rule-based policy

dqn_agent.py

def _rule_action(self):
   # self.rule_current_slot_index points to current slot
   # rule_requests defined in dialogue_config.py
   if self.rule_current_slot_index < len(rule_requests):
       slot = rule_requests[self.rule_current_slot_index]
       self.rule_current_slot_index += 1
       rule_response = {'intent': 'request', 'inform_slots': {}, 'request_slots': {slot: 'UNK'}}
   # self.rule_phase used to indicate if we are at second to last round or last round
   elif self.rule_phase == 'not done':
       rule_response = {'intent': 'match_found', 'inform_slots': {}, 'request_slots': {}}
       self.rule_phase = 'done'
   elif self.rule_phase == 'done':
       rule_response = {'intent': 'done', 'inform_slots': {}, 'request_slots': {}}
   else:
       raise Exception('Should not have reached this clause')

   # self._map_action_to_index(rule_response) takes an action and gets its index from all possible agent actions
   index = self._map_action_to_index(rule_response)
   return index, rule_response