def get_next_state(self, player_ids: List[int], state: State,
actions: List[int]) -> State:
next_state = State.from_state(state)
for _ in range(self.action_sampling_rate):
for i, player in enumerate(player_ids):
angle = self.calculate_new_angle(state.get_angle(player),
actions[i])
next_state.set_angle(player, angle)
position = self.calculate_new_position(
state.get_position(player), angle)
next_state.set_position(player, position)
next_state.draw_head(self.get_head_position(position, angle))
next_state.draw_player(player, True)
# TODO: Look into saving in designated memory space by specifying destination of copy.
return next_state
def get_action(self, state):
state_copy = State.from_state(state)
if 1 < len(state_copy.alive) < self.n_of_opp or self.first_round:
self.update_opponents(state_copy)
values = []
initial_position = self.get_initial_positions([self.id], state_copy)
for action in [RIGHT, LEFT, STRAIGHT]:
self.successors_generated += 1
self.update_successor_state([self.id], state_copy, [action],
initial_position, True)
values.append(
self.alpha_beta(state_copy, self.depth, -np.inf, np.inf, False,
action))
self.update_successor_state([self.id], state_copy, [action],
initial_position, False)
max_vals = [i for i, val in enumerate(values) if val == max(values)]
chosen_action = min(max_vals) if len(max_vals) <= 2 else random.choice(
max_vals)
return chosen_action