2021-08-20

数据推断

目标： 知道数据推断使用的业务场景使用SQL 完成关键指标推断逻辑

业务背景 确失关键指标
场景一家餐厅想要通过数据分析数据挖掘提升销量但是在历史记录的数据中缺少了重要的一个维度 用餐人数

梳理思路：
1、如何根据数据确定就餐人数
- 可以将菜单数据分成几类根据不同种类食物的点单量来推断就餐人数
1、例如主食类饮料小吃酱料主菜
- 确定判断的规则：
1、1份主食对应1个人（1:1）
2、1份主菜对应1个人（1:1）
3、1分饮料对应1个人（1:1）
4、2份小吃对应1个人（0.5:1）
5、蘸料作料不做计算
判断逻辑图

上升层面 SQL 的处理流程
SQL数据推断处理业务流程
看一个业务
数据图
数据表详情图

菜单表
- 商品名称
- 商品价格
原始的数据中缺少菜品的分类标签需要先添加菜品类别标签
- 方法使用locate 函数进行字符串匹配使用 case when 语句判断不同条件创建新字段 food_category(菜品类别)
- locate 函数介绍 locate（substr ， str）返回substr 在str中的位置
- substr 要查询的字符串
- str 字段名字或者是字符串

# 检查表里是否有food_category这个字段 如果有将其删除
DROP TABLE IF EXISTS food_category;
CREATE TABLE food_category AS SELECT   #as 是起别名用的
Item_Name AS item_name,
`Product_Price` AS price,
Restaurant_id AS restaurant_id,
#利用case when 给产品打标签
CASE
	WHEN locate( 'Dahi', Item_Name )> 0 THEN '酸奶'
	WHEN locate( 'wine', Item_Name )> 0
		OR locate( 'COBRA', Item_Name )> 0 THEN '酒'
	WHEN locate( 'water', Item_Name )> 0
		OR locate( 'Coke', Item_Name )> 0
		OR locate( 'Lemonade', Item_Name )> 0 THEN '饮料'
	WHEN locate( 'Rice', Item_Name )> 0 THEN '米饭'
	WHEN locate( 'Chapati', Item_Name )> 0
		OR locate( 'Paratha', Item_Name )> 0OR locate( 'Naan', Item_Name )> 0
		OR locate( 'roti', Item_Name )> 0
		OR locate( 'Papadum', Item_Name )> 0 THEN '饼'
	WHEN locate ( 'Main', Item_Name )> 0 THEN '主菜'
	WHEN locate ( 'Chaat', Item_Name )> 0
		OR locate( 'Muttar', Item_Name )> 0 THEN '小吃'
	WHEN locate( 'Chicken', Item_Name )> 0 THEN '鸡肉类'
	WHEN locate( 'Lamb', Item_Name )> 0 THEN '羊肉类'
	WHEN locate( 'Fish', Item_Name )> 0 THEN '鱼肉类'
	WHEN locate( 'Prawn', Item_Name )> 0
		OR locate( 'Jinga', Item_Name )> 0 THEN '虾类'
	WHEN locate( 'Pakora', Item_Name )> 0 THEN '炸素丸子'
	WHEN LOCATE( 'Saag', Item_Name )> 0 THEN '绿叶菜胡胡'
	WHEN locate( 'Paneer', Item_Name )> 0 THEN '芝士菜'
	WHEN locate( 'Pickle', Item_Name )> 0
		OR locate( 'Chutney', Item_Name )> 0 THEN '腌菜'
	WHEN locate( 'Aloo', Item_Name )> 0 THEN '土豆类'
	WHEN locate( 'Salad', Item_Name )> 0 THEN  '沙拉'
	WHEN locate( 'Tikka', Item_Name )> 0 THEN '烤串'
	WHEN locate( 'Chana', Item_Name )> 0 THEN '豆类'
	WHEN locate( 'Dall', Item_Name )> 0
		OR locate( 'Hari Mirch', Item_Name )> 0 THEN '素菜'
	WHEN locate( 'Puree', Item_Name )> 0 THEN '胡胡'
	WHEN locate( 'Raitha', Item_Name )> 0
		OR locate( 'Raita', Item_Name )> 0 THEN '酸奶沙拉'
	WHEN locate( 'French Fries', Item_Name )> 0 THEN '炸薯条'
	WHEN locate( 'Samosa', Item_Name )> 0 THEN '咖喱角'
	WHEN locate( 'Kehab', Item_Name )> 0
		OR locate( 'Kebab', Item_Name )> 0 THEN '烤串(小食)'
	WHEN locate( 'Bhajee', Item_Name )> 0
		OR locate( 'Bhaji', Item_Name )> 0 THEN '油炸蔬菜团'
	WHEN locate( 'Mushroom', Item_Name )> 0
		OR locate( 'Vegetable', Item_Name )> 0 THEN '蔬菜'
	WHEN locate( 'Starter', Item_Name)> 0 THEN '开胃小吃'
	WHEN locate( 'Sauce', Item_Name)> 0 THEN '酱' ELSE '咖喱菜'
	END AS food_category
FROM restaurant_products_price;

