def test_issue_187(self):
"""test the range of the reward class"""
for env_name in grid2op.list_available_test_env():
if env_name == "blank":
continue
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
with grid2op.make(env_name,
test=True,
reward_class=RedispReward) as env:
obs = env.reset()
obs, reward, done, info = env.step(env.action_space())
assert reward <= env.reward_range[
1], f"error for reward_max for {env_name}"
assert reward >= env.reward_range[
0], f"error for reward_min for {env_name}"
def test_custom_reward(self):
"""test i can generate the reward and use it in the envs"""
reward_cls = RedispReward.generate_class_custom_params(
alpha_redisph=2,
min_load_ratio=0.15,
worst_losses_ratio=0.05,
min_reward=-10.,
reward_illegal_ambiguous=0.,
least_losses_ratio=0.015)
for env_name in grid2op.list_available_test_env():
if env_name == "blank":
continue
with warnings.catch_warnings():
warnings.filterwarnings("ignore")
with grid2op.make(env_name, test=True,
reward_class=reward_cls) as env:
obs = env.reset()
obs, reward, done, info = env.step(env.action_space())
# test that reward is in the correct range
assert reward <= env.reward_range[
1], f"error reward > reward_max for {env_name}"
assert reward >= env.reward_range[
0], f"error reward < reward_min for {env_name}"
# test the parameters are effectively changed
# what should be computed
_alpha_redisph = dt_float(2)
_min_load_ratio = dt_float(0.15)
_worst_losses_ratio = dt_float(0.05)
_min_reward = dt_float(-10.)
_reward_illegal_ambiguous = dt_float(0.)
_least_losses_ratio = dt_float(0.015)
worst_marginal_cost = np.max(env.gen_cost_per_MW)
worst_load = dt_float(np.sum(env.gen_pmax))
# it's not the worst, but definitely an upper bound
worst_losses = dt_float(_worst_losses_ratio) * worst_load
worst_redisp = _alpha_redisph * np.sum(
env.gen_pmax) # not realistic, but an upper bound
max_regret = (worst_losses +
worst_redisp) * worst_marginal_cost
reward_min = dt_float(_min_reward)
least_loads = dt_float(
worst_load *
_min_load_ratio) # half the capacity of the grid
least_losses = dt_float(_least_losses_ratio *
least_loads) # 1.5% of losses
least_redisp = dt_float(0.0) # lower_bound is 0
base_marginal_cost = np.min(
env.gen_cost_per_MW[env.gen_cost_per_MW > 0.])
min_regret = (least_losses +
least_redisp) * base_marginal_cost
reward_max = dt_float(
(max_regret - min_regret) / least_loads)
assert abs(env.reward_range[1] - reward_max) <= self.tol, \
f"wrong reward max computed for {env_name}"
assert abs(env.reward_range[0] - reward_min) <= self.tol, \
f"wrong reward min computed for {env_name}"
def get_list_env(self):
res = grid2op.list_available_test_env()
res.append(ENV_WITH_ALARM_NAME)
return res
def get_list_env(self):
res = grid2op.list_available_test_env()
# env_with_alarm = os.path.join(PATH_DATA_TEST, "l2rpn_neurips_2020_track1_with_alert")
# res.append(env_with_alarm)
return res