def main():
FLAGS(sys.argv)
with sc2_env.SC2Env(
map_name="DefeatZerglingsAndBanelings",
step_mul=step_mul,
visualize=True,
players=[sc2_env.Agent(sc2_env.Race.terran)],
agent_interface_format=sc2_env.AgentInterfaceFormat(
feature_dimensions=sc2_env.Dimensions(screen=64, minimap=64)),
game_steps_per_episode=steps * step_mul) as env:
model = deepq.models.cnn_to_mlp(convs=[(32, 8, 4), (64, 4, 2),
(64, 3, 1)],
hiddens=[256],
dueling=True)
demo_replay = []
act = dqfd.learn(env,
q_func=model,
num_actions=3,
lr=1e-4,
max_timesteps=10000000,
buffer_size=100000,
exploration_fraction=0.5,
exploration_final_eps=0.01,
train_freq=2,
learning_starts=100000,
target_network_update_freq=1000,
gamma=0.99,
prioritized_replay=True,
demo_replay=demo_replay)
act.save("defeat_zerglings.pkl")
def main():
FLAGS(sys.argv)
with sc2_env.SC2Env(
"DefeatZerglingsAndBanelings",
step_mul=step_mul,
visualize=True,
game_steps_per_episode=steps * step_mul) as env:
model = deepq.models.cnn_to_mlp(
convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
hiddens=[256],
dueling=True
)
demo_replay = []
act = dqfd.learn(
env,
q_func=model,
num_actions=3,
lr=1e-4,
max_timesteps=10000000,
buffer_size=100000,
exploration_fraction=0.5,
exploration_final_eps=0.01,
train_freq=2,
learning_starts=100000,
target_network_update_freq=1000,
gamma=0.99,
prioritized_replay=True,
demo_replay=demo_replay
)
act.save("defeat_zerglings.pkl")
def main():
FLAGS(sys.argv)
logdir = "tensorboard"
if(FLAGS.algorithm == "deepq"):
logdir = "./tensorboard/zergling/%s/%s_%s_prio%s_duel%s_lr%s/%s" % (
FLAGS.algorithm,
FLAGS.timesteps,
FLAGS.exploration_fraction,
FLAGS.prioritized,
FLAGS.dueling,
FLAGS.lr,
start_time
)
if(FLAGS.log == "tensorboard"):
Logger.DEFAULT \
= Logger.CURRENT \
= Logger(dir='log.txt',
output_formats=[TensorBoardOutputFormat(logdir)])
elif(FLAGS.log == "stdout"):
os.mkdir(logdir)
Logger.DEFAULT \
= Logger.CURRENT \
= Logger(dir=None,
output_formats=[HumanOutputFormat(logdir+"/log.txt")])
with sc2_env.SC2Env(
map_name="DefeatZerglingsAndBanelings",
minimap_size_px = (FLAGS.minimap_size_px, FLAGS.minimap_size_px),
step_mul=FLAGS.step_mul,
visualize=FLAGS.visualize,
game_steps_per_episode= FLAGS.episode_steps) as env:
model = deepq.models.cnn_to_mlp(
convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1), (64, 3, 1), (64, 3, 1), (32, 3, 1)],
hiddens=[256],
dueling=True
)
act = dqfd.learn(
env,
q_func=model,
num_actions=FLAGS.num_actions,
lr=FLAGS.lr,
print_freq= FLAGS.print_freq,
max_timesteps=FLAGS.timesteps,
buffer_size=FLAGS.buffer_size,
exploration_fraction=FLAGS.exploration_fraction,
exploration_final_eps=FLAGS.exploration_final_eps,
train_freq=FLAGS.train_freq,
learning_starts=FLAGS.learning_starts,
target_network_update_freq=FLAGS.target_network_update_freq,
gamma=FLAGS.gamma,
prioritized_replay=FLAGS.prioritized,