以上是给产品品进行分组通过 case when 方便后续做数据推断

下一步：首先查看一下产品明细表，知道我们数据库内现有的产品划分找到与我们推理时用到的分类之间的差异。

SELECT
#使用count 对item_name 字段进行去重计数  取别名
	count( DISTINCT item_name ) AS item_num,
	food_category  #是我们给产品分类创建的字段
FROM
	food_category
GROUP BY
	food_category
ORDER BY
	item_num DESC;

在现有的产品明细表里重新定义一列字段，作为我们就餐人数判断的产品分类

drop table if exists food_type;
create table food_type as
SELECT
item_name,
price,
restaurant_id,
food_category,
CASE
WHEN food_category IN ( '鸡肉类', '羊肉类', '虾类', '咖喱菜', '鱼肉类', '主菜', '芝
士菜' ) THEN
'主菜'
WHEN food_category IN ( '饼', '米饭' ) THEN
'主食'
WHEN food_category IN ( '饮料', '酒', '酸奶' ) THEN
'饮料'
WHEN food_category IN ( '酱', '腌菜' ) THEN
'佐料' ELSE '小食'
END AS food_type
FROM
food_category;

再来看一下重新定义过后产品的分类情况

SELECT
count( DISTINCT item_name ) AS item_num,
food_type,
food_category
FROM
food_type
GROUP BY
food_type,
food_category
ORDER BY
food_type,
food_category,
item_num DESC;

将交易明细表与我们刚定义的产品明细表进行关联,把产品分类和价格加入明细当中

SELECT
a.*,
b.food_type,
b.price
FROM
restaurant_orders a
JOIN food_type b ON a.`Item_Name` = b.item_name
AND a.Restaurant_Id = b.restaurant_id;

目前一行记录是一笔订单的一种产品的售卖情况，如果一笔订单有若干样产品会分成若干行，我们希望把一笔订单的详情，从多行统一到一行中，同时用我们事先定义的系数计算

select
a.`Order_Number`,a.`Order_Date`,a.restaurant_id,round(sum(a.Quantity*b.price),2) as total_amount,
sum(case when food_type='主食' then a.Quantity*1 else 0 end) as staple_food_count,
sum(case when food_type='主菜' then a.Quantity*1 else 0 end) as main_course_count,
sum(case when food_type='饮料' then a.Quantity*1 else 0 end) as drink_count,
sum(case when food_type='小食' then a.Quantity*0.5 else 0 end) as snack_count
from restaurant_orders a join food_type b
on a.`Item_Name`=b.item_name and a.Restaurant_Id=b.restaurant_id
group by a.`Order_Number`,a.`Order_Date`,a.Restaurant_Id;

比较主食,主菜，饮料，小食中的最大者

select
c.*,
GREATEST(c.staple_food_count,c.main_course_count,c.drink_count,c.snack_count) as max_count
from
(select
a.`Order_Number`,a.`Order_Date`,a.restaurant_id,round(sum(a.Quantity*b.price),2) as
total_amount,
sum(case when food_type='主食' then a.Quantity*1 else 0 end) as staple_food_count,
sum(case when food_type='主菜' then a.Quantity*1 else 0 end) as main_course_count,
sum(case when food_type='饮料' then a.Quantity*1 else 0 end) as drink_count,
sum(case when food_type='小食' then a.Quantity*0.5 else 0 end) as snack_count
from restaurant_orders a join food_type b
on a.`Item_Name`=b.item_name and a.Restaurant_Id=b.restaurant_id
group by a.`Order_Number`,a.`Order_Date`,a.Restaurant_Id) c;

