def test_train_eval(self):
tp = TrainingParam()
tp.buffer_size = 100
tp.minibatch_size = 8
tp.update_freq = 32
tp.min_observation = 32
tmp_dir = tempfile.mkdtemp()
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
env = grid2op.make("rte_case5_example", test=True)
# neural network architecture
li_attr_obs_X = ["prod_p", "load_p", "rho"]
li_attr_obs_Tau = ["line_status"]
sizes = [100, 50, 10]
x_dim = NNParam.get_obs_size(env, li_attr_obs_X)
tau_dims = [
NNParam.get_obs_size(env, [el]) for el in li_attr_obs_Tau
]
kwargs_archi = {
'sizes': sizes,
'activs': ["relu" for _ in sizes],
'x_dim': x_dim,
'tau_dims': tau_dims,
'tau_adds': [0.0 for _ in range(len(tau_dims))],
'tau_mults': [1.0 for _ in range(len(tau_dims))],
"list_attr_obs": li_attr_obs_X,
"list_attr_obs_tau": li_attr_obs_Tau
}
kwargs_converters = {
"all_actions": None,
"set_line_status": False,
"change_bus_vect": True,
"set_topo_vect": False
}
nm_ = "AnneOnymous"
train_leap(env,
name=nm_,
iterations=100,
save_path=tmp_dir,
load_path=None,
logs_dir=tmp_dir,
training_param=tp,
verbose=False,
kwargs_converters=kwargs_converters,
kwargs_archi=kwargs_archi)
baseline_2 = eval_leap(env,
name=nm_,
load_path=tmp_dir,
logs_path=tmp_dir,
nb_episode=1,
nb_process=1,
max_steps=30,
verbose=False,
save_gif=False)
def test_train_eval(self):
if has_SACOld is not None:
raise ImportError(
f"TestSACOld is not available with error:\n{has_SACOld}")
tp = TrainingParam()
tp.buffer_size = 100
tp.minibatch_size = 8
tp.update_freq = 32
tp.min_observation = 32
tmp_dir = tempfile.mkdtemp()
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
env = grid2op.make("rte_case5_example", test=True)
li_attr_obs_X = ["prod_p", "load_p", "rho"]
# neural network architecture
observation_size = NNParam.get_obs_size(env, li_attr_obs_X)
sizes_q = [100, 50, 10] # sizes of each hidden layers
sizes_v = [100, 100] # sizes of each hidden layers
sizes_pol = [100, 10] # sizes of each hidden layers
kwargs_archi = {
'observation_size': observation_size,
'sizes': sizes_q,
'activs': ["relu" for _ in range(len(sizes_q))],
"list_attr_obs": li_attr_obs_X,
"sizes_value": sizes_v,
"activs_value": ["relu" for _ in range(len(sizes_v))],
"sizes_policy": sizes_pol,
"activs_policy": ["relu" for _ in range(len(sizes_pol))]
}
kwargs_converters = {
"all_actions": None,
"set_line_status": False,
"change_bus_vect": True,
"set_topo_vect": False
}
nm_ = "AnneOnymous"
train_sacold(env,
name=nm_,
iterations=100,
save_path=tmp_dir,
load_path=None,
logs_dir=tmp_dir,
training_param=tp,
verbose=False,
kwargs_converters=kwargs_converters,
kwargs_archi=kwargs_archi)
baseline_2 = eval_sacold(env,
name=nm_,
load_path=tmp_dir,
logs_path=tmp_dir,
nb_episode=1,
nb_process=1,
max_steps=30,
verbose=False,
save_gif=False)
def test_train_eval(self):
tp = TrainingParam()
tp.buffer_size = 100
tp.minibatch_size = 8
tp.update_freq = 32
tmp_dir = tempfile.mkdtemp()
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
env = grid2op.make("rte_case5_example", test=True)
li_attr_obs_X = [
"day_of_week", "hour_of_day", "minute_of_hour", "prod_p",
"prod_v", "load_p", "load_q", "actual_dispatch",
"target_dispatch", "topo_vect", "time_before_cooldown_line",
"time_before_cooldown_sub", "rho", "timestep_overflow",
"line_status"
]
# neural network architecture
observation_size = NNParam.get_obs_size(env, li_attr_obs_X)
sizes = [100, 50, 10] # sizes of each hidden layers
kwargs_archi = {
'observation_size': observation_size,
'sizes': sizes,
'activs':
["relu" for _ in sizes], # all relu activation function
"list_attr_obs": li_attr_obs_X
}
kwargs_converters = {
"all_actions": None,
"set_line_status": False,
"change_bus_vect": True,
"set_topo_vect": False
}
nm_ = "AnneOnymous"
train_d3qn(env,
name=nm_,
iterations=100,
save_path=tmp_dir,
load_path=None,
logs_dir=tmp_dir,
nb_env=1,
training_param=tp,
verbose=False,
kwargs_converters=kwargs_converters,
kwargs_archi=kwargs_archi)
baseline_2 = eval_d3qn(env,
name=nm_,
load_path=tmp_dir,
logs_path=tmp_dir,
