def test_log(self):
log_dir = "./newtest/test_log.txt"
if os.path.exists(log_dir):
shutil.rmtree(log_dir)
logger = Logger(log_dir)
logger.log("test text")
logger.log_performance(1, 1)
logger.log_performance(2, 2)
logger.log_performance(3, 3)
logger.close_files()
logger.plot('aaa')
# logger.log(f'\nTrain Agents:{get_agent_str(env_agent_list)}')
# logger.log(f'\nEval Agents:{get_agent_str(eval_agent_list)}')
for episode in range(episode_num):
# Generate data from the environment
trajectories, _ = env.run(is_training=True)
# Feed transitions into agent memory, and train the agent
for ts in trajectories[0]:
agent.feed(ts)
# Evaluate the performance. Play with random agents.
if episode % evaluate_every == 0:
logger.log_performance(env.timestep,
tournament(eval_env, evaluate_num)[0],
episode=episode)
# Save model
save_dir = 'models/mocsar_dqn_ra_pytorch'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
state_dict = agent.get_state_dict()
logger.log('\n########## Pytorch Save model ##########')
logger.log('\n' + str(state_dict.keys()))
torch.save(state_dict, os.path.join(save_dir, 'model.pth'))
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('DQN RA PyTorch')
env.game.num_players, env.game.num_cards, episode_num))
# logger.log(f'\nTrain Agents:{get_agent_str(env_agent_list)}')
# logger.log(f'\nEval Agents:{get_agent_str(eval_agent_list)}')
for episode in range(episode_num):
# Generate data from the environment
trajectories, _ = env.run(is_training=True)
# Feed transitions into agent memory, and train the agent
for ts in trajectories[0]:
agent.feed(ts)
# Evaluate the performance. Play with random agents.
if episode % evaluate_every == 0:
logger.log_performance(env.timestep,
tournament(eval_env, evaluate_num)[0],
episode=episode)
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('DQN RA')
# Save model
save_dir = 'models/mocsar_dqn_ra'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
saver = tf.compat.v1.train.Saver()
saver.save(sess, os.path.join(save_dir, 'model'))
# First sample a policy for the episode
for agent in agents:
agent.sample_episode_policy()
# Generate data from the environment
trajectories, _ = env.run(is_training=True)
# Feed transitions into agent memory, and train the agent
for i in range(env.player_num):
for ts in trajectories[i]:
agents[i].feed(ts)
# Evaluate the performance. Play with random agents.
if episode % evaluate_every == 0:
logger.log_performance(env.timestep,
tournament(eval_env, evaluate_num)[0])
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('NFSP')
# Save model
save_dir = 'models/limit_holdem_nfsp'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
saver = tf.train.Saver()
saver.save(sess, os.path.join(save_dir, 'model'))
# Init a Logger to plot the learning curve
logger = Logger(log_dir)
for episode in range(episode_num):
# Generate data from the environment
trajectories, _ = env.run(is_training=True)
# Feed transitions into agent memory, and train the agent
for ts in trajectories[0]:
agent.feed(ts)
# Evaluate the performance. Play with random agents.
if episode % evaluate_every == 0:
logger.log_performance(env.timestep,
tournament(eval_env, evaluate_num)[0])
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('DQN')
# Save model
save_dir = 'models/gin_rummy_dqn'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
saver = tf.train.Saver()
saver.save(sess, os.path.join(save_dir, 'model'))
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='')
# Evaluate the performance. Play with NFSP agents.
if episode % evaluate_every == 0:
logger.log_performance(env.timestep,
tournament(eval_env, evaluate_num)[0])
# Close files in the logger
logger.close_files()
logger.plot('BR')
# 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,
tournament(eval_env, evaluate_num)[0])
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('CFR')
agent.sample_episode_policy()
# Generate data from the environment
trajectories, _ = env.run(is_training=True)
# Feed transitions into agent memory, and train the agent
for i in range(env.player_num):
for ts in trajectories[i]:
agents[i].feed(ts)
# Evaluate the performance. Play with random agents.
if episode % evaluate_every == 0:
logger.log_performance(env.timestep,
tournament(eval_env, evaluate_num)[0],
episode=episode)
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('NFSP Torch')
# Save model
save_dir = 'models/mocsar_nfsp_pytorch'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
state_dict = {}
for agent in agents:
state_dict.update(agent.get_state_dict())
torch.save(state_dict, os.path.join(save_dir, 'model.pth'))
# First sample a policy for the episode
agent.sample_episode_policy()
# Generate data from the environment
trajectories, _ = env.run(is_training=True)
# Feed transitions into agent memory, and train the agent
for ts in trajectories[0]:
agent.feed(ts)
# Evaluate the performance. Play with random agents.
if episode % evaluate_every == 0:
logger.log_performance(env.timestep, tournament(eval_env, evaluate_num)[0], episode=episode)
# Save model
save_dir = 'models/mocsar_nfsp_pytorch_ra'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
state_dict = agent.get_state_dict()
logger.log('\n########## Pytorch Save model ##########')
logger.log('\n' + str(state_dict.keys()))
torch.save(state_dict, os.path.join(save_dir, 'model.pth'))
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('NFSP Torch Rule Agent')