最后，增加向下取整的逻辑，并且确保最小就餐人数为1

select c.*,
GREATEST(floor(GREATEST(c.staple_food_count,c.main_course_count,c.drink_count,c.snack
_count)),1) as customer_count from
(select
a.`Order_Number`,a.`Order_Date`,a.restaurant_id,round(sum(a.Quantity*b.price),2) as
total_amount,
sum(case when food_type='主食' then a.Quantity*1 else 0 end) as staple_food_count,
sum(case when food_type='主菜' then a.Quantity*1 else 0 end) as main_course_count,
sum(case when food_type='饮料' then a.Quantity*1 else 0 end) as drink_count,
sum(case when food_type='小食' then a.Quantity*0.5 else 0 end) as snack_count
from restaurant_orders a join food_type b
on a.`Item_Name`=b.item_name and a.Restaurant_Id=b.restaurant_id
group by a.`Order_Number`,a.`Order_Date`,a.Restaurant_Id) c;

新建一张表格，将两家店每一笔交易的时间，金额，就餐人数，主食，饮料，小食数存储进去

DROP TABLE IF EXISTS restaurants_orders_customer_count;outlier_count total_count outlier_rate
13 33055 0.0004
检查一下之前担心的饮料过多导致的推测异常占比
有了就餐人数之后，我们还可以进行进一步分析
CREATE TABLE restaurants_orders_customer_count AS SELECT
c.*, GREATEST( floor( GREATEST( c.staple_food_count, c.main_course_count,
c.drink_count, c.snack_count )), 1 ) AS customer_count
FROM
(SELECT
a.`Order_Number`,
a.`Order_Date`,
a.restaurant_id,
round( sum( a.Quantity * b.price ), 2 ) AS total_amount,
sum( CASE WHEN food_type = '主食' THEN a.Quantity * 1 ELSE 0 END ) AS
staple_food_count,
sum( CASE WHEN food_type = '主菜' THEN a.Quantity * 1 ELSE 0 END ) AS
main_course_count,
sum( CASE WHEN food_type = '饮料' THEN a.Quantity * 1 ELSE 0 END ) AS
drink_count,
sum( CASE WHEN food_type = '小食' THEN a.Quantity * 0.5 ELSE 0 END ) AS
snack_count
FROM
restaurant_orders a
JOIN food_type b ON a.`Item_Name` = b.item_name
AND a.Restaurant_Id = b.restaurant_id
GROUP BY
a.`Order_Number`,
a.`Order_Date`,
a.Restaurant_Id
) c;

检查一下之前担心的饮料过多导致的推测异常占比

SELECT
count( CASE WHEN drink_count >= 5 THEN `Order_Number` ELSE NULL END ) AS
outlier_count,
count(*) AS total_count,
round( count( CASE WHEN drink_count >= 5 THEN `Order_Number` ELSE NULL END )/
count(*), 5 ) AS outlier_rate
FROM
restaurants_orders_customer_count;

在这里插入图片描述

有了就餐人数之后，我们还可以进行进一步分析

SELECT
restaurant_id,
avg( customer_count ) AS avg_cc,
avg( total_amount ) AS ta,
avg( total_amount / customer_count ) AS avg_scc,
avg( staple_food_count / customer_count ) AS avg_staple,
avg( main_course_count / customer_count ) AS avg_main,
avg( drink_count / customer_count ) AS avg_drink,
avg( snack_count / customer_count ) AS avg_snake
FROM
restaurants_orders_customer_count
group by restaurant_id;

数据推断的价值：
数据本身就是资产，而获取新数据的成本不断增加，数据推断的价值愈发突出
数据推断是基于已经存在的数据资产，发掘出来的“新”数据，相当于是已较低的成本增加了数据资产

数据推断

猜你喜欢