def main(args):
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.InteractiveSession(config=config)
import lf2gym
from lf2gym import WebDriver, Character
env = lf2gym.make(startServer=True,
driverType=WebDriver.Chrome,
characters=[Character.Firen, Character[args.player]],
versusPlayer=True)
env.start_recording()
num_actions = env.action_space.n
global_network = Worker(-1,
num_actions,
1e-4,
sess,
AIchar=env.characters[0])
saver = tf.train.Saver()
latest_ckpt = tf.train.latest_checkpoint(args.loadpath)
saver.restore(sess, latest_ckpt)
print("restore from", latest_ckpt)
global_network.test(env)
from brain.fbrain import DQN
elif args.mode == 'mix':
from brain.mbrain import DQN
else:
print("Error mode! Use picture instead.")
from brain.pbrain import DQN
# game setup
AGENT = 'Firen'
OPPOENENT = 'Dennis'
# env setup
env = lf2gym.make(
startServer=True,
wrap='skip4',
driverType=lf2gym.WebDriver.PhantomJS,
characters=[lf2gym.Character[AGENT], lf2gym.Character[OPPOENENT]],
difficulty=lf2gym.Difficulty.Crusher,
debug=True,
action_options=['Basic', 'AJD'])
options = env.get_reset_options()
print('Original reset options: %s' % options)
options['hp_full'] = 100
options['mp_start'] = 250
print('Custom reset options: %s' % options)
# initial obervation number
mode = args.mode
if mode == 'picture':
state_n = env.observation_space.n
elif mode == 'feature':
t = time()
print('%s: %f seconds.' % (msg, t - s))
s = t
# Add import path and import the lf2gym
import os, sys
sys.path.append(os.path.abspath('..'))
# Import lf2gym
import lf2gym
# Make an env
env = lf2gym.make(
startServer=True,
wrap='skip4',
driverType=lf2gym.WebDriver.PhantomJS,
characters=[lf2gym.Character[AGENT], lf2gym.Character[OPPOENENT]],
difficulty=lf2gym.Difficulty.Dumbass,
debug=True)
# Set the reset options
options = env.get_reset_options()
print('Original reset options: %s' % options)
options['hp_full'] = 100
options['mp_start'] = 250
print('Custom reset options: %s' % options)
# Reset the env
env.reset(options)
# Start to record the screen
def main(args):
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
config.intra_op_parallelism_threads = 8
config.inter_op_parallelism_threads = 8
sess = tf.InteractiveSession(config=config)
if args.train:
workers = []
import lf2gym
from lf2gym import Difficulty
# url = 'http://127.0.0.1:' + str(args.port)
envs = [
lf2gym.make(startServer=True,
difficulty=Difficulty.Challengar,
port=args.port + i) for i in range(args.thread)
]
# [env.reduce_action_space(13) for env in envs]
try:
num_actions = envs[0].action_space.n
global_network = Worker(-1,
num_actions,
args.learning_rate,
sess,
AIchar=envs[0].characters[0])
for i in range(args.thread):
workers.append(
Worker(i,
num_actions,
args.learning_rate,
sess,
glob_net=global_network,
save_path=args.savepath,
AIchar=envs[i].characters[0]))
saver = tf.train.Saver(max_to_keep=35)
if args.retrain:
latest_ckpt = tf.train.latest_checkpoint(args.loadpath)
saver.restore(sess, latest_ckpt)
print("restore from", latest_ckpt)
else:
sess.run(tf.global_variables_initializer())
if not os.path.exists(args.savepath):
os.mkdir(args.savepath)
path = args.savepath + 'output_thread.log'
t_writer = open(path, 'w')
t_writer.write(args.comment)
t_writer.write('\nstart training a3c, thread 1 log...\n')
actor_learner_threads = [
threading.Thread(target=workers[i].train,
args=(envs[i], args.checkpoint_interval,
saver, t_writer))
for i in range(args.thread)
]
for t in actor_learner_threads:
t.start()
for t in actor_learner_threads:
t.join()
except KeyboardInterrupt:
print("W: interrupt received, stopping...")
finally:
[env.close() for env in envs]
if args.test:
import lf2gym # difficulty='CRUSHER 1.0'
from lf2gym import Character, Difficulty
# env = lf2gym.make(startServer=False, autoStart=True, rewardList=['hp'], difficulty=Difficulty.Challengar, wrap='skip4' )
env = lf2gym.make(startServer=True,
difficulty=Difficulty.Challengar,
port=args.port,
action_options=['Basic', 'AJD', 'No Combos'])
env.start_recording()
num_actions = env.action_space.n
global_network = Worker(-1,
num_actions,
args.learning_rate,
sess,
AIchar=env.characters[0])
saver = tf.train.Saver()
latest_ckpt = tf.train.latest_checkpoint(args.loadpath)
saver.restore(sess, latest_ckpt)
print("restore from", latest_ckpt)
global_network.test(env, render=False)
if args.test_chrome:
import lf2gym
from lf2gym import WebDriver, Character, Difficulty
env = lf2gym.make(startServer=False,
driverType=WebDriver.Chrome,
canvasSize=(600, 800),
characters=[Character.Firen, Character.Freeze],
versusPlayer=False,
difficulty=Difficulty.Crusher)
# env.reduce_action_space(13)
num_actions = env.action_space.n
global_network = Worker(-1,
num_actions,
args.learning_rate,
sess,
AIchar=env.characters[0])
saver = tf.train.Saver()
latest_ckpt = tf.train.latest_checkpoint(args.loadpath)
saver.restore(sess, latest_ckpt)
print("restore from", latest_ckpt)
global_network.test(env, render=False)
parser.add_argument("--interval", default=0.2, help="your enemy")
args = parser.parse_args()
# Add import path and import the lf2gym
import os, sys
sys.path.append(os.path.abspath('..'))
# Import lf2gym
import lf2gym
from time import sleep
# Make an env
env = lf2gym.make(startServer=True,
driverType=lf2gym.WebDriver.Chrome,
characters=[
lf2gym.Character[args.opponent],
lf2gym.Character[args.player]
],
versusPlayer=True)
# Reset the env
env.reset()
# Game starts
done = False
while not done:
# Sample a random action
action = env.action_space.sample()
# Take the chosen action
_, reward, done, _ = env.step(action)
print('Enemy took action %d.' % action)
sleep(args.interval)
player1_info.get('vy'), player1_info.get('vz'),
player2_info.get('hp'), player2_info.get('mp'), player2_info.get('x'),
player2_info.get('y'), player2_info.get('z'), player2_info.get('vx'),
player2_info.get('vy'), player2_info.get('vz')))
return player_info
if __name__ == "__main__":
# initialize gym environment and the agent
env = lf2gym.make(
startServer=True,
driverType=lf2gym.WebDriver.Chrome,
background=lf2gym.Background.HK_Coliseum,
characters=[args.player, args.opponent], #[Me/AI, bot]
difficulty=lf2gym.Difficulty.Crusher,
action_options=[
'Basic', 'AJD', 'Full Combos'
], # ['Basic', 'AJD', 'Full Combos'] # defines action space
headless=HEADLESS,
rewardList=['hp'],
versusPlayer=False
) # versusPlayer= False means Agent is playing against bot instedad of user
env.start_recording()
state_size_x = env.observation_space.n[0]
state_size_y = env.observation_space.n[1]
print("state_size: (" + str(state_size_x) + ", " + str(state_size_y) + ")")
action_size = int(env.action_space.n)
print("action_size: " + str(action_size))