def dqn_train_keras_rl(self, num_par_agents, model_name):
"""Implementation of kreras-rl deep q learing."""
from agents.agent_consider_equity import Player as EquityPlayer
from agents.agent_keras_rl_dqn import Player as DQNPlayer
from agents.agent_random import Player as RandomPlayer
env_name = 'neuron_poker-v0'
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(123)
env.seed(123)
env.add_player(EquityPlayer(name='equity/50/70',
min_call_equity=.5, min_bet_equity=.7))
env.add_player(EquityPlayer(name='equity/20/30',
min_call_equity=.2, min_bet_equity=.3))
env.add_player(RandomPlayer())
env.add_player(RandomPlayer())
env.add_player(RandomPlayer())
# shell is used for callback to keras rl
env.add_player(PlayerShell(name='keras-rl', stack_size=self.stack))
env.reset()
env_names = np.full((1, num_par_agents), model_name)
dqn = DQNPlayer()
with multiprocessing.Pool(num_par_agents) as pool:
pool.apply_async(parallel_dqn_train(dqn, env, env_name))
def dqn_train_custom_q1(self):
"""Create 6 players, 4 of them equity based, 2 of them random"""
from agents.agent_consider_equity import Player as EquityPlayer
from agents.agent_custom_q1 import Player as Custom_Q1
from agents.agent_random import Player as RandomPlayer
env_name = 'neuron_poker-v0'
self.env = gym.make(env_name, initial_stacks=self.stack, render=self.render)
# self.env.add_player(EquityPlayer(name='equity/50/50', min_call_equity=.5, min_bet_equity=-.5))
# self.env.add_player(EquityPlayer(name='equity/50/80', min_call_equity=.8, min_bet_equity=-.8))
# self.env.add_player(EquityPlayer(name='equity/70/70', min_call_equity=.7, min_bet_equity=-.7))
self.env.add_player(EquityPlayer(name='equity/20/30', min_call_equity=.2, min_bet_equity=-.3))
# self.env.add_player(RandomPlayer())
self.env.add_player(RandomPlayer())
self.env.add_player(RandomPlayer())
self.env.add_player(Custom_Q1(name='Deep_Q1'))
for _ in range(self.num_episodes):
self.env.reset()
self.winner_in_episodes.append(self.env.winner_ix)
league_table = pd.Series(self.winner_in_episodes).value_counts()
best_player = league_table.index[0]
print("League Table")
print("============")
print(league_table)
print(f"Best Player: {best_player}")
def dqn_train_keras_rl(self, model_name):
"""Implementation of kreras-rl deep q learing."""
env_name = 'neuron_poker-v0'
stack = 100
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(
EquityPlayer(name='equity/20/30',
min_call_equity=.2,
min_bet_equity=.3))
env.add_player(RandomPlayer())
env.add_player(RandomPlayer())
env.add_player(RandomPlayer())
env.add_player(PlayerShell(
name='keras-rl',
stack_size=stack)) # shell is used for callback to keras rl
env.reset()
dqn = DQNPlayer()
dqn.initiate_agent(env)
dqn.train(env_name=model_name)
def equity_vs_random(self):
"""Create 6 players, 4 of them equity based, 2 of them random"""
env_name = 'neuron_poker-v0'
stack = 500
self.env = gym.make(env_name, initial_stacks=stack, render=self.render)
self.env.add_player(
EquityPlayer(name='equity/50/50',
min_call_equity=.5,
min_bet_equity=-.5))
self.env.add_player(
EquityPlayer(name='equity/50/80',
min_call_equity=.8,
min_bet_equity=-.8))
self.env.add_player(
EquityPlayer(name='equity/70/70',
min_call_equity=.7,
min_bet_equity=-.7))
self.env.add_player(
EquityPlayer(name='equity/20/30',
min_call_equity=.2,
min_bet_equity=-.3))
self.env.add_player(RandomPlayer())
self.env.add_player(RandomPlayer())
for _ in range(self.num_episodes):
self.env.reset()
self.winner_in_episodes.append(self.env.winner_ix)
league_table = pd.Series(self.winner_in_episodes).value_counts()
best_player = league_table.index[0]
print("League Table")
print("============")
print(league_table)
print(f"Best Player: {best_player}")
def dqn_play_keras_rl(self):
"""Create 6 players, one of them a trained DQN"""
env_name = 'neuron_poker-v0'
stack = 500
self.env = gym.make(env_name, initial_stacks=stack, render=self.render)
self.env.add_player(
EquityPlayer(name='equity/50/50',
min_call_equity=.5,
min_bet_equity=.5))
self.env.add_player(
EquityPlayer(name='equity/50/80',
min_call_equity=.8,
