排除滚动平均值中的值（删除高值和低值）按子句分区

Question

我正在尝试计算 6 周内的滚动平均值，其中我在特定时间范围内放弃了高销售周和低销售周。

我使用 windows 函数来确定销售高峰和低谷周，但我无法在 6 周窗口内运行平均值并排除高低。

我尝试在 avg 函数中使用 case 语句，但它返回了错误的结果。

这是我的代码：

;with average_daily_sales
as
(
select sum(SalesUnits) as total_sales_units, 
ItemNumber, 
Store, 
END_OF_WEEK,
BEGIN_OF_WEEK
from CONFORM_MOVEMENT
where Store = 10 and ItemNumber =1026295
group by ItemNumber, Store, END_OF_WEEK, BEGIN_OF_WEEK


)

--USING windows function to accomplish a 6 week rolling average to identify the high and low selling weeks
--PARTITION BY creates pairs of stores and item numbers to analyze per each window (6 records represent 6 weeks)
--Set the window being analyzed by replacing the integer value between "ROWS BETWEEEN N PRECEEDING"
,highs_lows_identifier
as
(

select
max(total_sales_units) 
    over (PARTITION BY Store, ItemNumber ORDER BY END_OF_WEEK
            ROWS BETWEEN 5 PRECEDING AND CURRENT ROW ) as highs,
min(total_sales_units) 
    over (PARTITION BY Store, ItemNumber ORDER BY END_OF_WEEK
            ROWS BETWEEN 5 PRECEDING AND CURRENT ROW ) as lows,
Store,
ItemNumber,
END_OF_WEEK,
BEGIN_OF_WEEK,
total_sales_units
from average_daily_sales
group by Store, ItemNumber, END_OF_WEEK, BEGIN_OF_WEEK, total_sales_units


)
--Remove highs and lows from their respective record
,remove_highs_and_lows
as
(
select          
        avg(case when total_sales_units = highs 
                        or total_sales_units = lows
                        then null else total_sales_units end) 
            over ( partition by Store, ItemNumber ORDER BY END_OF_WEEK
                    rows between 5 preceding and current row) as average_sales_units,

        Store,
        ItemNumber,
        BEGIN_OF_WEEK,
        END_OF_WEEK,
        highs,
        lows,
        total_sales_units,
        total_sales_units /7 as daily_sales_units
from highs_lows_identifier
)

select * from remove_highs_and_lows
order by END_OF_WEEK asc

结果图片：

预期：记录 19 中的 average_sales_units 应为 61.5（不包括记录 16 和 17）。但是，没有发生排除，我的结果是 64。这个逻辑应该出现在每条记录中（例如记录 18 average_sales_units 应该排除记录 16 和 13）。

任何建议都会很棒！

谢谢

Answer 1

最简单的方法大概是apply。这对数据的外观做了一些假设，但想法是：

select cm.*, rolling_avg_sales
from CONFORM_MOVEMENT cm outer apply
     (select avg(cm2.sales) as rolling_avg_sales
      from (select cm2.*,
                   row_number() over (order by sales desc) as seqnum_asc,
                   row_number() over (order by sales desc) as seqnum_desc
            from CONFORM_MOVEMENT cm2
            where cm2.store = cm.store and cm2.item = cm.item and
                  cm2.end_of_week <= cm.end_of_week and
                  cm2.end_of_week > dateadd(week, -5, cm.end_of_week)
           ) cm2
      where 1 not in (seqnum_desc, seqnum_asc)
     ) cm2

Answer 2

谢谢 Gordon - 再次为没有提供示例数据而道歉，但你让我感动，我想我把这个解决方案复杂化了。

鉴于我可以使用我的窗口函数识别滚动 6 周中的高点和低点，我可以简单地将滚动 6 周的总和减去高点和低点，除以 4 得到每周平均值，然后再除以 7 得到获取每日平均值。

为了遵守业务逻辑，我将忽略前 6 周，因为前几周没有 6 周的时间跨度可供使用。

这里是：

--Remove highs and lows from their respective record - subtracting highs and lows to get the average
,remove_highs_and_lows
as
(
select
    --case
    --  when highs = lows then END_OF_WEEK
    --  when highs = total_sales_units 
    --      or lows = total_sales_units then END_OF_WEEK else null end as is_highs_lows,
        
        ((sum(total_sales_units)
            over ( partition by Store, ItemNumber ORDER BY END_OF_WEEK
                    rows between 5 preceding and current row) - highs - lows) / 4)
                    / 7 as daily_sales_units,
        sum(total_sales_units)
            over ( partition by Store, ItemNumber ORDER BY END_OF_WEEK
                    rows between 5 preceding and current row) as total_weekly_sales_units,
        (sum(total_sales_units)
            over ( partition by Store, ItemNumber ORDER BY END_OF_WEEK
                    rows between 5 preceding and current row) - highs - lows) / 4 as avg_weekly_sales_units,
        total_sales_units,
        highs,
        lows,
        Store,
        ItemNumber,
        BEGIN_OF_WEEK,
        END_OF_WEEK
from highs_lows_identifier

)

select * from remove_highs_and_lows