callback=deepq_callback
)
act.save("defeat_zerglings.pkl")
def main():
FLAGS(sys.argv)
logdir = "tensorboard"
if (FLAGS.algorithm == "deepq"):
logdir = "tensorboard/zergling/%s/%s_%s_prio%s_duel%s_lr%s/%s" % (
FLAGS.algorithm, FLAGS.timesteps, FLAGS.exploration_fraction,
FLAGS.prioritized, FLAGS.dueling, FLAGS.lr, start_time)
elif (FLAGS.algorithm == "acktr"):
logdir = "tensorboard/zergling/%s/%s_num%s_lr%s/%s" % (
FLAGS.algorithm, FLAGS.timesteps, FLAGS.num_cpu, FLAGS.lr,
start_time)
if (FLAGS.log == "tensorboard"):
Logger.DEFAULT \
= Logger.CURRENT \
= Logger(dir=None,
output_formats=[TensorBoardOutputFormat(logdir)])
elif (FLAGS.log == "stdout"):
Logger.DEFAULT \
= Logger.CURRENT \
= Logger(dir=None,
output_formats=[HumanOutputFormat(sys.stdout)])
with sc2_env.SC2Env(
map_name="DefeatZerglingsAndBanelings",
step_mul=step_mul,
visualize=True,
agent_interface_format=sc2_env.AgentInterfaceFormat(
feature_dimensions=sc2_env.Dimensions(screen=32, minimap=32)),
game_steps_per_episode=steps * step_mul) as env:
obs = env.reset()
#print(obs[0].observation)
model = deepq.models.cnn_to_mlp(convs=[(32, 8, 4), (64, 4, 2),
(64, 3, 1)],
hiddens=[256],
dueling=True)
demo_replay = []
act = dqfd.learn(env,
q_func=model,
num_actions=3,
lr=1e-4,
max_timesteps=10000000,
buffer_size=100000,
exploration_fraction=0.5,
exploration_final_eps=0.01,
train_freq=2,
learning_starts=100000,
target_network_update_freq=1000,
gamma=0.99,
prioritized_replay=True,
callback=deepq_callback)
act.save("defeat_zerglings.pkl")
def main():
FLAGS(sys.argv)
logdir = "tensorboard"
if(FLAGS.algorithm == "deepq"):
logdir = "tensorboard/zergling/%s/%s_%s_prio%s_duel%s_lr%s/%s" % (
FLAGS.algorithm,
FLAGS.timesteps,
FLAGS.exploration_fraction,
FLAGS.prioritized,
FLAGS.dueling,
FLAGS.lr,
start_time
)
elif(FLAGS.algorithm == "acktr"):
logdir = "tensorboard/zergling/%s/%s_num%s_lr%s/%s" % (
FLAGS.algorithm,
FLAGS.timesteps,
FLAGS.num_cpu,
FLAGS.lr,
start_time
)
if(FLAGS.log == "tensorboard"):
Logger.DEFAULT \
= Logger.CURRENT \
= Logger(dir=None,
output_formats=[TensorBoardOutputFormat(logdir)])
elif(FLAGS.log == "stdout"):
Logger.DEFAULT \
= Logger.CURRENT \
= Logger(dir=None,
output_formats=[HumanOutputFormat(sys.stdout)])
with sc2_env.SC2Env(
map_name="DefeatZerglingsAndBanelings",
step_mul=step_mul,
visualize=True,
game_steps_per_episode=steps * step_mul) as env:
model = deepq.models.cnn_to_mlp(
convs=[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
hiddens=[256],
dueling=True
)
demo_replay = []
act = dqfd.learn(
env,
q_func=model,
num_actions=3,
lr=1e-4,
max_timesteps=10000000,
buffer_size=100000,
exploration_fraction=0.5,
exploration_final_eps=0.01,
train_freq=2,
learning_starts=100000,
target_network_update_freq=1000,
gamma=0.99,
prioritized_replay=True,
callback=deepq_callback
)
act.save("defeat_zerglings.pkl")