min_bet_equity=.8))
self.env.add_player(
EquityPlayer(name='equity/70/70',
min_call_equity=.7,
min_bet_equity=.7))
self.env.add_player(
EquityPlayer(name='equity/20/30',
min_call_equity=.2,
min_bet_equity=.3))
self.env.add_player(RandomPlayer())
self.env.add_player(PlayerShell(name='keras-rl', stack_size=stack))
self.env.reset()
dqn = DQNPlayer(load_model='dqn1', env=self.env)
dqn.play(nb_episodes=self.num_episodes, render=self.render)
def dqn_play(self):
"""Create 6 players, one of them a trained DQN"""
env_name = 'neuron_poker-v0'
stack = 500
num_of_plrs = 6
self.env = gym.make(env_name,
num_of_players=num_of_plrs,
initial_stacks=stack,
render=self.render)
self.env.add_player(
EquityPlayer(name='equity/50/50',
min_call_equity=.5,
min_bet_equity=-.5))
self.env.add_player(
EquityPlayer(name='equity/50/80',
min_call_equity=.8,
min_bet_equity=-.8))
self.env.add_player(
EquityPlayer(name='equity/70/70',
min_call_equity=.7,
min_bet_equity=-.7))
self.env.add_player(
EquityPlayer(name='equity/20/30',
min_call_equity=.2,
min_bet_equity=-.3))
self.env.add_player(RandomPlayer())
self.env.add_player(DQNPlayer(load_model='neuron_poker-v0'))
for _ in range(self.num_episodes):
self.env.reset()
def dqn_play_keras_rl(self, model_name):
"""Create 6 players, one of them a trained DQN"""
from agents.agent_consider_equity import Player as EquityPlayer
from agents.agent_keras_rl_dqn import Player as DQNPlayer
from agents.agent_random import Player as RandomPlayer
env_name = 'neuron_poker-v0'
self.env = gym.make(env_name,
initial_stacks=self.stack,
render=self.render)
self.env.add_player(
EquityPlayer(name='equity/50/50',
min_call_equity=.5,
min_bet_equity=.5))
self.env.add_player(
EquityPlayer(name='equity/50/80',
min_call_equity=.8,
min_bet_equity=.8))
self.env.add_player(
EquityPlayer(name='equity/70/70',
min_call_equity=.7,
min_bet_equity=.7))
self.env.add_player(
EquityPlayer(name='equity/20/30',
min_call_equity=.2,
min_bet_equity=.3))
self.env.add_player(RandomPlayer())
self.env.add_player(PlayerShell(name='keras-rl',
stack_size=self.stack))
self.env.reset()
dqn = DQNPlayer(load_model=model_name, env=self.env)
dqn.play(nb_episodes=self.num_episodes, render=self.render)
def random_agents(self):
"""Create an environment with 6 random players"""
num_of_plrs = 6
self.env = HoldemTable(num_of_players=num_of_plrs, initial_stacks=500)
for _ in range(num_of_plrs):
player = RandomPlayer(500)
self.env.add_player(player)
self.run_episode()
def equity_vs_random(self):
"""Create 6 players, 4 of them equity based, 2 of them random"""
self.env = HoldemTable(num_of_players=5, initial_stacks=500)
self.env.add_player(EquityPlayer(name='equity/50/50', min_call_equity=.5, min_bet_equity=-.5))
self.env.add_player(EquityPlayer(name='equity/50/80', min_call_equity=.8, min_bet_equity=-.8))
self.env.add_player(EquityPlayer(name='equity/70/70', min_call_equity=.7, min_bet_equity=-.7))
self.env.add_player(EquityPlayer(name='equity/20/30', min_call_equity=.2, min_bet_equity=-.3))
self.env.add_player(RandomPlayer())
self.env.add_player(RandomPlayer())
for _ in range(self.num_episodes):
self.run_episode()
self.winner_in_episodes.append(self.env.winner_ix)
league_table = pd.Series(self.winner_in_episodes).value_counts()
best_player = league_table.index[0]
print(league_table)
print(f"Best Player: {best_player}")
def random_agents(self):
"""Create an environment with 6 random players"""
from agents.agent_random import Player as RandomPlayer
env_name = 'neuron_poker-v0'
num_of_plrs = 2
self.env = gym.make(env_name, initial_stacks=self.stack, render=self.render)
for _ in range(num_of_plrs):
player = RandomPlayer()
self.env.add_player(player)
self.env.reset()
def random_agents(self):
"""Create an environment with 6 random players"""
env_name = 'neuron_poker-v0'
stack = 500
num_of_plrs = 6
self.env = gym.make(env_name, num_of_players=num_of_plrs, initial_stacks=stack, render=self.render)
for _ in range(num_of_plrs):
player = RandomPlayer()
self.env.add_player(player)
self.env.reset()
def dqn_train():
"""Implementation of kreras-rl deep q learing."""
