def validation_experiment(base_folder, env_entry_point):
experiment_name = "setup_validation"
experiment_folder = prepare_experiment(base_folder, experiment_name)
if experiment_folder is None:
return
description_file = "{}/experiment_description.txt".format(
experiment_folder)
with open(description_file, "w") as f:
f.write(gen_exp_descr(experiment_name, None))
run_ps_agent_experiment_with_result_files(
agent_train_test_once=lambda env: ps_train_test_osp_ql(
ps_env=env,
max_number_of_steps=MAX_NUMBER_OF_STEPS,
n_episodes=N_EPISODES,
rand_qtab=RAND_QTAB,
learning_rate=LEARNING_RATE,
discount_factor=DISCOUNT_FACTOR,
exploration_rate=EXPLORATION_RATE,
exploration_decay_rate=EXPLORATION_DECAY_RATE,
k_s=ACTIONS,
visualize=VISUALIZE),
base_folder=experiment_folder,
n_repeat=N_REPEAT,
time_step=TIME_STEP,
p_reff=P_REF,
log_level=LOG_LEVEL,
env_entry_point=env_entry_point,
compute_reward=None,
)
def exploration_experiment(
base_folder,
explor_params,
env_entry_point,
experiment_name="exploration(rate_and_discount)_variation"):
experiment_folder = prepare_experiment(base_folder, experiment_name)
if experiment_folder is None:
return
description_file = "{}/experiment_description.txt".format(
experiment_folder)
with open(description_file, "w") as f:
param_i_correspondence = "\n".join([
"{} - exploration rate = {}, exploration rate decay = {}".format(
i, params[0], params[1])
for i, params in enumerate(explor_params)
])
f.write(gen_exp_descr(experiment_name, param_i_correspondence))
for expl_rate, expl_decay in explor_params:
subfolder = "{}/expl_rate={}_decay={}".format(experiment_folder,
expl_rate, expl_decay)
os.mkdir(subfolder)
run_ps_agent_experiment_with_result_files(
agent_train_test_once=lambda env: ps_train_test_osp_ql(
ps_env=env,
max_number_of_steps=MAX_NUMBER_OF_STEPS,
n_episodes=N_EPISODES,
rand_qtab=RAND_QTAB,
learning_rate=LEARNING_RATE,
discount_factor=DISCOUNT_FACTOR,
exploration_rate=expl_rate,
exploration_decay_rate=expl_decay,
k_s=ACTIONS,
visualize=VISUALIZE,
n_test_episodes=N_TEST_EPISODES),
base_folder=subfolder,
n_repeat=N_REPEAT,
time_step=TIME_STEP,
p_reff=P_REF,
log_level=LOG_LEVEL,
env_entry_point=env_entry_point,
compute_reward=None,
)
def reward_experiment(base_folder,
env_entry_point,
compute_reward_s,
experiment_name="reward_variation"):
experiment_folder = prepare_experiment(base_folder, experiment_name)
if experiment_folder is None:
return
description_file = "{}/experiment_description.txt".format(
experiment_folder)
param_i_correspondence = "\n".join([
"{} - reward = {}".format(i, comp_rew[0])
for i, comp_rew in enumerate(compute_reward_s)
])
with open(description_file, "w") as f:
f.write(gen_exp_descr(experiment_name, param_i_correspondence))
for comp_rew in compute_reward_s:
name, func = comp_rew
subfolder = "{}/reward-{}".format(experiment_folder, name)
os.mkdir(subfolder)
run_ps_agent_experiment_with_result_files(
agent_train_test_once=lambda env: ps_train_test_osp_ql(
ps_env=env,
max_number_of_steps=MAX_NUMBER_OF_STEPS,
n_episodes=N_EPISODES,
rand_qtab=RAND_QTAB,
learning_rate=LEARNING_RATE,
discount_factor=DISCOUNT_FACTOR,
exploration_rate=EXPLORATION_RATE,
exploration_decay_rate=EXPLORATION_DECAY_RATE,
k_s=ACTIONS,
visualize=VISUALIZE,
n_test_episodes=N_TEST_EPISODES),
base_folder=subfolder,
n_repeat=N_REPEAT,
time_step=TIME_STEP,
p_reff=P_REF,
log_level=LOG_LEVEL,
env_entry_point=env_entry_point,
compute_reward=func,
)
def best_combination_experiment(base_folder,
env_entry_point,
t_s,
experiment_name="best_parameters_combination",
**kwargs):
experiment_folder = prepare_experiment(base_folder, experiment_name)
if experiment_folder is None:
return
description_file = "{}/experiment_description.txt".format(
experiment_folder)
with open(description_file, "w") as f:
param_i_correspondence = "\n".join(
["{} - time_step = {}".format(i, t) for i, t in enumerate(t_s)])
f.write(gen_exp_descr(experiment_name, param_i_correspondence))
for t in t_s:
subfolder = "{}/time_step={}".format(experiment_folder, t)
os.mkdir(subfolder)
max_n_steps = MAX_NUMBER_OF_STEPS // t
run_ps_agent_experiment_with_result_files(
agent_train_test_once=lambda env: ps_train_test_osp_ql(
ps_env=env,
max_number_of_steps=max_n_steps,
n_episodes=N_EPISODES,
rand_qtab=RAND_QTAB,
learning_rate=LEARNING_RATE,
discount_factor=DISCOUNT_FACTOR,
exploration_rate=EXPLORATION_RATE,
exploration_decay_rate=EXPLORATION_DECAY_RATE,
k_s=ACTIONS,
visualize=VISUALIZE,
n_test_episodes=N_TEST_EPISODES,
n_test_steps=N_TEST_STEPS),
base_folder=subfolder,
n_repeat=N_REPEAT,
time_step=t,
p_reff=P_REF,
log_level=LOG_LEVEL,
env_entry_point=env_entry_point,
compute_reward=None,
**kwargs)