def evaluate_meta_method_for_blocks(self):
"""
Evaluating heuristic blocks.
:return: str, the name of the new heuristic selected.
"""
slow_model_index = self.logs.get_slow_iters()[-1] + 1
meta_games = self.get_meta_game()
slow_model_regrets = regret(meta_games, slow_model_index,
self._slow_model_nash)
slow_model_nashconv = np.sum(slow_model_regrets)
if len(self._block_nashconv) == 0:
base_model_regrets = regret(meta_games, 1)
base_model_nashconv = np.sum(base_model_regrets)
delta_nashconv = base_model_nashconv - slow_model_nashconv
else:
delta_nashconv = self._block_nashconv[-1] - slow_model_nashconv
self._block_nashconv.append(slow_model_nashconv)
self._heuristic_selector.update_weights(delta_nashconv,
NE_list=self._NE_list)
new_heuristic_index = self._heuristic_selector.sample(self._iterations)
return self._heuristic_list[new_heuristic_index]
def evaluate_meta_method(self):
"""
Evaluating the performance of each heuristic by calculating the regret descent
based on the base model.
:return: str, the name of the new heuristic selected.
"""
if len(self.logs.get_slow_iters()
) == self._slow_oracle_period: #first evaluation
base_model_index = 1
else:
#TODO: check the correctness of the index.
base_model_index = self.logs.get_slow_iters(
)[-self._slow_oracle_period]
slow_model_index = self.logs.get_slow_iters()[-1] + 1
meta_games = self.get_meta_game()
if self._standard_regret:
base_model_regrets = regret(meta_games, base_model_index,
self._base_model_nash)
slow_model_regrets = regret(meta_games, slow_model_index,
self._slow_model_nash)
else:
base_model_regrets = strategy_regret(meta_games, base_model_index,
self.get_nash_strategies(),
self._base_model_nash)
slow_model_regrets = strategy_regret(meta_games, slow_model_index,
self.get_nash_strategies(),
self._slow_model_nash)
base_model_nashconv = np.sum(base_model_regrets)
slow_model_nashconv = np.sum(slow_model_regrets)
delta_nashconv = base_model_nashconv - slow_model_nashconv
self._heuristic_selector.update_weights(delta_nashconv)
new_heuristic_index = self._heuristic_selector.sample(self._iterations)
return self._heuristic_list[new_heuristic_index]