predictor_bonus_base.py 文件源码

def get_bonus(self,path):
        if self._fit_steps > self._yield_zeros_until:
            bonus = self._coeff * self._f_predict(path['observations']).reshape(-1)
            if self._filter_bonuses:
                bonus = bonus  * (np.invert(self._wrapped_constraint.evaluate(path)))
            return bonus
        else:
            return np.zeros(path["rewards"].size)