env_name = 'neuron_poker-v0'
stack = 100
env = gym.make(env_name, num_of_players=2, initial_stacks=stack, funds_plot=False)
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(EquityPlayer(name='equity/20/30', min_call_equity=.2, min_bet_equity=.3))
env.add_player(RandomPlayer())
env.add_player(RandomPlayer())
env.add_player(RandomPlayer())
env.add_player(PlayerShell(name='keras-rl', stack_size=stack)) # shell is used for callback to keras rl
env.reset()
dqn = DQNPlayer()
dqn.initiate_agent(env)
dqn.train(env_name='dqn1')
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()
def uto_plays(self):
"""Create an environment with 6 random players"""
env_name = 'neuron_poker-v0'
stack = 500
num_of_plrs = 6
self.env = gym.make(env_name,
num_of_players=num_of_plrs,
initial_stacks=stack,
render=self.render)
self.env.add_player(RandomPlayer())
self.env.add_player(
EquityPlayer(name='equity/50/80',
min_call_equity=.8,
min_bet_equity=.4))
self.env.add_player(
EquityPlayer(name='equity/70/70',
min_call_equity=.7,
min_bet_equity=.5))
self.env.add_player(
EquityPlayer(name='equity/20/30',
min_call_equity=.2,
min_bet_equity=.6))
self.env.add_player(UtoPlayer(name='Uto1 1'))
self.env.add_player(
UtoPlayer(name='Uto1 2',
min_call_equity=0.46,
min_bet_equity=0.56,
min_call_equity_allin=0.7))
for _ in range(self.num_episodes):
self.env.reset()
self.winner_in_episodes.append(self.env.winner_ix)
league_table = pd.Series(self.winner_in_episodes).value_counts()
best_player = league_table.index[0]
print(league_table)
print(f"Best Player: {best_player}")
def deep_q_learning():
"""Implementation of kreras-rl deep q learing."""
env_name = 'neuron_poker-v0'
stack = 100
env = gym.make(env_name, num_of_players=5, initial_stacks=stack)
np.random.seed(123)
env.seed(123)
env.add_player(
EquityPlayer(name='equity/50/50',
min_call_equity=.5,
min_bet_equity=-.5))
env.add_player(
EquityPlayer(name='equity/50/80',
min_call_equity=.8,
min_bet_equity=-.8))
env.add_player(
EquityPlayer(name='equity/70/70',
min_call_equity=.7,
min_bet_equity=-.7))
env.add_player(
EquityPlayer(name='equity/20/30',
min_call_equity=.2,
min_bet_equity=-.3))
env.add_player(RandomPlayer())
env.add_player(PlayerShell(
name='keras-rl',
stack_size=stack)) # shell is used for callback to keras rl
env.reset()
nb_actions = len(env.action_space)
# Next, we build a very simple model.
from keras import Sequential
from keras.optimizers import Adam
from keras.layers import Dense, Dropout
from rl.memory import SequentialMemory
from rl.agents import DQNAgent
from rl.policy import BoltzmannQPolicy
model = Sequential()
model.add(
Dense(64, activation='relu', input_shape=env.observation_space))
model.add(Dropout(0.2))
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(nb_actions, activation='linear'))
print(model.summary())
# Finally, we configure and compile our agent. You can use every built-in Keras optimizer and
# even the metrics!
memory = SequentialMemory(limit=50000, window_length=1)
policy = BoltzmannQPolicy()
dqn = DQNAgent(model=model,
nb_actions=nb_actions,
memory=memory,
nb_steps_warmup=10,
target_model_update=1e-2,
policy=policy)
dqn.compile(Adam(lr=1e-3), metrics=['mae'])
# Okay, now it's time to learn something! We visualize the training here for show, but this
# slows down training quite a lot. You can always safely abort the training prematurely using
# Ctrl + C.
dqn.fit(env, nb_steps=50000, visualize=True, verbose=2)
# After training is done, we save the final weights.
dqn.save_weights('dqn_{}_weights.h5f'.format(env_name), overwrite=True)
# Finally, evaluate our algorithm for 5 episodes.
dqn.test(env, nb_episodes=5, visualize=True)