nb_episode=1,
nb_process=1,
max_steps=30,
verbose=False,
save_gif=False)
def test_train_eval_multiprocess(self):
# test only done for this baselines because the feature is coded in base class in DeepQAgent
tp = TrainingParam()
tp.buffer_size = 100
tp.minibatch_size = 8
tp.update_freq = 32
tp.min_observation = 32
tmp_dir = tempfile.mkdtemp()
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
env_init = grid2op.make("rte_case5_example", test=True)
env = make_multi_env(env_init=env_init, nb_env=2)
li_attr_obs_X = ["prod_p", "load_p", "rho"]
# neural network architecture
observation_size = NNParam.get_obs_size(env, li_attr_obs_X)
sizes = [100, 50, 10] # sizes of each hidden layers
kwargs_archi = {
'observation_size': observation_size,
'sizes': sizes,
'activs':
["relu" for _ in sizes], # all relu activation function
"list_attr_obs": li_attr_obs_X
}
kwargs_converters = {
"all_actions": None,
"set_line_status": False,
"change_bus_vect": True,
"set_topo_vect": False
}
nm_ = "AnneOnymous"
train_dqn(env,
name=nm_,
iterations=100,
save_path=tmp_dir,
load_path=None,
logs_dir=tmp_dir,
training_param=tp,
verbose=False,
kwargs_converters=kwargs_converters,
kwargs_archi=kwargs_archi)
baseline_2 = eval_dqn(env_init,
name=nm_,
load_path=tmp_dir,
logs_path=tmp_dir,
nb_episode=1,
nb_process=1,
max_steps=30,
verbose=False,
save_gif=False)
# NN training
tp.lr = 1e-5
tp.lr_decay_steps = 300000
tp.minibatch_size = 32 * int(args.nb_env)
tp.update_freq = tp.minibatch_size / 2
# limit the number of time steps played per scenarios
tp.step_increase_nb_iter = None # None to deactivate it
tp.min_iter = None
tp.update_nb_iter = None # once 100 scenarios are solved, increase of "step_increase_nb_iter"
# oversampling hard scenarios
tp.oversampling_rate = None # None to deactivate it
# experience replay
tp.buffer_size = 1000000
# just observe the data for a while
tp.min_observe = None # int(10000)
# e greedy
tp.min_observation = 128
tp.initial_epsilon = 0.2
tp.final_epsilon = 1./(288.)
tp.step_for_final_epsilon = int(1e5)
# TODO add the "i dont do anything for a few time steps at the beginning of the training"
# don't start always at the same hour (if not None) otherwise random sampling, see docs
tp.random_sample_datetime_start = None
# saving, logging etc.
def test_train_eval(self):
tp = TrainingParam()
tp.buffer_size = 100
tp.minibatch_size = 8
tp.update_freq = 32
tp.min_observation = 32
tmp_dir = tempfile.mkdtemp()
if has_LeapNetEncoded is not None:
raise ImportError(
f"TestLeapNetEncoded is not available with error:\n{has_LeapNetEncoded}"
)
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
env = grid2op.make("rte_case5_example", test=True)
kwargs_converters = {
"all_actions": None,
"set_line_status": False,
"change_line_status": True,
"change_bus_vect": False,
"set_topo_vect": False,
"redispacth": False
}
# nn architecture
li_attr_obs_X = ["prod_p", "prod_v", "load_p", "load_q"]
li_attr_obs_input_q = [
"time_before_cooldown_line", "time_before_cooldown_sub",
"actual_dispatch", "target_dispatch", "day_of_week",
"hour_of_day", "minute_of_hour", "rho"
]
li_attr_obs_Tau = ["line_status", "timestep_overflow"]
list_attr_gm_out = [
"a_or", "a_ex", "p_or", "p_ex", "q_or", "q_ex", "prod_q",
"load_v"
] + li_attr_obs_X
kwargs_archi = {
'sizes': [],
'activs': [],
'x_dim': -1,
"list_attr_obs": li_attr_obs_X,
"list_attr_obs_tau": li_attr_obs_Tau,
"list_attr_obs_x": li_attr_obs_X,
"list_attr_obs_input_q": li_attr_obs_input_q,
"list_attr_obs_gm_out": list_attr_gm_out,
'dim_topo': env.dim_topo,
"sizes_enc": (10, 10, 10, 10),
"sizes_main": (50, ),
"sizes_out_gm": (50, ),
"sizes_Qnet": (
50,
50,
)
}
nm_ = "AnneOnymous"
train_leapenc(env,
name=nm_,
iterations=100,
save_path=tmp_dir,
load_path=None,
logs_dir=tmp_dir,
training_param=tp,
verbose=False,
kwargs_converters=kwargs_converters,
kwargs_archi=kwargs_archi)
baseline_2 = eval_leapenc(env,
name=nm_,
load_path=tmp_dir,
logs_path=tmp_dir,
nb_episode=1,
nb_process=1,
max_steps=30,
verbose=False,
save_gif=False)