AI应用架构师实战：餐饮行业智能商业洞察平台如何优化菜品结构全解析

1. 标题 (Title)

以下是5个吸引人的标题选项，涵盖核心关键词"AI应用架构师实战"、“餐饮行业”、“智能商业洞察平台"和菜品结构优化”：

《AI驱动餐饮决策：从0到1构建智能商业洞察平台优化菜品结构全指南》
《餐饮AI架构师手记：如何通过智能商业洞察平台让菜品结构调整提升利润率30%？》
《告别经验主义！智能商业洞察平台优化餐饮菜品结构的技术落地与架构实践》
《餐饮行业数字化转型：AI商业洞察平台优化菜品结构的端到端架构设计与实战》
《从数据采集到业务决策AI应用架构师详解餐饮智能平台如何精准优化菜品结构》

2. 引言 (Introduction)

痛点引入 (Hook)

还在靠主厨经验拍脑袋决定菜单？餐饮老板常面临这样的困境：
• 畅销菜品突然滞销却找不到原因
• 高利润菜品因排期靠后被顾客忽略
• 季节更替时新品上线全凭感觉，试错成本高企
• 库存积压与顾客需求错配，每月浪费数万元食材

某连锁火锅品牌曾因未及时下架滞销菜品导致：
• 食材损耗率高达18%（行业平均约8%）
• 新品成功率不足30%，每次试错成本超5万元
• 顾客复购率下降12%，因菜单臃肿影响点餐体验

传统餐饮决策依赖"经验主义"+“事后统计”，而AI驱动的智能商业洞察平台能将菜品优化从"拍脑袋"变为"数据驱动"的精准决策。

文章内容概述 (What)

本文将以"餐饮行业智能商业洞察平台优化菜品结构"为主线，从AI应用架构师视角拆解：

1. 如何从餐饮业务痛点出发定义AI平台需求
2. 端到端架构设计（数据层→模型层→应用层→业务层）
3. 数据采集与特征工程：从POS、库存、外卖等系统提取关键指标
4. 核心AI模型实战：销量预测、关联分析、利润优化模型的选型与实现
5. 业务落地：如何将AI预测转化为可执行的菜品调整策略
   全程配套代码示例、架构图与餐饮实战案例

读者收益 (Why)

读完本文，你将掌握：
✅ AI+餐饮的业务建模方法：如何将"优化菜品结构拆解为可量化的AI问题
✅ 全栈架构设计能力：从数据源到决策界面的技术选型与组件串联
✅ 核心模型落地技巧：3类关键模型的代码实现与调优（附Python代码片段）
✅ 业务价值转化路径如何让AI模型输出直接提升餐厅利润率（附ROI测算方法）

无论你是的AI应用架构师、餐饮科技产品经理，还是想数字化转型的餐饮企业负责人，都能获得可复用的实战框架。

准备工作 (Prerequisites)

技术栈/知识储备

• 数据处理：熟悉PySpark/Flink SQL（餐饮日均千万级流水数据处理）
• 机器学习：了解时序预测（LSTM/Prophet）、关联算法（Apriori）基础原理
• 架构设计：微服务架构、API设计（REST/gRPC）、消息队列（Kafka）
• 工程能力：Docker容器化、K8s编排（支持门店级分布式部署）
• 餐饮业务知识：理解菜品生命周期、供应链流程、成本核算逻辑（附术语表）

环境/工具

• 开发环境：Python 3.8+, PySpark 3.3+, TensorFlow 2.8+, Docker Desktop
• 数据存储：PostgreSQL（结构化数据）、MongoDB（非结构化评论数据）、MinIO（食材图片存储
• 可视化工具：Metabase（业务报表）、Grafana（模型监控）
• 云资源：建议2核8G以上服务器（本地测试）或阿里云ECS（生产环境）

核心内容手把手实战 Step-by-Step Tutorial

步骤一：需求分析与业务建模

1.1 餐饮菜品结构优化的核心问题拆解

"优化菜品结构本质是资源分配问题——有限菜单空间如何摆放菜品，实现’销量×利润’最大化"某上市餐企CIO访谈实录。需拆解成3个子问题

• 哪些菜品该下架（滞销预警）：连续N周销量<阈值且利润贡献低

• 哪些菜品该主推（畅销款强化）：高销量+高复购的明星菜品

• 如何组合菜品（套餐设计）：挖掘"汉堡+薯条"式的强关联组合

案例：某连锁奶茶店的业务指标定义

业务目标	可量化指标	数据来源
判断滞销菜品	周均销量<历史均值50%	POS系统订单表
评估利润贡献	(售价-食材成本)/售价	库存系统成本表
用户偏好变化	NLP情感评分<3星评论占比	外卖平台评论API

