def __init__(self, explore_policy='constant epsilon', eps=0.05, natural=False,
eps_decay=0.9999, show_every=1000000, evaluate_iter=1000, temp_decay=0.9999, init_temp=20):
self.Q = self.initializeQ()
self.winrates = []
self.natural = natural
self.show_every = show_every
self.evaluate_iter = evaluate_iter
self.env = blackjack.BlackjackEnv(natural=self.natural)
self.n_sub_optimals = []
self.min_eps = 0.001
self.min_temp = 0.1
if explore_policy == 'constant_epsilon':
self.explore_policy = self.e_greedy
self.eps = eps
self.eps_decay = 1
self.temp = init_temp
self.temp_decay = 1
elif explore_policy == 'decay_epsilon':
self.explore_policy = self.e_greedy
self.eps = 1
self.eps_decay = eps_decay
self.temp = init_temp
self.temp_decay = 1
elif explore_policy == 'boltzmann_exploration':
self.explore_policy = self.boltzmann_exploration
self.eps = 1
self.eps_decay = 1
self.temp = init_temp
self.temp_decay = temp_decay
def __init__(self):
# main model # gets trained every step
self.model = self.create_model()
# Target model this is what we .predict against every step
self.target_model = self.create_model()
self.target_model.set_weights(self.model.get_weights())
self.replay_memory = deque(maxlen=self.REPLAY_MEMORY_SIZE)
# self.tensorboard = ModifiedTensorBoard(log_dir=f"logs/{self.MODEL_NAME}-{int(time.time())}")
self.target_update_counter = 0
self.env = blackjack.BlackjackEnv(natural=self.NATURAL)
self.ep_rewards = []
self.eps = 1
def evaluate_model(self, n_episodes):
results = {-1: 0, 0: 0, 1: 0, 1.5: 0}
q_table = self.get_q_table()
policy = self.get_best_policy(q_table)
game = blackjack.BlackjackEnv(natural=self.NATURAL)
for i in range(n_episodes):
state = game.reset()
done = False
while not done:
if state[0] < 12:
new_state, reward, done, _ = game.step(1)
else:
action = policy[state]
new_state, reward, done, _ = game.step(action)
state = new_state
results[reward] += 1
winrate = (results[1] + results[1.5]) / n_episodes * 100
print('\nWin Rate: {:.2f} % ({} games)'.format(winrate, n_episodes))
n_sub_optimal = visualization.compare2Optimal(policy)
print('Suboptimal Actions: {}/200\n'.format(n_sub_optimal))
def evaluate_policy(self):
results = {-1: 0, 0: 0, 1: 0, 1.5: 0}
policy = self.get_best_policy()
game = blackjack.BlackjackEnv(natural=self.natural)
for i in range(self.evaluate_iter):
state = game.reset()
done = False
while not done:
if state[0] < 12:
new_state, reward, done, _ = game.step(1)
else:
action = policy[state]
new_state, reward, done, _ = game.step(action)
state = new_state
results[reward] += 1
winrate = (results[1] + results[1.5]) / self.evaluate_iter * 100
print('\nWin Rate: {:.2f} % ({} games)'.format(winrate,
self.evaluate_iter))
n_sub_optimal = visualization.compare2Optimal(policy)
print('Suboptimal Actions: {}/200\n'.format(n_sub_optimal))
self.winrates.append(winrate)
"""
Gagan Heer A00933997
Decision Making: Rule Based
Please look over the README.md file if there is any trouble using this file
"""
import time
import gym
import blackjack as bj
import random
env = bj.BlackjackEnv()
STAND = 0
HIT = 1
numGames = 1001
# 1 = hit, 0 = stand
def rule_based_action(playerTotal, dealerCard):
nextAction = None
if playerTotal <= 11:
nextAction = HIT
elif playerTotal >= 17:
nextAction = STAND
elif dealerCard >= 7 or dealerCard == 1:
nextAction = HIT
elif (dealerCard <= 6 and dealerCard != 1) and playerTotal >= 13:
nextAction = STAND
elif dealerCard == 2 or dealerCard == 3:
nextAction = HIT
else:
nextAction = STAND
if done:
if reward >= 1:
win += 1
elif reward == 0:
tie += 1
elif reward == -1:
loss += 1
break
# percentage of winning games
return 100 * win / n_test, 100 * tie / n_test, 100 * loss/n_test
# TEST
# Results
print("WITH REPLACEMENT")
env = bj.BlackjackEnv(1000000)
results = main(algo='random', nb_games = 50000, txt = 'replacement.csv', txt_to_read = 'replacement.csv', n_test = 10000)
print("Wins:", results[0], "% || Ties:", results[1], "% || Losses:", results[2], "%")
print("Espérance : " + str(results[0] - results[2]))
print("WITHOUT REPLACEMENT")
env = bj.BlackjackEnv(nb_deck)
results = main(algo='random', nb_games = 50000, txt = 'no_replacement.csv', txt_to_read = 'no_replacement.csv', n_test = 10000)
print("Wins:", results[0], "% || Ties:", results[1], "% || Losses:", results[2], "%")
print("Espérance : " + str(results[0] - results[2]))