def analyse_suboptimal_arm_pulls(self):
# Compute deltas and theoretical upper bound of playing each sub-optimal arm.
self.best_arm = mh.get_maximum_index(self.true_means)
mean_of_best_arm = self.true_means[self.best_arm]
for i in range(self.K):
self.deltas[i] = mean_of_best_arm - self.true_means[i]
del_sq_invs = mh.get_instance_dependent_square_inverses(
self.deltas, self.best_arm)
addi_constant = rvh.func_of_pi(add=1, power=2, mult=1 / 3)
time_series = np.arange(self.T + 1)
logarithmic_time_series = rvh.natural_logarithm(time_series)
a = np.array(del_sq_invs)
del_sq_inv_row_matrix = np.reshape(a, (1, -1))
logarithmic_time_series_column_matrix = np.reshape(
logarithmic_time_series, (-1, 1))
matrix = np.dot(logarithmic_time_series_column_matrix,
del_sq_inv_row_matrix)
self.theoretical_bounds_arm_pulls = matrix + addi_constant
def analyse_common_stats(self):
# Compute deltas and theoretical upper bound of regret of UCB1.
self.best_arm = mh.get_maximum_index(self.true_means)
mean_of_best_arm = self.true_means[self.best_arm]
for i in range(self.K):
self.deltas[i] = mean_of_best_arm - self.true_means[i]
sum_del_inv, sum_del = mh.get_instance_dependent_values(
self.best_arm, self.deltas)
mult_constant, addi_constant = mh.get_theoretical_constants(
sum_del_inv, sum_del)
time_series = np.arange(self.T + 1)
self.cum_regret_theo_bound = mult_constant * rvh.natural_logarithm(
time_series) + addi_constant
self.cum_optimal_reward = time_series * mean_of_best_arm