num_episodes = (num_train_ep + num_test_ep) * num_repetitions
# Directory
save_dir = args.save_dir or mkdir()
print("Starting Training - id={}; num_opponents={}; num_teammates={}; "
"num_episodes={};".format(agent_id, num_op, num_team, num_episodes))
# Initialize connection with the HFO server
hfo_interface = HFOAttackingPlayer(agent_id=agent_id,
num_opponents=num_op,
num_teammates=num_team)
hfo_interface.connect_to_server()
# Agent set-up
reward_function = basic_reward
features_manager = DiscreteFeatures(num_team, num_op)
actions_manager = DiscreteActionsV5()
agent = QLearningAgentTest(num_states=features_manager.get_num_states(),
num_actions=actions_manager.get_num_actions(),
learning_rate=0.1,
discount_factor=0.9,
epsilon=0.8)
# Save metrics structures
avr_rewards_list = []
avr_epsilons_list = []
trained_eps_list = []
q_tables_list = []
# Test one first time without previous train:
av_reward = test(num_episodes=num_test_ep,
agent=agent,
game_interface=hfo_interface,
features=features_manager,
save_dir = args.save_dir
print("Q Table file: ", load_file)
print("Starting Test - num_opponents={}; num_teammates={}; "
"num_episodes={};".format(num_op, num_team, num_episodes))
# Initialize connection with the HFO server
hfo_interface = HFOAttackingPlayer(agent_id=agent_id,
num_opponents=num_op,
num_teammates=num_team)
hfo_interface.connect_to_server()
# Agent set-up
reward_function = reward_functions.basic_reward
features_manager = discrete_features_v2.DiscreteFeaturesV2(
num_team, num_op)
actions_manager = DiscreteActionsV5(
origin_pos=learning_agent.ORIGIN_POSITIONS["MID LEFT"])
agent = learning_agent.QLearningAgentV5(
num_states=features_manager.get_num_states(),
num_actions=actions_manager.get_num_actions(),
dir=save_dir)
agent.load_q_table(load_file)
# Run training using Q-Learning
av_reward = learning_agent.test(num_episodes=num_episodes,
agent=agent,
game_interface=hfo_interface,
features=features_manager,
actions=actions_manager,
reward_funct=reward_function)
print("Av reward = {}".format(av_reward))