def param_optimization():
starttime = time.time()
epsilons = [0.1, 0.05]
discount_factors = [0.9, 0.5]
learning_rate = [0.5, 0.25, 0.1, 0.05]
combination = []
multiprocess = []
for e in epsilons:
for d in discount_factors:
for l in learning_rate:
p = multiprocessing.Process(
target=play_with_on_QLearning_thread,
args=(10000, e, d, l))
multiprocess.append(p)
p.start()
for index, process in enumerate(multiprocess):
process.join()
print('That took {} seconds'.format(time.time() - starttime))
Plot = PlotStatistics()
Plot.plotMultiple(pathToFolder="Param_optimization", numMovAvg=1000)
def plot():
Plot = PlotStatistics()
labels = [
r'$\theta_{1}$', r'$\theta_{2}$', r'$\theta_{3}$', r'$\theta_{4}$'
]
Plot.plot_chromosome_2D(
path_to_csv='GA_simple_logs/simple_ga_chromosomes.csv', labels=labels)
def normal_training():
players = [LudoPlayerRandom() for _ in range(3)]
# GA optimized param
# e = 0.40463712
# d = 0.14343606
# a = 0.10783296
# Exhaustive search param
e = 0.05
d = 0.5
a = 0.25
# players.append(LudoPlayerQLearningToken("epsilon greedy", QtableName='Qlearning_token_logs/QTable_token_40000', RewardName='Qlearning_token_logs/Reward_token_40000', epsilon=e, discount_factor=d, learning_rate=a))
# for i, player in enumerate(players):
# player.id = i # selv tildele atributter uden defineret i klassen
# score = [0, 0, 0, 0]
# n = 40000
# start_time = time.time()
# tqdm_1 = tqdm(range(n), ascii=True)
# for i in tqdm_1:
# tqdm_1.set_description_str(f"win rates {np.around(score/np.sum(score),decimals=2)*100}")
# random.shuffle(players)
# ludoGame = LudoGame(players)
# winner = ludoGame.play_full_game()
# score[players[winner].id] += 1
# for player in players: # Saving reward for QLearning player
# if type(player)==LudoPlayerQLearningToken:
# player.append_reward()
# for player in players:
# if type(player)==LudoPlayerQLearningToken:
# player.saveQTable() # only one player that is Qlearning
# player.saveReward()
# duration = time.time() - start_time
# print('win distribution:', score)
# print('win distribution percentage', (score/np.sum(score))*100)
# print('games per second:', n / duration)
Plot = PlotStatistics()
Plot.plotReward(
pathToCSV=
f'Qlearning_token_logs/Reward_token_40000_e-{e}_d-{d}_a-{a}.csv',
numMovAvg=1000)
def normal_play():
players = []
players = [LudoPlayerRandom() for _ in range(3)]
epsilon = 0.05 #0.40463712 #0.05 #
discount_factor = 0.5 #0.14343606 #0.5 #
learning_rate = 0.25 #0.10783296 #0.25 #
parameters = [epsilon, discount_factor, learning_rate]
t1 = LudoPlayerQLearningAction(
parameters,
chosenPolicy="epsilon greedy",
QtableName='Qlearning_action_logs/1QTable_action_r_win',
RewardName='Qlearning_action_logs/1Reward_action_r_win')
players.append(t1)
for i, player in enumerate(players):
player.id = i # selv tildele atributter uden defineret i klassen
score = [0, 0, 0, 0]
# n = 40000
# start_time = time.time()
# tqdm_1 = tqdm(range(n), ascii=True)
# for i in tqdm_1:
# tqdm_1.set_description_str(f"win rates {np.around(score/np.sum(score),decimals=2)*100}")
# random.shuffle(players)
# ludoGame = LudoGame(players)
# winner = ludoGame.play_full_game()
# score[players[winner].id] += 1
# for player in players: # Saving reward for QLearning player
# if type(player)==LudoPlayerQLearningAction:
# player.append_reward()
# for player in players:
# if type(player)==LudoPlayerQLearningAction:
# player.saveQTable()
# player.saveReward()
# duration = time.time() - start_time
# print('win distribution percentage', (score/np.sum(score))*100)
# print('win distribution:', score)
Plot = PlotStatistics()
Plot.plotReward(
pathToCSV=
f'Qlearning_action_logs/1Reward_action_r_win_e-{epsilon}_d-{discount_factor}_a-{learning_rate}.csv',
numMovAvg=1000)
def main():
players = []
players = [LudoPlayerRandom() for _ in range(3)]
epsilon = 0.1
discount_factor = 0.5
learning_rate = 0.1
parameters = [epsilon, discount_factor, learning_rate]
t1 = LudoPlayerQLearningSimple(parameters,
chosenPolicy="greedy",
