shutil.copyfile(episode_generate_script_path, episode_generate_script_output_path)
# copy round-robin match script to output dir
round_robin_script_path = os.path.join(root, "scripts", "round_robin_match.py")
round_robin_script_output_path = os.path.join(OUTPUT_DIR, os.path.basename(round_robin_script_path))
shutil.copyfile(round_robin_script_path, round_robin_script_output_path)
TEST_LENGTH = 1000000
# Setup algorithm
value_func = ApproxActionValueFunction()
task = TexasHoldemTask(final_round=1000, scale_reward=True, lose_penalty=True, shuffle_position=True)
task.set_opponent_value_functions([value_func]*9)
policy = EpsilonGreedyPolicy(eps=0.99)
policy.set_eps_annealing(0.99, 0.1, int(TEST_LENGTH*0.8))
algorithm = MonteCarlo()
algorithm.setup(task, policy, value_func)
# load last training result
LOAD_DIR_NAME = ""
LOAD_DIR_PATH = os.path.join(os.path.dirname(__file__), "results", LOAD_DIR_NAME, "checkpoint", "gpi_finished")
if len(LOAD_DIR_NAME) != 0:
algorithm.load(LOAD_DIR_PATH)
# Setup callbacks
callbacks = []
save_interval = 50000
save_dir_path = os.path.join(OUTPUT_DIR, "checkpoint")
os.mkdir(save_dir_path)
learning_recorder = LearningRecorder(algorithm, save_dir_path, save_interval)
# record log on terminal in log file sys.stdout = Logger(os.path.join(OUTPUT_DIR, "training.log")) # copy training script to output dir script_output_path = os.path.join(OUTPUT_DIR, os.path.basename(__file__)) shutil.copyfile(__file__, script_output_path) TEST_LENGTH = 10000 # Setup algorithm value_func = ApproxActionValueFunction() task = TexasHoldemTask(scale_reward=True, lose_penalty=True) task.set_opponent_value_functions([value_func] * 9) policy = EpsilonGreedyPolicy(eps=0.99) policy.set_eps_annealing(0.99, 0.1, TEST_LENGTH) algorithm = MonteCarlo(gamma=0.99) algorithm.setup(task, policy, value_func) # Setup callbacks callbacks = [] save_interval = 1000 save_dir_path = os.path.join(OUTPUT_DIR, "checkpoint") os.mkdir(save_dir_path) learning_recorder = LearningRecorder(algorithm, save_dir_path, save_interval) callbacks.append(learning_recorder) monitor_file_path = os.path.join(OUTPUT_DIR, "stop.txt") manual_interruption = ManualInterruption(monitor_file_path) callbacks.append(manual_interruption)