overpull_res = []
### SET K arms
arms = [
[
RestedRottingGaussian, {
'decayingFunction': lambda n: mu if n <= HORIZON / 4 else -mu,
'sigma': sigma,
}
],
[
RestedRottingGaussian, {
'decayingFunction': lambda n: 0,
'sigma': sigma,
}
],
]
rew, noisy_rew, time, pulls, cumul_pulls = repetedRuns(policy,
arms,
rep=REPETITIONS,
T=HORIZON,
parallel=PARALLEL)
oracle_rew, noisy_oracle_rew, oracle_time, oracle_pull, oracle_cumul_pulls = repetedRuns(
[GreedyOracle, {}], arms, rep=1, T=HORIZON, oracle=True)
regret = oracle_rew - rew
regret_res.append(regret)
time_res.append(time)
logging.info("EVENT : SAVING ... ")
np.save(regret_path, np.array(regret_res))
np.save(time_path, np.array(regret_res))
logging.info("EVENT : END ... ")
'per_arm_restart': True,
'sig2': SIGMA**2
}
],
[Exp3S, {
'alpha': 1 / T,
'gamma': (K * V / T)**(1 / 3)
}],
]
policy_ind = 2 if len(sys.argv) == 1 else int(sys.argv[2])
policy = policies[policy_ind]
policy_name = str(policy[0](nbArms=2, **policy[1]))
policy_name_nospace = policy_name.replace(' ', '_')
logging.info("CONSTANT CONFIG : POLICY " + str(i) + " " + policy_name)
rew, noisy_rew, time, pulls, cumul_pulls = repetedRuns(policy,
arms,
rep=REPETITIONS,
T=HORIZON,
parallel=PARALLEL)
oracle_rew = df.max(axis=1).cumsum().to_numpy()
regret = DRAWS * oracle_rew - rew
logging.info("EVENT : SAVING ... ")
path_regret = os.path.join('./data/DAY_%s_REGRET_%s_%s' %
(DAY, policy_name_nospace, date))
path_time = os.path.join('./data/DAY_%s_TIME_%s_%s' %
(DAY, policy_name_nospace, date))
np.save(path_regret, regret)
np.save(path_time, time)
logging.info("EVENT : END ... ")