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')
env.model.create_agents({"mocsar_min": 4})
env_agent_list = [env.model.rule_agents[i] for i in range(1, 4)]
env_agent_list.insert(0, agent)
env.set_agents(env_agent_list)
# Evaluation agent
eval_env.model.create_agents({"mocsar_random": 4})
eval_agent_list = [eval_env.model.rule_agents[i] for i in range(1, 4)]
eval_agent_list.insert(0, agent)
eval_env.set_agents(eval_agent_list)
# Init a Logger to plot the learning curve
logger = Logger(log_dir)
# Log Game info
logger.log('\n########## Game information ##########')
logger.log('\nNumPlayers: {}, NumCards: {}, Episodes: {}'.format(
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:
env.model.create_agents({"mocsar_min": 4})
env_agent_list = [env.model.rule_agents[i] for i in range(1, 4)]
env_agent_list.insert(0, agent)
env.set_agents(env_agent_list)
# Evaluation agent
eval_env.model.create_agents({"mocsar_random": 4})
eval_agent_list = [eval_env.model.rule_agents[i] for i in range(1, 4)]
eval_agent_list.insert(0, agent)
eval_env.set_agents(eval_agent_list)
# Init a Logger to plot the learning curve
logger = Logger(log_dir)
## Log Game info
logger.log('\n########## Game information ##########')
logger.log('\nNumPlayers: {}, NumCards: {}, Episodes: {}'.format(
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:
trajectories = OUTPUT_QUEUE.get()
# Feed transitions into agent memory, and train
for ts in trajectories[0]:
agent.feed(ts)
# Evaluate the performance
reward = 0
tasks = assign_task(evaluate_num, PROCESS_NUM)
variables = tf.contrib.slim.get_variables(scope="dqn", collection=tf.GraphKeys.TRAINABLE_VARIABLES)
variables = [var.eval() for var in variables]
for task in tasks:
INPUT_QUEUE.put((task, False, variables, agent.total_t))
for _ in range(evaluate_num):
payoffs = OUTPUT_QUEUE.get()
reward += payoffs[0]
logger.log('\n########## Evaluation ##########')
logger.log('Average reward is {}'.format(float(reward)/evaluate_num))
# Close files in the logger
logger.close_files()
# Plot the learning curve
logger.plot('DQN')
# Save model
save_dir = 'models/blackjack_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'))
env.model.create_agents({"mocsar_min": 4})
env_agent_list = [env.model.rule_agents[i] for i in range(1, 4)]
env_agent_list.insert(0, agent)
env.set_agents(env_agent_list)
# Evaluation agent
eval_env.model.create_agents({"mocsar_random": 4})
eval_agent_list = [eval_env.model.rule_agents[i] for i in range(1, 4)]
eval_agent_list.insert(0, agent)
eval_env.set_agents(eval_agent_list)
# Init a Logger to plot the learning curve
logger = Logger(log_dir)
# Log Game info
logger.log('\n########## Game information, NFSP, RuleAgents, Pytorch ##########')
logger.log('\nNumPlayers: {}, NumCards: {}, Episodes: {}'.format(env.game.num_players,
env.game.num_cards,
conf.get_int('episode_num')))
for episode in range(conf.get_int('episode_num')):
# 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)