def choose_move(self, state: CatanState):
self.scores_by_player = state.get_scores_by_player_indexed()
next_moves = state.get_next_moves()
if len(next_moves) <= 1:
return next_moves[0]
if state.is_initialisation_phase():
best_move, best_score = None, 0
for move in next_moves:
state.make_move(move)
score = self.initialization_phase_heuaristic(state)
state.unmake_move(move)
if score > best_score:
best_move = move
best_score = score
return best_move
mcts = MCTS(MCTSNode(state), next_moves, self.exploration_param)
mcts.do_n_rollouts(self.iterations)
return mcts.choose().move
def winning_heuristic(self, state: CatanState):
"""
our heuristic for winning the game! it is composed of 3 different heuristics.
:param state: the state of the game.
:return: a heuristic score for this state.
"""
self.scores_by_player = state.get_scores_by_player_indexed()
my_score = self.scores_by_player[self.get_id()]
if my_score >= 10:
return inf
max_score = max(self.scores_by_player)
if max_score >= 10:
return -inf
if state.is_initialisation_phase():
return self.heuristic_initialisation_phase(state)
if self.in_first_phase():
return self.heuristic_first_phase(state,
self.winner_first_phase_weights)
return self.heuristic_final_phase(state,
self.winner_last_phase_weights)
def default_heuristic(self, state: CatanState):
if state.is_initialisation_phase():
return self._random_choice([i for i in range(10)])
# as discussed with Shaul, this isn't zero-sum heuristic, but a max-gain approach where only own player's
# value is is taken in account
return float(state.get_scores_by_player()[self])