.full(shape=(act_sp.n_line, ), fill_value=1,
dtype=dt_int), (act_sp.n_line, ), dt_int),
"change_line_status":
(np.full(shape=(act_sp.n_line, ), fill_value=0, dtype=dt_int),
np.full(shape=(act_sp.n_line, ), fill_value=1,
dtype=dt_int), (act_sp.n_line, ), dt_int),
"set_bus": (np.full(shape=(act_sp.dim_topo, ),
fill_value=-1,
dtype=dt_int),
np.full(shape=(act_sp.dim_topo, ),
fill_value=1,
dtype=dt_int), (act_sp.dim_topo, ), dt_int),
"change_bus": (np.full(shape=(act_sp.dim_topo, ),
fill_value=0,
dtype=dt_int),
np.full(shape=(act_sp.dim_topo, ),
fill_value=1,
dtype=dt_int), (act_sp.dim_topo, ), dt_int),
"redispatch": (low_gen, high_gen, (act_sp.n_gen, ), dt_float),
"set_storage": (-1.0 * act_sp.storage_max_p_prod,
1.0 * act_sp.storage_max_p_absorb,
(act_sp.n_storage, ), dt_float),
"curtail": (curtail,
np.full(shape=(act_sp.n_gen, ),
fill_value=1.,
dtype=dt_float), (act_sp.n_gen, ), dt_float),
"curtail_mw":
(curtail_mw, 1.0 * act_sp.gen_pmax, (act_sp.n_gen, ), dt_float),
}
self._key_dict_to_proptype = {
"set_line_status": dt_int,
"change_line_status": dt_bool,
"set_bus": dt_int,
"change_bus": dt_bool,
"redispatch": dt_float,
"set_storage": dt_float,
"curtail": dt_float,
"curtail_mw": dt_float
}
if add is not None:
self._add = add
else:
self._add = {}
if multiply is not None:
self._multiply = multiply
else:
self._multiply = {}
# handle the "functional" part
self.__func = {}
self._dims = None
self._dtypes = None
if functs is None:
functs = {}
low, high, shape, dtype = self._get_info(functs)
# initialize the base container
Box.__init__(self, low=low, high=high, shape=shape, dtype=dtype)