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 __play_against_random(self, best_pop_player, game_iterations, p_id, win_rate):
players = [LudoPlayerRandom() for _ in range(3)]
players.append(best_pop_player)
for i, player in enumerate(players):
player.id = i # selv tildele atributter uden defineret i klassen
score = [0, 0, 0, 0]
for i in range(game_iterations):
random.shuffle(players)
ludoGame = LudoGame(players)
winner = ludoGame.play_full_game()
score[players[winner].id] += 1
win_rate[p_id] = score[3]
print(f'Chromosome {best_pop_player.chromosome} scored win percentage against random {np.around(score/np.sum(score),decimals=2)*100}')
def play_with_on_QLearning_thread(num_games, epsilon, discount_factor,
learning_rate):
players = [LudoPlayerRandom() for _ in range(3)]
players.append(
LudoPlayerQLearningToken("epsilon greedy",
QtableName='Param_optimization/QTable',
RewardName='Param_optimization/Reward',
epsilon=epsilon,
discount_factor=discount_factor,
learning_rate=learning_rate))
for i, player in enumerate(players):
player.id = i # selv tildele atributter uden defineret i klassen
score = [0, 0, 0, 0]
n = num_games
for i in range(n):
random.shuffle(players)
ludoGame = LudoGame(players)
for player in players: # Saving reward for QLearning player
if type(player) == LudoPlayerQLearningToken:
player.append_reward()
winner = ludoGame.play_full_game()
score[players[winner].id] += 1
if i % 2500 == 0:
print('Game ', i, ' done')
for player in players:
if type(player) == LudoPlayerQLearningToken:
player.saveQTable()
player.saveReward()
print(
f'Player with eps={epsilon}, discountfactor={discount_factor} and learningrate={learning_rate} won {np.around(score/np.sum(score),decimals=2)*100}'
)
def eval(best_player):
# Evaluting the performance of the best player
players = [LudoPlayerRandom() for _ in range(2)]
players.append(best_player)
players.append(best_player)
n_games = 10000
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}"
)
def play_on_thread(num_games, player_type, opponent_type):
# Player initilization
if player_type is LudoPlayerRandom:
players = [player_type() for _ in range(2)]
elif player_type is LudoPlayerQLearningAction:
# Exhuastive search parameters
epsilon = 0.05
discount_factor = 0.5
learning_rate = 0.25
parameters = [epsilon, discount_factor, learning_rate]
players = [
LudoPlayerQLearningAction(
parameters,
chosenPolicy="greedy",
QtableName='Qlearning_action_logs/QTable_action_r_win',
RewardName='Qlearning_action_logs/Reward_action_r_win')
for _ in range(2)
]
elif player_type is LudoPlayerQLearningToken:
# Exhuastive search parameters
epsilon = 0.05
discount_factor = 0.5
learning_rate = 0.25
players = [
LudoPlayerQLearningToken(
"greedy",
QtableName='Qlearning_token_logs/QTable_token_40000',
RewardName='Qlearning_token_logs/Reward_token_40000',
epsilon=epsilon,
discount_factor=discount_factor,
learning_rate=learning_rate) for _ in range(2)
]
elif player_type is simple_GA_player:
chromosome = [0.800085, 2.05562201, 0.55735083, -0.9978861]
players = [simple_GA_player(chromosome) for _ in range(2)]
elif player_type is MathiasPlayer:
players = [MathiasPlayer() for _ in range(2)]
elif player_type is smart_player:
players = [smart_player() for _ in range(2)]
else:
print("player not found please check correctly added")
return
# Opponent initilization
if opponent_type is LudoPlayerRandom:
players += [LudoPlayerRandom() for _ in range(2)]
elif opponent_type is LudoPlayerQLearningAction:
# Exhuastive search parameters
epsilon = 0.05
discount_factor = 0.5
learning_rate = 0.25
parameters = [epsilon, discount_factor, learning_rate]
players += [
LudoPlayerQLearningAction(
parameters,
chosenPolicy="greedy",
QtableName='Qlearning_action_logs/QTable_action_r_win',
RewardName='Qlearning_action_logs/Reward_action_r_win')
for _ in range(2)
]
elif opponent_type is LudoPlayerQLearningToken:
# Exhuastive search parameters
epsilon = 0.05
discount_factor = 0.5
learning_rate = 0.25
players += [
LudoPlayerQLearningToken(
"greedy",
QtableName='Qlearning_token_logs/QTable_token_40000',
RewardName='Qlearning_token_logs/Reward_token_40000',
epsilon=epsilon,
discount_factor=discount_factor,
learning_rate=learning_rate) for _ in range(2)
]
elif opponent_type is simple_GA_player:
chromosome = [0.800085, 2.05562201, 0.55735083, -0.9978861]
players += [simple_GA_player(chromosome) for _ in range(2)]
elif opponent_type is MathiasPlayer:
players += [MathiasPlayer() for _ in range(2)]
elif player_type is smart_player:
players = [smart_player() for _ in range(2)]
else:
print("player not found please check correctly added")
return
for i, player in enumerate(players):
player.id = i # selv tildele atributter uden defineret i klassen
score = [0, 0, 0, 0]
n = num_games
for i in range(n):
random.shuffle(players)
ludoGame = LudoGame(players)
winner = ludoGame.play_full_game()
score[players[winner].id] += 1
# if i%2500 == 0:
# print('Game ', i, ' done')
win_rate_percent_player = np.sum((score / np.sum(score))[0:2]) * 100
win_rate_percent_opponent = np.sum((score / np.sum(score))[2:]) * 100
significant_player = binominal_test(num_games, np.sum(score[0:2]))
significant_opponent = binominal_test(num_games, np.sum(score[2:]))
print(
f'Player: {player_type.name}, Won: {win_rate_percent_player:.2f}, Significant: {significant_player} VERSUS. Opponent: {opponent_type.name}, Won: {win_rate_percent_opponent:.2f}, Significant: {significant_opponent}'
)