def main(argv):
roms_path = "roms/"
env = Environment("env1", roms_path)
policy_gradient = PolicyGradientBuilder(100800, 100800, False)
rl = AgentProxy(env, 100800)
dqn = DQNFlyweight(agent=rl)
net = ReinforcementLearning(rl)
env.start()
while True:
move_action = random.randint(0, 8)
p_move_action = rl.action_space_down_sample(move_action)
steps_move_action = net.steps_action(p_move_action)
attack_action = random.randint(0, 9)
p_attack_action = rl.action_space_down_sample(attack_action)
steps_attack_action = net.steps_action(p_attack_action)
#frames, reward, round_done, stage_done, game_done = env.step(move_action, attack_action)
frames, reward, \
round_done, stage_done, \
game_done = policy_gradient.learn(steps_move_action, steps_attack_action)
if game_done:
env.new_game()
elif stage_done:
env.next_stage()
elif round_done:
env.next_round()
def __init__(self, index, monitor=None):
roms_path = "/home/zhangchao/Downloads/"
self.env = Environment("env{}".format(index), roms_path)
if monitor:
self.monitor = monitor
else:
self.monitor = None
self.env.start()
def __init__(self, index, monitor=None):
roms_path = "roms/"
self.env = Environment("env{}".format(index), roms_path)
if monitor:
self.monitor = monitor
else:
self.monitor = None
self.env.start()
def run_env(worker_id, roms_path):
env = Environment(f"env{worker_id}", roms_path)
env.start()
while True:
move_action = random.randint(0, 8)
attack_action = random.randint(0, 9)
frames, reward, round_done, stage_done, game_done = env.step(
move_action, attack_action)
if game_done:
env.new_game()
elif stage_done:
env.next_stage()
elif round_done:
env.next_round()
def __init__(self, index, difficulty, monitor=None):
roms_path = "roms/"
self.env = Environment("env{}".format(index),
roms_path,
difficulty=difficulty)
if monitor:
self.monitor = monitor
else:
self.monitor = None
self.env.start()
self.action_space = gym.spaces.Discrete(90)
self.observation_space = gym.spaces.Box(low=0,
high=1,
shape=(3 + self.action_space.n,
128, 128),
dtype=np.float32)
def run(self):
try:
logger.info("Starting Worker")
self.env = Environment(self.env_id, self.roms_path, difficulty=self.difficulty, frame_ratio=self.frameRatio, frames_per_step=self.framesPerStep, throttle=False)
frames = self.env.start()
while True:
self.model.eval()
observations, histories, frames = self.generate_playthrough(frames)
self.model.train()
dataset = wu.compileHistories(observations, histories)
wu.train(self.model, self.optim, self.criterion, dataset)
except Exception as identifier:
logger.error(identifier)
logger.error(traceback.format_exc())
def run(worker_no, roms_path, learning_rate, cluster, data_bins, stats,
saves_path, save_frequency):
name = "worker%d" % worker_no
with tf.device(
tf.train.replica_device_setter(
worker_device="/job:worker/task:%d" % worker_no,
cluster=cluster)):
Model("global", learning_rate)
saver = tf.train.Saver(var_list=tf.get_collection(
tf.GraphKeys.TRAINABLE_VARIABLES, scope="global"))
local_model = Model(name, learning_rate)
update_local_ops = update_target_graph('global', name)
gpu_options = tf.GPUOptions(allow_growth=True)
config = tf.ConfigProto(gpu_options=gpu_options)
server = tf.train.Server(cluster,
job_name="worker",
task_index=worker_no,
config=config)
with tf.train.MonitoredTrainingSession(master=server.target) as sess:
try:
if stats.get_episode() != 0:
saver.restore(sess._sess._sess._sess._sess,
f'{saves_path}/model.ckpt')
sess.run(update_local_ops)
print("Started Worker Updates")
env = Environment(name, roms_path, difficulty=3)
frames = env.start()
while True:
history = {
"observation": [],
"move_action": [],