QtableName='1_QTable_simple',
RewardName='1_Reward_simple')
players.append(t1)
for i, player in enumerate(players):
player.id = i # selv tildele atributter uden defineret i klassen
score = [0, 0, 0, 0]
n = 5000
start_time = time.time()
tqdm_1 = tqdm(range(n), ascii=True)
for i in tqdm_1:
tqdm_1.set_description_str(
f"win rates {np.around(score/np.sum(score),decimals=2)*100}")
random.shuffle(players)
ludoGame = LudoGame(players)
winner = ludoGame.play_full_game()
score[players[winner].id] += 1
for player in players: # Saving reward for QLearning player
if type(player) == LudoPlayerQLearningSimple:
player.append_reward()
player.reset_upd_val()
# player.rewards.append(player.total_reward)
# player.total_reward = 0
for player in players:
if type(player) == LudoPlayerQLearningSimple:
player.saveQTable()
player.saveReward()
duration = time.time() - start_time
print('win distribution:', score)
Plot = PlotStatistics()
Plot.plotReward(pathToCSV='1_Reward_simple_e-0.1_d-0.5_a-0.1.csv',
numMovAvg=1000)
def normal_training():
players = [LudoPlayerRandom() for _ in range(3)]
players.append(
LudoPlayerQLearningFullAction("epsilon greedy",
QtableName='1_QTable_action_full_r_cum',
RewardName='1_Reward_action_full_r_cum',
epsilon=0.1,
discount_factor=0.5,
learning_rate=0.1))
for i, player in enumerate(players):
player.id = i # selv tildele atributter uden defineret i klassen
score = [0, 0, 0, 0]
n = 15000
start_time = time.time()
tqdm_1 = tqdm(range(n), ascii=True)
for i in tqdm_1:
tqdm_1.set_description_str(
f"win rates {np.around(score/np.sum(score),decimals=2)*100}")
random.shuffle(players)
ludoGame = LudoGame(players)
winner = ludoGame.play_full_game()
score[players[winner].id] += 1
for player in players: # Saving reward for QLearning player
if type(player) == LudoPlayerQLearningFullAction:
player.append_reward()
for player in players:
if type(player) == LudoPlayerQLearningFullAction:
player.saveQTable() # only one player that is Qlearning
player.saveReward()
duration = time.time() - start_time
print('win distribution:', score)
print('win distribution percentage', (score / np.sum(score)) * 100)
print('games per second:', n / duration)
Plot = PlotStatistics()
Plot.plotReward(
pathToCSV='1_Reward_action_full_r_cum_e-0.1_d-0.5_a-0.1.csv',
numMovAvg=1000)
def plot():
Plot = PlotStatistics()
labels = [r'$\epsilon$', r'$\gamma$', r'$\alpha$']
#Plot.plot_chromosome_2D(path_to_csv='GA_qlearning_param_opti_logs/qlean_param_opti_ga_chromosomes.csv', labels=labels)
Plot.plot_chromosome_3D(
path_to_csv=
'GA_qlearning_param_opti_logs/qlean_param_opti_ga_chromosomes.csv')
Plot.plot_chromosome_2D_qlearning(
path_to_csv=
'GA_qlearning_param_opti_logs/qlean_param_opti_ga_chromosomes.csv',
labels=labels)
# for i, player in enumerate(players):
# player.id = i # selv tildele atributter uden defineret i klassen
# score = [0, 0, 0, 0]
# tqdm_1 = tqdm(range(n_games), ascii=True)
# for i in tqdm_1:
# random.shuffle(players)
# ludoGame = LudoGame(players)
# winner = ludoGame.play_full_game()
# score[players[winner].id] += 1
# tqdm_1.set_description_str(f"Validating best player: win rates {np.around(score/np.sum(score),decimals=2)*100}")
Plot = PlotStatistics()
Plot.plot_chromosome_2D(path_to_csv='chromosomes_plot.csv')
##### TESTING FOR QLEARNING PARAMETERS #####
# basetournement_test = basetournement(LudoPlayerQLearningSimple, 20, type_player='qlearning') # 32 chromomes max out the thread count of this computer.
# chromosomes = np.random.uniform(low=0, high=1, size=(20, 3)) # 32 chromomes max out the thread count of this computer.
# print(chromosomes)
# best_player = basetournement_test.play_for_generations(chromosomes, tournament_it=100, generations_it=20, validation_it=1) # problem gets the same 5 chromosome as the best
# print(f'The best player chosen has chromosome {best_player.chromosome}')
# Evaluting the performance of the best player
# players = [LudoPlayerRandom() for _ in range(3)]
# players.append(best_player)