def test_main(agent, env):
env, _, out_params = env
runner = Runner(agent[1], env)
runner.run(1)
# env.history.df.to_hdf('tests/test_main.hd5', 'hist')
df = env.history.df.head(100)
df = df.reindex(sorted(df.columns), axis=1)
df2 = pd.read_hdf('tests/test_main.hd5', 'hist').head(100) # noqa
df2 = df2.reindex(sorted(df2.columns), axis=1)
assert df[out_params].to_numpy() == approx(df2[out_params].to_numpy(), 5e-2)
def test_main_paramchange(agent, env):
params, agent = agent
env, _, out_params = env
runner = Runner(agent, env)
params['voltP'].val = 4
runner.run(1)
# env.history.df.to_hdf('tests/test_main2.hd5', 'hist')
df = env.history.df.head(50)
df = df.reindex(sorted(df.columns), axis=1)
df2 = pd.read_hdf('tests/test_main.hd5', 'hist').head(50) # noqa
df2 = df2.reindex(sorted(df2.columns), axis=1)
assert df[out_params].to_numpy() != approx(df2[out_params].to_numpy(), 5e-3)
df2 = pd.read_hdf('tests/test_main2.hd5', 'hist').head(50) # noqa
df2 = df2.reindex(sorted(df2.columns), axis=1)
assert df[out_params].to_numpy() == approx(df2[out_params].to_numpy(), 5e-2)
def test_simpleagent(env):
np.random.seed(1)
env, inputs, out_params = env
class RndAgent(Agent):
def act(self, obs: pd.Series) -> np.ndarray:
return np.random.random(len(inputs))
agent = RndAgent()
runner = Runner(agent, env)
runner.run(1)
# env.history.df.to_hdf('tests/test_main3.hd5', 'hist')
df = env.history.df.head(50)
df = df.reindex(sorted(df.columns), axis=1)
df2 = pd.read_hdf('tests/test_main3.hd5', 'hist').head(50) # noqa
df2 = df2.reindex(sorted(df2.columns), axis=1)
assert df[out_params].to_numpy() == approx(df2[out_params].to_numpy(), 5e-3)
for i in '123'], [f'slave.SPI{i}' for i in 'dq0']],
callback=update_legend,
color=[['b', 'r', 'g'], ['b', 'r', 'g']],
style=[[None], ['--']],
title=
'Example of using an timevariant external current reference'),
],
log_level=logging.INFO,
max_episode_steps=max_episode_steps,
model_params={
'rl1.resistor1.R': partial(load_step, gain=20),
'rl1.resistor2.R': partial(load_step, gain=20),
'rl1.resistor3.R': partial(load_step, gain=20),
'rl1.inductor1.L': 0.001,
'rl1.inductor2.L': 0.001,
'rl1.inductor3.L': 0.001
},
model_path='../omg_grid/grid.network.fmu',
net=net)
# User runner to execute num_episodes-times episodes of the env controlled by the agent
runner = Runner(agent, env)
def timeshift(component, t):
if t > .1:
return dict(i_ref=np.array([30, 0, 0]))
return dict(i_ref=np.array([5, 0, 0]))
net['inverter2'].post_calculate_hook = timeshift
runner.run(num_episodes, visualise=True)
'rl1.resistor3.R': callback.load_step_resistance,
'rl1.inductor1.L': callback.load_step_inductance,
'rl1.inductor2.L': callback.load_step_inductance,
'rl1.inductor3.L': callback.load_step_inductance
},
max_episode_steps=max_episode_steps,
net=net,
model_path='../../omg_grid/grid.network_singleInverter.fmu',
history=FullHistory()
)
#####################################
# Execution of the experiment
# Using a runner to execute 'num_episodes' different episodes (i.e. SafeOpt iterations)
runner = Runner(agent, env, callback)
runner.run(num_episodes, visualise=True)
print('\n Experiment finished with best set: \n\n {}'.format(agent.history.df[:]))
print('\n Experiment finished with best set: \n')
print('\n {} = {}'.format(adjust, agent.history.df.at[np.argmax(agent.history.df['J']), 'Params']))
print(' Resulting in a performance of J = {}'.format(np.max(agent.history.df['J'])))
print('\n\nBest experiment results are plotted in the following:')
# Show best episode measurment (current) plot
best_env_plt = runner.run_data['best_env_plt']
if best_env_plt:
ax = best_env_plt[0].axes[0]
ax.set_title('Best Episode')