[Part 1 Multi-Armed Bandit Problem] #강화학습

Part 1 Multi-Armed Bandit Problem.py

import numpy as np

# action 별 reward 반환 함수
def pull_bandit_arm(bandits, bandit_number):
	# Pull arm in position bandit_number and return the obtained reward.
	result = np.random.uniform()
	return int(result <= bandits[bandit_number]) # 0 또는 1을 반환한다.

# action 을 결정하는 함수
# 탐험률 반영
def take_epsilon_greedy_action(epsilon, average_rewards):
	# Take random action with probability epsilon, else take best action.
	result = np.random.uniform()
	if result < epsilon:
		return np.random.randint(0, len(average_rewards)) # Random action
	else:
		return np.argmax(average_rewards) # Greedy action

if __name__ == "__main__":
	
	# Probability of success of each bandit
	bandits = [0.1, 0.3, 0.05, 0.55, 0.4]
	num_iterations = 1000
	epsilon = 0.1

	# Store info to know which one is the best action in each moment
	total_rewards = [0 for _ in range(len(bandits))]
	total_attempts = [0 for _ in range(len(bandits))]
	average_rewards = [0.0 for _ in range(len(bandits))]

	#print(total_rewards, total_attempts, average_rewards)
	for iteration in range(num_iterations + 1):
		action = take_epsilon_greedy_action(epsilon, average_rewards)
		reward = pull_bandit_arm(bandits, action)
		#print("action: {}, reward; {}".format(action, reward))

		# Store result
		total_rewards[action] += reward
		total_attempts[action] += 1
		average_rewards[action] = total_rewards[action] / float(total_attempts[action])

		if iteration % 100 == 0:
			print('Average reward for bandits in iteration {} is {}'.format(iteration,
										['{:.2f}'.format(elem) for elem in average_rewards]))

# Print results
best_bandit = np.argmax(average_rewards)
print('\nBest bandit is {} with an average observed reward of {:.4f}'
						.format(best_bandit, average_rewards[best_bandit]))
print('Total observed reward in the {} episodes has been {}'
								.format(num_iterations, sum(total_rewards)))