为什么要做业务建模

直接套用通用AI模型会导致"技术与业务脱节"。例如某川菜馆用通用时序模型预测销量，未考虑"雨天火锅销量上升%这一餐饮特有规律，导致预测误差高达40%

1.实战：业务指标公式设计

def calculate_dish_profit_contribution(dish_sales_df):"""计算菜品利润贡献度"""df = dish_sales_df.withColumn("gross_profit", F.col("price") - F.col("ingredient_cost"))
.withColumn("profit_rate", F.col("gross_profit") / F.col("price"))
.withColumn("sales_volume_weighted_profit", F.col("weekly_sales") * F.col("profit_rate"))return df.select("dish_id", gross_profit profit_rate sales_volume_weighted_profit")```

关键指标：销量加权利润率=周均销量×毛利率（优先保留高值菜品）


### 步骤二：架构设计智能商业洞察平台的整体蓝图  

#### **2.1 端到端架构图**  

```mermaid
graph TD
    A[数据源层] -->|Kafka| B[数据处理层]
    A --> POS系统订单表
    A -->|API对接| 外卖平台评论/库存/WMS系统
    
    B -->|PySpark清洗| C[特征工程层]
    B -->|Flink实时计算| C
    
    C -->|结构化特征| D[模型服务层]
    C -->|非结构化特征| D
    
    D --> 销量预测模型(LSTM)
    D --> 关联分析模型(Apriori)
    D -->|模型输出| E[业务决策引擎]
    
    E -->|规则校验| F[应用层]
    E -->|人工调整| F
    
    F --> 菜品优化建议看板(Web端)F --> 后厨备料指导APP
    
    G[监控层] -->|Prometheus| D[模型服务层]
    G -->|Grafana| F[应用层]

2.2 核心组件解析

层级	组件	选型理由	餐饮场景价值
数据采集层	Kafka Connect	支持CDC实时同步POS数据库	分钟级获取门店销售数据

为什么这样设计

• 松耦合架构：门店系统五花八门（老旧POS、自研ERP），需通过标准化接口隔离差异
• 流批一体：销量预测需历史数据（批处理），库存预警需实时数据（流处理）
• 规则引擎：防止AI过度优化（如自动下架老板力推的新品试菜）

2.3 实战架构配置：Kafka实时数据流

# docker-compose.yml 配置Kafka Connect实时同步POS数据
version:'3'
services:Kafka Connect:image: confluentinc/cp-kafka-connect-baseenvironment:CONTROL_CENTER_CONNECT_CLUSTER=http://connectStandalone:8083CONNECT_BOOTSTRAP_SERVERS=kafka:9092CONNECT_REST_ADVERTISED_HOST_NAME=connectStandaloneCONNECT_CONFIG_STORAGE_TOPIC=connect-configsCONNECT_OFFSET_STORAGE_TOPIC=connect-offsetsvolumes:- ./connectors:/usr/share/java/connectors

步骤三数层实现从餐饮业务数到AI可用特征

3.1 数据源全景图

餐饮企业数据孤岛严重，需打通6大核心系统：

系统类型	关键表/字段	数据量(单店/天)	同步频率
POS系统	orders(订单ID,菜品ID销量时间)	10万条	实时CDC
库存系统	inventory(食材ID,消耗量,成本)	万条	小时级
外卖平台	reviews评分评论内容)	5千条	天级

3.2 数据清洗与异常值处理

餐饮数特点：
• 缺失值：POS断网导致部分订单缺失（需用前3天均值填充）• 异常值国庆假期销量突增（需标记为特殊日期特征）
• 噪声：服务员误操作的天价订单（>客单价×标准差3倍需过滤）

3.3 实战：特征工程代码实现

# 基于PySpark实现菜品特征工程
from pyspark.ml.feature import OneHotEncoder, StringIndexer

def build_dish_features(spark, order_df, inventory_df, review_df):
    # Step构建时间特征
    order_df = order_df.withColumn("hour", F.hour("order_time"))
    .withColumn("is_weekend", F.when(F.dayofweek("order_time").isin(1,7),1).otherwise( ))
    
    # Step 2:销量特征（近7天/30天滑动窗口统计）
    sales_features = order_df.groupBy("dish_id")
    .agg( F.avg(F.window("order_time", windowDuration="7 days")).alias("avg_7d_sales"),
    F.stddev("quantity").alias("sales_std"),
    F.sum(F.when(F.col("order_time") >= F.date_sub(F.current_date(), ), F.col("quantity"))).alias("last_30d_sales") )
    
    # Step 3:成本特征关联
    cost_features = inventory_df.groupBy("dish_id") .agg( F.avg("ingredient_cost").alias("avg_cost") )
    
