def extract_double_object_actions()

in agent/affordance_extractors/lm_affordance_extractor.py [0:0]

• 21 lines of code
• 5 McCabe index (conditional complexity)

    def extract_double_object_actions(self, entity1, entity2):
        key = (id(entity1), id(entity2))
        if key in self.cached_double_object_actions:
            return self.cached_double_object_actions[key]

        action_prob_dict = {}
        num_actions = 0
        tie_breaker = 0.00000001
        for verb, prep in complex_verbs:
            phrase = verb + ' the ' + entity1.name + ' ' + prep + ' the ' + entity2.name
            log_prob = self.get_joint_log_prob(self.forward_model, phrase, 5)
            num_actions += 1
            action_prob_dict[DoubleAction(verb, entity1, prep, entity2)] = log_prob + num_actions * tie_breaker

        sorted_actions = sorted(action_prob_dict, key=action_prob_dict.get, reverse=True)
        scored_actions = []
        for action in sorted_actions:
            log_prob = action_prob_dict[action]
            prob = 1. / (1. + math.exp(min(20., -(log_prob + self.double_object_action_lp_offset) * self.double_object_action_lp_scale)))
            if prob > self.double_object_action_extraction_threshold:
                scored_actions.append((action, prob))

        self.cached_double_object_actions[key] = scored_actions
        return scored_actions