def __init__(self):
self.env = game_environment.ConnectXEnvironment(7, 6, 4)
self.trainer = 0
self.columns = self.env.num_columns
self.rows = self.env.num_rows
self.actions = gym.spaces.Discrete(self.columns)
self.positions = gym.spaces.Discrete(self.columns * self.rows)
self.list_of_trainers = ["new2", "new", "new1", "new3", "new4"]
self.score_list = {i: 0 for i in self.list_of_trainers}
self.games_list = {i: 0 for i in self.list_of_trainers}
self.change_trainer_at_random()
def __init__(self):
self.env = game_environment.ConnectXEnvironment(7, 6, 4)
self.trainer = 0
self.columns = self.env.num_columns
self.rows = self.env.num_rows
self.actions = gym.spaces.Discrete(self.columns)
self.positions = gym.spaces.Discrete(self.columns * self.rows)
self.list_of_trainers = [
"variety14.0", "variety5.0", "variety3.0", "variety4.0",
"variety8.0", "variety9.0", "variety19.0",
"lookahead_vs_verticalbot1", "variety20.0", "variety12.0",
"variety13.0"
]
self.score_list = {i: 0 for i in self.list_of_trainers}
self.games_list = {i: 0 for i in self.list_of_trainers}
self.change_trainer_at_random()
def __init__(self, num_states, num_actions, gamma, max_exp, min_exp,
batch_size, learning_rate):
self.num_actions = num_actions
self.batch_size = batch_size
self.gamma = gamma
self.model = network.ConnectXNetwork2(num_states, num_actions)
self.optimizer = optim.Adam(self.model.parameters(), lr=learning_rate)
self.criterion = nn.MSELoss()
self.mark = 1 #placeholder for verticalbot funticonality
self.name = 0
self.EVALenv = game_environment.ConnectXEnvironment(7, 6, 4)
self.experience = {
'prev_obs': [],
'a': [],
'r': [],
'obs': [],
'done': []
}
self.max_exp = max_exp
self.min_exp = min_exp
def dojo(games, gym, TrainNet, TargetNet, min_epsilon, epsilon, copy_step):
total_loss = 0
even_match = 0
test_match = game_environment.ConnectXEnvironment(7, 6, 4)
_, test_state = test_match.reset()
print(TrainNet.predict(test_state))
decay = 0.9995
for i in range(games):
rewards, loss = gym.generate_data(TrainNet, TargetNet, epsilon,
copy_step)
if rewards == 0:
even_match += 1
print("motstander", gym.trainer.mark)
print("SCORE:", rewards)
gym.score_list[gym.trainer.name] += rewards
gym.games_list[gym.trainer.name] += 1
total_loss += loss
print(i)
if i % 10 == 0 and i != 0:
gym.change_trainer_at_random()
print(TrainNet.predict(test_state))
if i % 2 == 0 and i != 0:
epsilon = max(min_epsilon, epsilon * decay)
if i % 100 == 0 and i != 0:
print('Total Loss:', total_loss)
print('Even matches:', even_match)
gym.print_scores()
gym.reset_scores()
even_match = 0
total_loss = 0
print("games", i)
print("epsilon", epsilon)
if i % copy_step == 0:
TargetNet.copy_weights(TrainNet)
if i % 50000 == 0 and i != 0:
plt = plot_grad_flow(TrainNet.model.named_parameters())
path = "plot" + str(i) + ".png"
plt.savefig(path)
def __init__(self, num_states, num_actions):
self.num_actions = num_actions
self.model = network.ConnectXNetwork2(num_states, num_actions)
self.mark = 6
self.name = 0
self.EVALenv = game_environment.ConnectXEnvironment(7, 6, 4)
[self.side_length + w * 100 + 50, h * 100 + 100], 45)
if self.game.board[space] == 2:
pygame.draw.circle(
self.screen, self.blue,
[self.side_length + w * 100 + 50, h * 100 + 100], 45)
w += 1
if w == self.width:
w = 0
h += 1
pygame.event.pump()
pygame.display.flip()
gamma = 0.99
copy_step = 25
max_exp = 100000
min_exp = 100
batch_size = 32
learning_rate = 0.00146
epsilon = 0.05
decay = 0.999
min_epsilon = 0.01
episodes = 200000
template_gym = DQN2.ConnectXGym2()
Opponent = DQN2.DQN(template_gym.positions.n, template_gym.actions.n, gamma,
max_exp, min_exp, batch_size, learning_rate)
Opponent.load_weights('lookahead_vs_verticalbot2')
game = game_environment.ConnectXEnvironment(7, 6, 4)
GameRendering(game, Opponent, 0)