def simple_regret(bandit, best):
max_expected = max([bandit.rewards[i] * bandit.probabilities[i] for i in range(bandit.get_num_arms())])
regrets = [0 for _ in range(len(best))]
for i in range(len(best)):
idx = best[i]
regrets[i] = max_expected - (bandit.rewards[idx] * bandit.probabilities[idx])
return regrets
def cumulative_regret(bandit, pulls):
max_expected = max([bandit.rewards[i] * bandit.probabilities[i] for i in range(bandit.get_num_arms())])
regrets = [0 for _ in range(len(pulls))]
for i in range(len(pulls)):
if i > 0:
idx = pulls[i]
regrets[i] = regrets[i - 1] + (max_expected - (bandit.rewards[idx] * bandit.probabilities[idx]))
else:
idx = pulls[i]
regrets[i] = max_expected - (bandit.rewards[idx] * bandit.probabilities[idx])
return regrets