def __init__(self):
''' Load pretrained model
'''
env = rlcard.make('leduc-holdem')
self.agent = CFRAgent(env,
model_path=os.path.join(ROOT_PATH,
'leduc_holdem_cfr'))
self.agent.load()
def test_train(self):
env = rlcard.make('leduc-holdem', config={'allow_step_back': True})
agent = CFRAgent(env)
for _ in range(100):
agent.train()
state = {
'obs': np.array([1., 1., 0., 0., 0., 0.]),
'legal_actions': [0, 2]
}
action, _ = agent.eval_step(state)
self.assertIn(action, [0, 2])
def test_save_and_load(self):
env = rlcard.make('leduc-holdem', config={'allow_step_back': True})
agent = CFRAgent(env)
for _ in range(100):
agent.train()
agent.save()
new_agent = CFRAgent(env)
new_agent.load()
self.assertEqual(len(agent.policy), len(new_agent.policy))
self.assertEqual(len(agent.average_policy),
len(new_agent.average_policy))
self.assertEqual(len(agent.regrets), len(new_agent.regrets))
self.assertEqual(agent.iteration, new_agent.iteration)
class LeducHoldemCFRModel(Model):
''' A pretrained model on Leduc Holdem with CFR
'''
def __init__(self):
''' Load pretrained model
'''
env = rlcard.make('leduc-holdem')
self.agent = CFRAgent(env, model_path=os.path.join(ROOT_PATH, 'leduc_holdem_cfr'))
self.agent.load()
@property
def agents(self):
''' Get a list of agents for each position in a the game
Returns:
agents (list): A list of agents
Note: Each agent should be just like RL agent with step and eval_step
functioning well.
'''
return [self.agent, self.agent]
eval_env = rlcard.make('leduc-holdem')
# Set the iterations numbers and how frequently we evaluate the performance and save model
evaluate_every = 100
save_plot_every = 1000
evaluate_num = 10000
episode_num = 10000
# The paths for saving the logs and learning curves
log_dir = './experiments/leduc_holdem_cfr_result/'
# Set a global seed
set_global_seed(0)
# Initilize CFR Agent
agent = CFRAgent(env)
agent.load() # If we have saved model, we first load the model
# Evaluate CFR against pre-trained NFSP
eval_env.set_agents([agent, models.load('leduc-holdem-nfsp').agents[0]])
# Init a Logger to plot the learning curve
logger = Logger(log_dir)
for episode in range(episode_num):
agent.train()
print('\rIteration {}'.format(episode), end='')
# Evaluate the performance. Play with NFSP agents.
if episode % evaluate_every == 0:
agent.save() # Save model
logger.log_performance(env.timestep,
def load_cfr_leduc_agent(model_path):
agent = CFRAgent(env, model_path=model_path)
agent.load()
return agent
evaluate_every = 100
save_plot_every = 1000
evaluate_num = 10000
episode_num = 10000000
# The paths for saving the logs and learning curves
root_path = './experiments/mahjong_cfr_result/'
log_path = root_path + 'log.txt'
csv_path = root_path + 'performance.csv'
figure_path = root_path + 'figures/'
# Set a global seed
set_global_seed(0)
# Initilize CFR Agent
agent = CFRAgent(env)
# Init a Logger to plot the learning curve
logger = Logger(root_path)
for episode in range(episode_num):
agent.train()
print('\rIteration {}'.format(episode), end='')
agent.save()
# Evaluate the performance. Play with NFSP agents.
if episode % evaluate_every == 0:
reward = 0
for eval_episode in range(evaluate_num):
_, payoffs = eval_env.run(is_training=False)
reward += payoffs[0]
})
# Set the iterations numbers and how frequently we evaluate/save plot
evaluate_every = 100
save_plot_every = 1000
evaluate_num = 10000
episode_num = 10000000
# The paths for saving the logs and learning curves
log_dir = './experiments/leduc_holdem_br_result/'
# Set a global seed
set_global_seed(0)
# Initilize CFR Agent
opponent = CFRAgent(env)
#opponent = RandomAgent(action_num=env.action_num)
#opponent.load() # If we have saved model, we first load the model
#agent = RandomAgent(action_num=env.action_num)
agent = BRAgent(eval_env, opponent)
#agent = CFRAgent(env)
eval_env.set_agents([agent, opponent])
# Init a Logger to plot the learning curve
logger = Logger(log_dir)
for episode in range(episode_num):
opponent.train()
#agent.train()
print('\rIteration {}'.format(episode), end='')