我家的树莓派在成为了"智能语音助手"后,经过rasa学习训练,已经可以帮忙查日期/时间,查天气预报,进行一些简单的闲聊。但是,我希望它的功能还可以再强大些,比如说,可以帮我记录todo任务。为了实现这一目标,又花了一周时间,终于在今天实现了这个功能。
要实现这个功能,说白了,就是定义一个todo class,然后通过rasa 的自定义actions来调用这个class,从而实现todo task的创建、查询、删除这几个基本功能。
插一句话:接下来分享的代码,都是基于我的1.4.0版rasa来说的,要在其他版本上使用,需要根据相应版本的规则自行适配。
1 增加nlu.md中的intents
2 domain.yml作相应调整(只列新增部分)
3 stories.md新增故事
4 在actions.py中定义domain中出现的actions和forms
4.1 显示任务列表
python
class ActionShowTasks(Action):
def name(self) -> Text:
return "action_show_tasks"
def run(
self,
dispatcher: CollectingDispatcher,
tracker: Tracker,
domain: Dict[Text, Any],
) -> List[Dict[Text, Any]]:
todos = todo.Todo()
todolist = todos.all()
result = ""
for todo_dict in todolist:
result += " 编号 " + str(todo_dict.id) + " 任务 " + todo_dict.content
dispatcher.utter_message(text=result)
return []
4.2 新增todo task
python
class AddTaskForm(FormAction):
def name(self) -> Text:
return "add_task_form"
@staticmethod
def required_slots(tracker: Tracker) -> List[Text]:
return ["content"]
def slot_mappings(self) -> Dict[Text, Union[Dict, List[Dict]]]:
return {
"content": [self.from_entity(entity="content"), self.from_text()]
}
def submit(
self,
dispatcher: CollectingDispatcher,
tracker: Tracker,
domain: Dict[Text, Any],
) -> Dict[Text, Any]:
todos = todo.Todo()
content = tracker.get_slot("content")
if content is not None and content != "":
todos.content = content
todos.save()
todolist = todos.all()
msg = "新增任务 编号 " + str(todolist[-1].id) + " 任务 " + todolist[-1].content
else:
msg = "任务创建失败"
dispatcher.utter_message(text=msg)
return [SlotSet("content", None)]
4.3 查询todo task
python
class ActionQueryTaskForm(FormAction):
def name(self) -> Text:
return "query_task_form"
@staticmethod
def required_slots(tracker: Tracker) -> List[Text]:
return ["id"]
def slot_mappings(self) -> Dict[Text, Union[Dict, List[Dict]]]:
return {
"id": [self.from_entity(entity="id"), self.from_text()]
}
def submit(
self,
dispatcher: CollectingDispatcher,
tracker: Tracker,
domain: Dict[Text, Any],
) -> Dict[Text, Any]:
todos = todo.Todo()
tid = tracker.get_slot("id")
msg = "任务不存在"
if tid is not None and tid != "":
if is_int(tid) and int(tid) >0:
task = todos.getById(tid)
if task is not None:
msg = "定位任务 编号 " + str(task.id) + " 任务 " + task.content
dispatcher.utter_message(text=msg)
return [SlotSet("id", None)]
4.4 删除指定todo task
python
class ActionDeleteTaskForm(FormAction):
def name(self) -> Text:
return "delete_task_form"
@staticmethod
def required_slots(tracker: Tracker) -> List[Text]:
return ["id"]
def slot_mappings(self) -> Dict[Text, Union[Dict, List[Dict]]]:
return {
"id": [self.from_entity(entity="id"), self.from_text()]
}
def submit(
self,
dispatcher: CollectingDispatcher,
tracker: Tracker,
domain: Dict[Text, Any],
) -> Dict[Text, Any]:
tid = tracker.get_slot("id")
msg = "任务不存在"
if tid is not None and tid != "":
if is_int(tid) and int(tid) > 0:
todos = todo.Todo()
taskid = int(tid)
todos.id = taskid
todos.delete()
msg = "任务已删除"
dispatcher.utter_message(text=msg)
return [SlotSet("id", None)]
如上就是四个功能的实现代码。其中,定位和删除需要判断输入的slot值是否为数字,就定义了一个检测函数来实现数字判断。
python
def is_int(string: Text) -> bool:
try:
int(string)
return True
except ValueError:
return False
actions.py文件头部需要引用的模块罗列如下:
python
from typing import Any, Text, Dict, List, Union, Optional
from datetime import datetime, timedelta
from rasa_sdk import Action, Tracker
from rasa_sdk.executor import CollectingDispatcher
from rasa_sdk.forms import FormAction
from rasa_sdk.events import SlotSet
import ssl
from urllib import request, parse
import json
import todo
完成上述四个文件的编写后,在模型训练前请确认一下config.yml,我的模型用的是MitieNLP和jieba,网上大多数人用的是SpacyNLP,所以你要根据自己的实际情况来修改。如下是我的config.yml。
python
# Configuration for Rasa NLU.
# https://rasa.com/docs/rasa/nlu/components/
language: zh
# pipeline: supervised_embeddings
pipeline:
- name: MitieNLP
model: data/total_word_feature_extractor_zh.dat
- name: JiebaTokenizer
- name: MitieEntityExtractor
- name: EntitySynonymMapper
- name: RegexFeaturizer
- name: MitieFeaturizer
- name: SklearnIntentClassifier
# Configuration for Rasa Core.
# https://rasa.com/docs/rasa/core/policies/
policies:
- name: MemoizationPolicy
- name: KerasPolicy
- name: MappingPolicy
- name: FormPolicy
执行rasa train完成模型训练,同时记得执行rasa run actions --actions actions来注册todo相关的四个actions。模型生成后,运行rasa shell测试通过,就可以让智能语音助手来执行对应的todo指令了。
写在后面:
1.最后的语音对话图请忽略我的识别耗时,直接用sherpa-ncnn的测试代码跑wav识别耗时很少,可能还是我的代码需要进一步优化。我所使用的语音助手整合代码请看这篇博文《树莓派智能语音助手之功能整合》
2.todo的新增,查询和删除使用了form,但我暂时没有找到rasa1.4.0对应form的全部完整定义规则,导致实际运行时会出现"This can throw of the prediction. Make sure to include training examples in your stories for the different types of slots this action can return.
"的warning,不影响运行结果,但应该是属于stories定义没写完整。有知道怎么解决的朋友也可以教教我。
-
由于我在查询和删除form中使用了同一个slot------"id",结果当我执行完查询马上执行删除时,tracker.get_slot('id')会直接读取前一个form中的slot值。因此需要reset slot,即把class结束的return[]改成"return [SlotSet("id", None)]"。
-
actions中import todo所引用的todo class代码请看这篇博文《用python实现todo功能》。
-
若不清楚怎么进行rasa的模型训练,可以看这篇博文《树莓派智能语音助手之首次RASA模型训练》。