def predict(self, game: Game, player: Player):
actions = []
e_x, e_y = game.get_closest_enemy_location(player.location[0], player.location[1], player.team_id)
actions.append((ATTACK_CLOSEST_TARGET, self.distance((e_x, e_y), player.location)))
r_x, r_y = game.get_nearest_resource_index(player.location[0], player.location[1])
actions.append((HARVEST_CLOSEST_RESOURCE, self.distance((r_x, r_y), player.location)))
b_x, b_y = game.get_enemy_base(player.team_id)
actions.append((ATTACK_BASE, self.distance((b_x, b_y), player.location)))
return min(actions, key=lambda x: x[1])[0]
def play(g: Game,
ddqn: DoubleDeepQNetwork,
NPC_Memory: NPC_History,
use_NPC=False):
# Initial 2 players
player1 = g.get_players(1)[0]
player2 = g.get_players(2)[0]
player1.main_player = True
# Setup action list
global action_list
if use_NPC:
action_list = collections.deque(maxlen=5)
start_game(g)
state = g.capture_grey_scale()
g.prev_stat = g.get_state_stat()
while True:
if g.is_terminal():
g.stop()
if use_NPC:
action_list.clear()
g.game_result()
print("Game over")
break
# Spots
print("Player 1 : " + str(player1.location) + " - " +
str(player1.health_p) + " - " + str(player1.gold))
print("Player 2 : " + str(player2.location) + " - " +
str(player2.health_p) + " - " + str(player2.gold))
# Player 1 action by model
# action = ddqn.predict_action(state)
action = 1
player1.do_action(ATTACK_CLOSEST_TARGET)
# Only for NPC as current model is no good
if use_NPC is True:
npc_action = np.random.randint(2)
if len(action_list) >= 5:
# TODO dummy_get_npc_state
player_health = player1.health_p
resource_matrix = g.get_resource_matrix()
# print(resource_matrix.shape)
x, y = g.get_closest_enemy_location(player1.location[0],
player1.location[1], 1)
dist_closest_enemy = (np.abs(player1.location[0] - x) +
np.abs(player2.location[1] - y))
reward_value = g.get_reward_value()
number_of_enemies = len(g.teams[Game.OPPONENTS[1]].players)
NPC_state = [
action_list[0], action_list[1], action_list[2],
action_list[3], action_list[4], player_health,
np.sum(np.sum(resource_matrix, axis=0), axis=0),
dist_closest_enemy, reward_value, number_of_enemies
]
# print(NPC_state)
# NPC_Memory.Add_Observation(NPC_state,action)
NPC_Memory.Add_Observation(NPC_state, npc_action)
print(NPC_Memory.num_obs)
if NPC_Memory.num_obs > 100 and NPC_Memory.do_once_flag is True:
print("GOT HERE __________----------")
NPC_History.do_once_flag = False
player_net = NPC_CatBoost('Follower')
player_net.train_(NPC_Memory.CAT_State,
NPC_Memory.CAT_Action)
player_net.eval_train()
if use_NPC is True:
action_list.append(action)
# Player 2 random action
# player2.do_action(np.random.randint(1, 4))
update_with_skip_rate(g, SKIP_RATE)
g.caption() # Show Window caption
g.update_state() # Update states to new model
g.view() # View the game state in the pygame window
next_state = g.capture_grey_scale()
reward = g.get_reward_value()
# ddqn.train(state, action, reward, next_state, g.is_terminal())
state = next_state