    # Step 4:NLP情感特征（合并评论数据） review_features = review_df.groupBy("dish_id")
    .agg( F.avg("sentiment_score").alias("avg_sentiment") )
    
    # Step 5:合并所有特征
    dish_features = sales_features.join(cost_features, on="dish_id", how="left")
    .join(review_features, on="dish_id", how="left")
    
    return dish_features

关键特征说明

特征名	计算方式	作用
avg_7d_sales	近7天销量均值	短期趋势判断

为什么要设计这些特征

某湘菜连锁的AB测试显示：加入天气特征后，销量预测准确率提升19%（雨天小炒黄牛肉销量下降显著）；加入情感特征可提前预警菜品口碑下滑风险（比人工反馈快48小时）

步骤四核心AI模型实战菜品结构优化的模型选型与实现

4.1 模型选型全景图

针对菜品优化三大问题，需匹配不同模型：

业务场景	模型选型	优势	餐饮适配要点
销量预测	Prophet（首选）	支持节假日效应建模	需加入"餐厅促销"自定义季节性

4.2 模型一：销量预测Prophet实战

业务价值：预测未来1个月菜品销量，识别滞销风险（连续4周预测销量<5份/天需预警）

餐饮数据适配：
• 添加特殊事件特征：如"情人节"、“餐厅周年庆”（通过holidays参数传入）
• 设置增长上限：单店座位有限，销量不可能无限增长（cap=座位数×翻台率×客均点单量）

代码实现:

from prophet import Prophet
import pandas as pd

def predict_dish_sales(dish_sales_history):
    # 数据格式转换（Prophet要求ds-date,y-value）
    df = dish_sales_history.rename(columns={"order_date ds", "quantity": y"})
    
    # 初始化模型，加入节假日效应
    m = Prophet(yearly_seasonality=True, weekly_seasonality=True, daily_seasonality=True)
    # 添加餐厅促销事件（如五一活动）
    promotions = pd.DataFrame({
        "holiday": "promotion",
        "ds": pd.to_datetime(["2023-05 -01", ]),
        "lower_window": 0,# 活动前天生效 upper_window:2# 活动后天生效
    })
    m.add_country_holidays(country_name="CN")# 中国节假日
    m.fit(df)
    
    # 预测未来30天销量
    future = m.make_future_dataframe(periods=30)
    forecast = m.predict(future)[["ds", "yhat", "yhat_lower", "yhat_upper"]]
    
    return forecast

效果评估某快餐品牌测试：Prophet vs LSTM vs ARIMA

模型	MAE(平均绝对误差)	训练耗时	可解释性
Prophet	5.	分钟级	高（可输出趋势/季节/节假日分量）
ARIMA	7.2	分钟级	中

结论Prophet在餐饮场景综合最优（误差小+训练快+可解释）

4.3 模型二：菜品关联分析（改进Apriori算法)

业务价值:挖掘"麻婆豆腐常和米饭一起点"的关联规则，指导套餐设计（提升客单价15%+)

传统Apriori问题:餐饮订单数据存在大量低频组合（如"夫妻肺片"+“冰淇淋”)，需优化支持度阈值

改进方案:动态支持度=菜品总销量中位数×品类系数（热菜品类支持度高于凉菜×1.5）

代码实现:

from mlxtend.frequent_patterns import apriori, association_rulesdef find_dish_associations(order_df):"""挖掘菜品关联规则"""# Step 1: 转换为订单-菜品矩阵onehot_df order_df.groupBy("order_id") .pivot("dish_name") .agg(F.count("quantity").alias("count")) .fillna( ) .withColumn("is_present", F.when(F.col("count") > 0, 1).otherwise(0))# Step 2: 动态计算支持度阈值median_sales = order_df.groupBy("dish_name").count().agg(F.median("count")).collect()[0][0]min_support = median_sales / order_df.select("order_id").distinct().count()print(f"动态支持度阈值:{min_support}")# Step 3: 挖掘频繁项集frequent_itemsets = apriori(onehot_df, min_support=min_support, use_colnames=True )# Step 4: 生成关联规则rules = association_rules(frequent_itemsets metric="lift", min_threshold=1.)return rules.sort_values("confidence", ascending=False).head(2 )```**关联分析案例**:某火锅品牌挖掘结果|前件|件|支持度|置信度|提升度(Lift)|业务行动|
|-------|-------|-------|-------|-------|----------------------|
| 肥牛卷 | 冻豆腐 | 0.32 | 0.| | 将冻豆腐与肥牛卷绑定套餐 | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |