从零开发AI天气查询技能：Rasa实战指南
1. 项目概述为什么需要掌握AI Skill开发AI Skill人工智能技能开发正在成为技术从业者的核心能力之一。不同于传统的API调用一个完整的AI Skill需要整合自然语言处理、意图识别、上下文管理、多轮对话等关键技术模块。我经历过从简单脚本到企业级AI Skill的完整开发周期发现大多数教程要么过于理论化要么缺乏工程实践指导。这个教程将带你从零开始构建一个可运行的天气查询AI Skill包含完整的项目结构、符合行业规范的代码实现以及我在实际开发中总结的12个关键避坑点。你会学到如何设计对话流、处理用户意图歧义、对接第三方API等实用技能最终产出可直接复用的代码仓库。2. 开发环境与工具链配置2.1 基础环境搭建推荐使用Python 3.8作为开发语言这是目前AI Skill开发的主流选择。不同于普通Python项目AI Skill需要特别注意依赖隔离# 创建虚拟环境Windows/Mac通用 python -m venv ai_skill_env source ai_skill_env/bin/activate # Linux/Mac ai_skill_env\Scripts\activate # Windows # 核心依赖安装 pip install rasa3.6.0 # 对话管理框架 pip install uvicorn fastapi # API服务 pip install python-dotenv # 环境变量管理注意Rasa框架的3.x版本与2.x存在架构差异本教程基于3.6版本编写。若使用其他版本需注意NLU管道配置的变化。2.2 项目目录结构规范一个标准的AI Skill项目应包含以下目录结构基于实际企业开发规范weather_skill/ ├── actions/ # 自定义动作代码 │ ├── __init__.py │ └── weather_actions.py ├── data/ # 训练数据 │ ├── nlu.yml # 意图样本 │ ├── rules.yml # 对话规则 │ └── stories.yml # 对话流程 ├── models/ # 训练好的模型 ├── tests/ # 测试用例 │ ├── test_actions.py │ └── test_dialogs.py ├── config.yml # 管道配置 ├── credentials.yml # 服务认证 ├── domain.yml # 技能定义 └── endpoints.yml # 服务端点这种结构的关键优势在于实现业务逻辑与对话逻辑解耦便于持续集成(CI)流水线处理符合主流AI平台部署要求3. 核心功能实现详解3.1 意图识别与实体提取在data/nlu.yml中定义天气查询意图的样本数据version: 3.1 nlu: - intent: ask_weather examples: | - 今天天气怎么样 - 北京明天会下雨吗 - 查询上海未来三天的天气 - 纽约现在气温多少度 - intent: provide_location examples: | - [北京](location) - 我在[上海](location) - [纽约市](location)的天气配置实体提取管道config.yml关键部分pipeline: - name: WhitespaceTokenizer - name: RegexFeaturizer - name: LexicalSyntacticFeaturizer - name: CountVectorsFeaturizer - name: DIETClassifier # 双意图实体识别模型 epochs: 100实战经验DIETClassifier的epochs设置需根据样本量调整。通常100-200个样本建议50-100轮超过500样本可减少到30-50轮。3.2 对话流设计实战在data/stories.yml中定义典型对话路径stories: - story: happy path weather query steps: - intent: ask_weather - action: utter_ask_location - intent: provide_location entities: - location: 北京 - action: action_get_weather对应的domain.yml需包含intents: - ask_weather - provide_location entities: - location actions: - utter_ask_location - action_get_weather templates: utter_ask_location: - text: 请问您想查询哪个城市的天气3.3 天气API对接实现在actions/weather_actions.py中实现核心业务逻辑import requests from typing import Dict, Text, Any, List from rasa_sdk import Action, Tracker from rasa_sdk.executor import CollectingDispatcher class ActionGetWeather(Action): def name(self) - Text: return action_get_weather async def run( self, dispatcher: CollectingDispatcher, tracker: Tracker, domain: Dict[Text, Any] ) - List[Dict[Text, Any]]: location tracker.get_slot(location) if not location: dispatcher.utter_message(未获取到有效位置信息) return [] # 使用心知天气API示例需申请免费key api_key YOUR_API_KEY url fhttps://api.seniverse.com/v3/weather/now.json?key{api_key}location{location}languagezh-Hans try: response requests.get(url, timeout5) data response.json() weather data[results][0][now] msg f{location}当前天气{weather[text]}温度{weather[temperature]}℃ except Exception as e: msg f获取天气信息失败{str(e)} dispatcher.utter_message(textmsg) return []关键实现细节使用异步async/await提高并发性能添加超时(timeout)避免长时间阻塞完善的异常处理机制符合Rasa SDK的Action接口规范4. 测试与调试技巧4.1 交互式测试方法启动Rasa shell进行实时测试rasa shell --debug调试会话示例? 北京今天天气怎么样 - 请问您想查询哪个城市的天气 ? 北京 - 北京当前天气晴温度23℃4.2 自动化测试方案在tests/test_actions.py中添加单元测试from actions.weather_actions import ActionGetWeather from rasa_sdk.executor import CollectingDispatcher from rasa_sdk import Tracker from rasa.shared.core.slots import Slot def test_weather_action(): dispatcher CollectingDispatcher() tracker Tracker( default, {}, {location: 北京}, [], False, None, {}, action_listen, ) action ActionGetWeather() result action.run(dispatcher, tracker, {}) assert len(dispatcher.messages) 1 assert 北京 in dispatcher.messages[0][text]4.3 性能优化技巧模型缓存在endpoints.yml中配置模型热加载models: url: http://localhost:5002 wait_time_between_pulls: 60 # 每分钟检查更新对话缓存使用Redis存储对话状态tracker_store: type: redis url: localhost port: 6379 db: 0 key_prefix: weather_skill:异步处理对耗时操作标记asynciostaticmethod async def fetch_weather(location: str) - dict: async with aiohttp.ClientSession() as session: async with session.get(url) as resp: return await resp.json()5. 部署与持续集成5.1 Docker化部署创建DockerfileFROM python:3.8-slim WORKDIR /app COPY . . RUN pip install -r requirements.txt CMD [rasa, run, --enable-api, --cors, *]构建并运行docker build -t weather-skill . docker run -p 5005:5005 weather-skill5.2 CI/CD流水线配置示例GitHub Actions配置.github/workflows/main.ymlname: CI on: [push] jobs: test: runs-on: ubuntu-latest steps: - uses: actions/checkoutv2 - name: Set up Python uses: actions/setup-pythonv2 with: python-version: 3.8 - name: Install dependencies run: | python -m pip install --upgrade pip pip install -r requirements.txt pip install pytest - name: Test with pytest run: | pytest tests/6. 常见问题解决方案6.1 意图识别不准现象用户说北京天气被识别为其他意图排查步骤检查nlu.yml中是否包含相似样本运行rasa data validate检查数据冲突使用rasa test nlu评估识别准确率优化方案- intent: ask_weather examples: | - [北京](location)天气 - 查下[上海](location)的天气情况 - [纽约](location)天气怎么样6.2 实体提取失败现象未正确提取朝阳区等细分地点解决方案添加同义词配置domain.ymlsynonyms: - synonym: 北京 examples: 朝阳区, 海淀区, 北京市使用正则表达式增强config.ymlpipeline: - name: RegexEntityExtractor patterns: - pattern: (海淀|朝阳|丰台)区 name: location6.3 对话流程中断现象用户未提供位置时直接报错修复方案在rules.yml中添加兜底逻辑rules: - rule: 主动询问位置 steps: - intent: ask_weather - action: utter_ask_location在action中添加空值检查location tracker.get_slot(location) or tracker.latest_message.get(text)7. 进阶开发建议7.1 多语言支持在config.yml中配置多语言管道language: zh pipeline: - name: LanguageModelTokenizer - name: LanguageModelFeaturizer model_name: bert model_weights: bert-base-chinese7.2 上下文感知增强实现跨对话轮次的记忆功能class ActionRememberPreference(Action): def name(self): return action_remember_preference def run(self, dispatcher, tracker, domain): preferred_city tracker.get_slot(location) if preferred_city: dispatcher.utter_message( f已记住您常查询的城市是{preferred_city}, buttons[{title: f{preferred_city}天气, payload: f{preferred_city}天气}] )7.3 可视化监控集成Prometheus监控指标from prometheus_client import start_http_server, Counter REQUEST_COUNT Counter(weather_requests, Total weather requests) class ActionGetWeather(Action): async def run(self, dispatcher, tracker, domain): REQUEST_COUNT.inc() location tracker.get_slot(location) ...启动监控服务器start_http_server(8000)
