前 10 名和后float_value名的平均值，带日期，按日期分组

Question

我有一个这样的表：

| prodID  | date     |  perm
---------------------------------   
|200      |8/7/2011  | 81.742 
|200      |8/7/2011  | 81.644
|200      |8/7/2011  | 81.302
|200      |8/7/2011  | 81.057
|201      |8/7/2011  | 80.932
|201      |8/7/2011  | 80.839
|201      |8/7/2011  | 80.622
|201      |8/7/2011  | 80.557
|201      |8/7/2011  | 80.541

（除了稍微大一点）发生的情况细分：我想取前10个值（和后10个值）的平均值，其中prodid=somevalue在本例中为200。

代码：

declare @myid int
set @myid = 200
;with high as  --top ten average
(
 select prodid, CONVERT(CHAR(10),  DATEADD(DAY, AVG(DATEDIFF(DAY, 0, CONVERT    
(SMALLDATETIME, [date]))), 0),     101) as date, max(perm)as max_perm, avg(perm) 
as   
high_perm from   
( select prodid, date, perm, 
 row_number() over(partition by date order by perm desc) as nt    
 from live_pilot_plant
 where prodid = @myid) as T 
 where nt <= 10
 group by prodid
),
low as   -- bottom ten average
(
select prodid, CONVERT(CHAR(10),  DATEADD(DAY, AVG(DATEDIFF(DAY, 0, CONVERT    
(SMALLDATETIME, [date]))), 0),101) as date, min(perm) as min_perm, avg(perm) 
as low_perm  from   
( select prodid, date, perm,  
  row_number() over(partition by date order by perm asc) as nt    
  from live_pilot_plant
  where prodid = @myid) as T 
  where nt <= 10
  group by prodid
)
select l.prodid, l.date, l.low_perm as low_avg, m.high_perm as high_avg,
(m.high_perm -    l.low_perm) as delta
from low l
left outer join high m
on l.prodid = m.prodid 

它产生了这样的东西：

|  prodID  |   date     |  low_avg   |  high_avg  |  delta   |
|   200    | 08/07/2011 |   68.752   |  79.1976   |  10.444  |

这些数字不准确--

这一切都很好，除了不太花枝招展。我的意思是有很多prodID，基于prodID做一个太慢了。如何根据日期获取low_avg和high_avg（按prodID分组）

类似这样的东西：

| date       | prodID  | low_avg  | high_avg  |  delta  |
| 08/07/2011 | 200     |  60      |  80       |  20     |
| 08/07/2011 | 201     |  70      |  100      | 100     |

注意：你可能已经注意到一个疯狂的转换日期。原因是一些prodID的日期重叠，即2011年8月7日和2011年8日的200，我需要对日期进行平均（这是一个varchar）。因此，类似于如果有100行2011年8月7日，然后有9行2011年7月8日，最终查询将产生日期为2011年8日的

Answer 1

以下查询一次对所有产品执行此操作：

select lpp.prod_id, lpp.date,
       AVG(case when seqnum_asc <= 10 then perm end) as avg_bottom10,
       AVG(case when seqnum_desc <= 10 then perm end) as avg_top10,
       (AVG(case when seqnum_desc <= 10 then perm end) - AVG(case when seqnum_asc <= 10 then perm end)) as delta
from (select lpp.*,
             ROW_NUMBER() over (partition by prodid, date order by perm) as seqnum_asc,
             ROW_NUMBER() over (partition by prodid, date order by perm desc) as seqnum_desc
      from live_pilot_plan lpp
     ) lpp
group by lpp.prod_id, lpp.ate