def dqn_train_keras_rl(self):
"""Implementation of kreras-rl deep q learing."""
env_name = 'neuron_poker-v0'
stack = 2000
env = gym.make(env_name, initial_stacks=stack, funds_plot=self.funds_plot, render=self.render,
use_cpp_montecarlo=self.use_cpp_montecarlo)
np.random.seed(123)
env.seed(123)
# env.add_player(EquityPlayer(name='equity/50/70', min_call_equity=.5, min_bet_equity=.7))
# env.add_player(RandomPlayer())
# env.add_player(RandomPlayer())
# env.add_player(RandomPlayer())
# env.add_player(PlayerShell(name='keras-rl-1', stack_size=stack), range=0.9) # shell is used for callback to keras rl
# env.add_player(PlayerShell(name='keras-rl-2', stack_size=stack), range=0.9) # shell is used for callback to keras rl
# env.add_player(PlayerShell(name='keras-rl-3', stack_size=stack), range=0.9) # shell is used for callback to keras rl
# env.add_player(PlayerShell(name='keras-rl-4', stack_size=stack), range=0.9) # shell is used for callback to keras rl
# env.add_player(PlayerShell(name='keras-rl-5', stack_size=stack), range=0.9) # shell is used for callback to keras rl
# env.add_player(PlayerShell(name='keras-rl-6', stack_size=stack), range=0.9) # shell is used for callback to keras rl
# env.add_player(PlayerShell(name='keras-rl-7', stack_size=stack), range=0.9) # shell is used for callback to keras rl
env.add_player(PlayerShell(name='LJY', stack_size=stack, range=0.33)) # shell is used for callback to keras rl
# dqn = DQNPlayer(name='DQN-1',stack_size=2000, range=0.9, env=env , load_model=None)
# env.add_player(dqn)
env.add_player(RandomPlayer(name='Random-1',range=1))
# env.add_player(RandomPlayer(name='Random-2',range=1))
# env.add_player(RandomPlayer(name='Random-3',range=1))
# env.add_player(RandomPlayer(name='Random-4',range=1))
# env.add_player(RandomPlayer(name='Random-5',range=1))
# env.add_player(RandomPlayer(name='Random-6',range=1))
# env.add_player(RandomPlayer(name='Random-7',range=1))
# env.add_player(DQNPlayer(name='DQN-2',stack_size=2000, range=0.9, env=env , load_model=None))
# env.add_player(DQNPlayer(name='DQN-3',stack_size=2000, range=0.9, env=env , load_model=None))
# env.add_player(DQNPlayer(name='DQN-4',stack_size=2000, range=0.9, env=env , load_model=None))
# env.add_player(DQNPlayer(name='DQN-5',stack_size=2000, range=0.9, env=env , load_model=None))
# env.add_player(DQNPlayer(name='DQN-6',stack_size=2000, range=0.9, env=env , load_model=None))
# env.add_player(DQNPlayer(name='DQN-7',stack_size=2000, range=0.9, env=env , load_model=None))
# env.add_player(DQNPlayer(name='DQN-8',stack_size=2000, range=0.9, env=env , load_model=None))
env.reset()
# print(env.players[0].range)
# print(env.players[1].range)
# print(env.players[2].range)
# print(env.players[3].range)
# print(env.players[4].range)
# print(env.players[5].range)
dqn = DQNPlayer()
# dqn.initiate_agent(env,load_model='3dqn_vs_3rd')
dqn.initiate_agent(env)
dqn.train(ckpt_name='LJY')
def key_press_agents(self):
"""Create an environment with 6 key press agents"""
env_name = 'neuron_poker-v0'
stack = 2000
# num_of_plrs = 6
env = gym.make(env_name, initial_stacks=stack, render=self.render)
player = KeyPressAgent(name="LJY",range=0.3)
env.add_player(player)
# self.env.add_player(EquityPlayer(name='equity/50/50', min_call_equity=.5, min_bet_equity=-.5))
env.add_player(RandomPlayer(name='Random-1',range=1))
# env.add_player(RandomPlayer(name='Random-2',range=1))
# env.add_player(RandomPlayer(name='Random-3',range=1))
# env.add_player(RandomPlayer(name='Random-4',range=1))
# env.add_player(RandomPlayer(name='Random-5',range=1))
# self.env.add_player(PlayerShell(name='dqn001', stack_size=stack))
# self.env.add_player(PlayerShell(name='dqn002', stack_size=stack))
# self.env.add_player(PlayerShell(name='dqn003', stack_size=stack))
# self.env.add_player(PlayerShell(name='dqn004', stack_size=stack))
# self.env.add_player(PlayerShell(name='dqn005', stack_size=stack))
env.reset()
def dqn_agent(self, mode):
my_import = __import__('agents.'+self.agent, fromlist=['Player'])
player = getattr(my_import, 'Player')
env_path = 'env'
if self.env_name != 'v0':
env_path += '_' + self.env_name
shell_import = __import__(
'gym_env.' + env_path, fromlist=['PlayerShell'])
PlayerShell_import = getattr(shell_import, 'PlayerShell')
env_name = 'neuron_poker-' + self.env_name
self.env = gym.make(env_name, initial_stacks=self.stack, funds_plot=self.funds_plot, render=self.render,
use_cpp_montecarlo=self.use_cpp_montecarlo)
np.random.seed(42)
self.env.seed(42)
count = 1
for player_type in self.players:
if player_type == 0:
self.env.add_player(RandomPlayer(env_path))
elif type(player_type) == tuple and len(player_type) == 2:
self.env.add_player(EquityPlayer(name='equity_' + str(count), env=env_path,
min_call_equity=player_type[0], min_bet_equity=player_type[1]))
count += 1
self.env.add_player(PlayerShell_import(
name='keras-rl', stack_size=self.stack))
self.env.reset()
if mode == 'train':
dqn = player()
dqn.initiate_agent(self.env)
dqn.train(env_name=self.model_name)
elif mode == 'play':
dqn = player(load_model=self.model_name, env=self.env)
dqn.play(nb_episodes=self.num_episodes, render=self.render)
def ai_vs_random(self, ai_num):
"""
Created by Xue Hongyan
Create an environment with provided number of ai players and random players
"""
env_name = 'neuron_poker-v0'
stack = 500
num_of_plrs = 6
self.env = gym.make(env_name, initial_stacks=stack, render=self.render)
player_pool = []
for _ in range(ai_num):
player = Custom_AI(env=self.env)
player_pool.append(player)
for _ in range(num_of_plrs - ai_num):
player = RandomPlayer()
player_pool.append(player)
random.shuffle(player_pool)
for player in player_pool:
self.env.add_player(player)
self.env.reset()