def test_reactor_engine_copies(self):
"""Do deepcopy, pickling and repr of ReactorEngine work as expected?"""
orig = engines.ReactorEngine(
nsteps=101,
Tinj=300.0,
rxnmech="dodecane_lu_nox.cti",
mdot=0.1,
max_minj=5e-5,
reward=rw.Reward(negative_reward=-0.05),
)
assert_deepcopy_pickle_repr(orig)
negative_reward=rwd_params["negative_reward"].value,
EOC_reward=rwd_params["EOC_reward"].value,
randomize=rwd_params["randomize"].value,
random_updates=rwd_params["random_updates"].value,
)
# Initialize the engine
eng_params = params.inputs["engine"]
if eng_params["engine"].value == "reactor-engine":
eng = engines.ReactorEngine(
Tinj=eng_params["Tinj"].value,
nsteps=eng_params["nsteps"].value,
mdot=eng_params["mdot"].value,
max_minj=eng_params["max_minj"].value,
injection_delay=eng_params["injection_delay"].value,
max_pressure=eng_params["max_pressure"].value,
ename=eng_params["ename"].value,
reward=reward,
fuel=eng_params["fuel"].value,
rxnmech=eng_params["rxnmech"].value,
observables=eng_params["observables"].value,
use_qdot=eng_params["use_qdot"].value,
)
elif eng_params["engine"].value == "EQ-engine":
eng = engines.EquilibrateEngine(
Tinj=eng_params["Tinj"].value,
nsteps=eng_params["nsteps"].value,
mdot=eng_params["mdot"].value,
max_minj=eng_params["max_minj"].value,
injection_delay=eng_params["injection_delay"].value,
max_pressure=eng_params["max_pressure"].value,
ename=eng_params["ename"].value,
def test_reactor_engine_with_complex_reward(self):
"""Does the ReactorEngine with complex reward work as expected?"""
# Initialize engine
reward = rw.Reward(
names=["work", "nox", "soot"],
norms=[1.0, 5e-8, 1e-9],
weights=[0.34, 0.33, 0.33],
negative_reward=-100.0,
randomize=False,
)
eng = engines.ReactorEngine(
nsteps=101,
Tinj=300.0,
rxnmech="dodecane_lu_nox.cti",
mdot=0.1,
max_minj=5e-5,
ename="Isooctane_MBT_DI_50C_Summ.xlsx",
reward=reward,
)
env = DummyVecEnv([lambda: eng])
variables = eng.observables + eng.internals + eng.histories
df = pd.DataFrame(
columns=list(
dict.fromkeys(
variables
+ eng.action.actions
+ ["rewards"]
+ eng.reward.get_rewards()
)
)
)
# Evaluate a dummy agent that injects at a fixed time
t0 = time.time()
done = False
cnt = 0
obs = env.reset()
df.loc[cnt, variables] = [eng.current_state[k] for k in variables]
df.loc[cnt, eng.action.actions] = 0
rwd = list(
eng.reward.compute(eng.current_state, eng.nsteps, False, False).values()
)
df.loc[cnt, eng.reward.get_rewards()] = rwd
df.loc[cnt, ["rewards"]] = [sum(rwd)]
while not done:
cnt += 1
# Agent tries to inject twice, but is not allowed the second time
action = (
[1]
if (eng.current_state["ca"] == 0) or eng.current_state["ca"] == 2
else [0]
)
obs, reward, done, info = env.step(action)
df.loc[cnt, variables] = [info[0]["current_state"][k] for k in variables]
df.loc[cnt, eng.action.actions] = eng.action.current
df.loc[cnt, ["rewards"]] = reward
df.loc[cnt, eng.reward.get_rewards()] = list(info[0]["rewards"].values())
for rwd in eng.reward.get_rewards() + ["rewards"]:
df[f"cumulative_{rwd}"] = np.cumsum(df[rwd])
elapsed = time.time() - t0
utilities.plot_df(env, df, idx=6, name="reactor")
# Test
npt.assert_allclose(np.linalg.norm(df.V), 0.002205916821815495)
npt.assert_allclose(np.linalg.norm(df.p), 34254670.52877185, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df["T"]), 18668.46491609, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.rewards), 54.47632708, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.r_work), 53.47224436, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.r_nox), 14.10312665, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.w_work), 3.41695771, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.w_nox), 3.31645895, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.w_soot), 3.31645895, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.mdot), 0.14142135623730953)
print(f"Wall time for ReactorEngine (complex reward) = {elapsed} seconds")
def test_reactor_engine(self):
"""Does the ReactorEngine work as expected?"""
# Initialize engine
eng = engines.ReactorEngine(
nsteps=101,
Tinj=300.0,
rxnmech="dodecane_lu_nox.cti",
mdot=0.1,
max_minj=5e-5,
ename="Isooctane_MBT_DI_50C_Summ.xlsx",
reward=rw.Reward(negative_reward=-0.05),
)
env = DummyVecEnv([lambda: eng])
variables = eng.observables + eng.internals + eng.histories
df = pd.DataFrame(
columns=list(
dict.fromkeys(
variables
+ eng.action.actions
+ ["rewards"]
+ eng.reward.get_rewards()
)
)
)
# Evaluate a dummy agent that injects at a fixed time
t0 = time.time()
done = False
cnt = 0
obs = env.reset()
df.loc[cnt, variables] = [eng.current_state[k] for k in variables]
df.loc[cnt, eng.action.actions] = 0
rwd = list(
eng.reward.compute(eng.current_state, eng.nsteps, False, False).values()
)
df.loc[cnt, eng.reward.get_rewards()] = rwd
df.loc[cnt, ["rewards"]] = [sum(rwd)]
while not done:
cnt += 1
# Agent tries to inject twice, but is not allowed the second time
action = (
[1]
if (eng.current_state["ca"] == -10) or eng.current_state["ca"] == 10
else [0]
)
obs, reward, done, info = env.step(action)
df.loc[cnt, variables] = [info[0]["current_state"][k] for k in variables]
df.loc[cnt, eng.action.actions] = eng.action.current
df.loc[cnt, ["rewards"]] = reward
df.loc[cnt, eng.reward.get_rewards()] = list(info[0]["rewards"].values())
for rwd in eng.reward.get_rewards() + ["rewards"]:
df[f"cumulative_{rwd}"] = np.cumsum(df[rwd])
elapsed = time.time() - t0
utilities.plot_df(env, df, idx=3, name="reactor")
# Test
npt.assert_allclose(np.linalg.norm(df.V), 0.002205916821815495)
npt.assert_allclose(np.linalg.norm(df.p), 35131769.27712808, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df["T"]), 18843.86576749, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.rewards), 156.55524985, rtol=1e-5)
npt.assert_allclose(np.linalg.norm(df.mdot), 0.14142135623730953)
print(f"Wall time for ReactorEngine = {elapsed} seconds")