"attack_action": [],
"reward": []
}
game_done = False
total_reward = 0
while not game_done:
observation = wu.prepro(frames)
history["observation"].append(observation)
move_out, attack_out = sess.run(
[local_model.move_out_sym, local_model.attack_out_sym],
feed_dict={local_model.observation_sym: observation})
move_action_hot = wu.choose_action(move_out)
attack_action_hot = wu.choose_action(attack_out)
history["move_action"].append(move_action_hot)
history["attack_action"].append(attack_action_hot)
frames, r, round_done, stage_done, game_done = env.step(
np.argmax(move_action_hot),
np.argmax(attack_action_hot))
total_reward += r["P1"]
history["reward"].append(r["P1"])
if round_done:
wu.store_history(data_bins, worker_no, history)
history = {
"observation": [],
"move_action": [],
"attack_action": [],
"reward": []
}
if game_done:
wu.train(sess, local_model,
*data_bins.empty_bin(worker_no))
sess.run(update_local_ops)
stats.update({
"score": total_reward,
"stage": env.stage
})
if stats.get_episode() > 0 and stats.get_episode(
) % save_frequency == 0:
saver.save(sess._sess._sess._sess._sess,
f'{saves_path}/model.ckpt')
frames = env.reset()
except:
error = traceback.format_exc()
print(error)
logging.error(error)
exit(1)
import random
import time
from MAMEToolkit.sf_environment import Environment
roms_path = 'roms/' # Replace this with the path to your ROMs
env = Environment('env1', roms_path)
print('[test] Loaded SF3 from ROM file')
# https://github.com/M-J-Murray/MAMEToolkit/blob/3041734391292376aa909938ea5b51030e3c0240/MAMEToolkit/sf_environment/Environment.py#L88
print('[test] Wait until learnable gameplay starts...')
frames = env.start()
print('[test] Start!')
env.close()
print('[test] Your installation is complete!')
import random
from MAMEToolkit.sf_environment import Environment
roms_path = 'rom/'
env = Environment("sfiii3n", roms_path)
env.start()
while True:
move_action = random.randint(0, 8)
attack_action = random.randint(0, 9)
frames, reward, round_done, stage_done, game_done = env.step(
move_action, attack_action)
if game_done:
env.new_game()
elif stage_done:
env.next_stage()
elif round_done:
env.next_round()
move_action = random.randint(0, 8)
attack_action = random.randint(0, 9)
frames, reward, round_done, stage_done, game_done = env.step(
move_action, attack_action)
if game_done:
env.new_game()
elif stage_done:
env.next_stage()
elif round_done:
env.next_round()
if __name__ == '__main__':
# Environments must be created outside of the threads
#threads = [Process(target=run_env, args=(i, roms_path)) for i in range(workers)]
#[thread.start() for thread in threads]
env = Environment("env1", roms_path)
env.start()
while True:
move_action = random.randint(0, 8)
attack_action = random.randint(0, 9)
frames, reward, round_done, stage_done, game_done = env.step(
move_action, attack_action)
if game_done:
env.new_game()
elif stage_done:
env.next_stage()
elif round_done:
env.next_round()
frame = frame[32:214, 12:372] # crop
frame = frame[::3, ::3]
if isGrey:
frame = 0.2989 * frame[:, :,
0] + 0.5870 * frame[:, :,
1] + 0.1140 * frame[:, :,
2] # greyscale
return frame
isGrey = True
roms_path = "../roms/" # Replace this with the path to your ROMs
env = Environment("env1",
roms_path,
frames_per_step=1,
frame_ratio=3,
throttle=False)
fig = plt.figure()
plt.ion()
im: AxesImage = plt.imshow(prepro(env.start(), isGrey),
cmap="gray" if isGrey else None)
plt.axis("off")
plt.show()
while True:
move_action = random.randint(0, 8)
attack_action = random.randint(0, 9)
frame, reward, round_done, stage_done, game_done = env.step(
move_action, attack_action)
im.set_data(prepro(frame, isGrey))