SQL 中分区的加权移动平均线

Question

我有一个简单的 table，其中包含按日期 (YYYYMMDD) 销售的不同产品。天数多，品多。 table 看起来像这样：

date      product     sales
...
20210101  Apples       112
20210101  Bananas      13
20210102  Apples       12
20210102  Bananas      101
20210103  Apples       18
20210103  Bananas      155
20210101  Lemons       14
...

我需要计算每个产品的 3 天加权移动平均值。主要棘手的事情是分区。我试图在这段代码中压缩它：

    SELECT table.date,
       max(table.sales) as sales,
       CASE 
       WHEN table.date - (select min(date) from table) >1
       THEN sum(weighting * ma_table.sales)
       END as WMA 
FROM table
JOIN table AS ma_table ON table.date - ma_table.date BETWEEN 0 AND 2
JOIN (SELECT id_1, 1/(2 + 1.) as weighting FROM (VALUES (0, 1, 2)) as t(id_1)) weights ON
  id_1 = table.date - ma_table.date
GROUP BY table.date
ORDER BY table.date

是否可以使用此代码完成任何操作以在其中包含分区？

Answer 1

下面是SQL计算每个产品的3天移动平均线，使用 window 函数。
这是你想要的吗？

with
a as ( -- serial numbers upto N (=3)
    select row_number() over () as i
    from TBL
    limit 3
),
b as (
  select
      TBL.date, TBL.product, TBL.sales,
      lag("sales", cast(a.i - 1 as integer)) 
        over(partition by TBL.product, a.i order by date) as sales2,
      (3 - a.i + 1) as weight
  from TBL
  cross join a
)
select date, product, sales,
    sum(weight*sales2) / (3*(3+1)/2) as avg
from b
group by date, product, sales
order by product, date

DEMO

Answer 2

所以我猜是这样的

create table yourtable (
 id int primary key, 
 "date" date, 
 product varchar(30), 
 sales decimal(18,2)
)

insert into yourtable 
(id, "date" , product, sales) values
  (1, '2021-01-01', 'Apples',  112)
, (2, '2021-01-02', 'Apples',   12)
, (3, '2021-01-03', 'Apples',   18)
, (4, '2021-01-01', 'Bananas',  13)
, (5, '2021-01-02', 'Bananas', 101)
, (6, '2021-01-03', 'Bananas', 155)
, (7, '2021-01-01', 'Lemons',   14)

select "date", product, sales
, cast((prev2+prev1*2+sales*3)/(1+2+3) as decimal(18,2)) as WMA
from (
select "date", product, sales
, lag(sales, 2, 0) over (partition by product order by "date") as prev2
, lag(sales, 1, 0) over (partition by product order by "date") as prev1
from yourtable
) q
order by product, "date"

date       | product |  sales |    wma
:--------- | :------ | -----: | -----:
2021-01-01 | Apples  | 112.00 |  56.00
2021-01-02 | Apples  |  12.00 |  43.33
2021-01-03 | Apples  |  18.00 |  31.67
2021-01-01 | Bananas |  13.00 |   6.50
2021-01-02 | Bananas | 101.00 |  54.83
2021-01-03 | Bananas | 155.00 | 113.33
2021-01-01 | Lemons  |  14.00 |   7.00

db<>fiddle here

无限加权滚动平均值的数字可能如下所示。

select "date", product, sales
, cast(sum(cnt * sales) 
       over (partition by product ROWS BETWEEN UNBOUNDED PRECEDING AND current row)
     / sum(cnt) 
       over (partition by product ROWS BETWEEN UNBOUNDED PRECEDING AND current row
  ) as decimal(18,2)) as UnboundedWeightedMovingAverage
from (
select "date", product, sales
, count(*) over (partition by product order by "date" asc) as cnt
from yourtable
) q
order by product, "date"

date       | product |  sales | unboundedweightedmovingaverage
:--------- | :------ | -----: | -----------------------------:
2021-01-01 | Apples  | 112.00 |                         112.00
2021-01-02 | Apples  |  12.00 |                          45.33
2021-01-03 | Apples  |  18.00 |                          31.67
2021-01-01 | Bananas |  13.00 |                          13.00
2021-01-02 | Bananas | 101.00 |                          71.67
2021-01-03 | Bananas | 155.00 |                         113.33
2021-01-01 | Lemons  |  14.00 |                          14.00

db<>fiddle here

SQL 中分区的加权移动平均线

Weighted moving average with partitions in SQL

sql

moving-average

weighted