Sql 在秩（）结束时计数时跳过空值

给定一组行，字段有时为空，有时不为空：

SELECT 
   Date, TheThing
FROM MyData
ORDER BY Date


Date                     TheThing
-----------------------  --------
2016-03-09 08:17:29.867  a
2016-03-09 08:18:33.327  a
2016-03-09 14:32:01.240  NULL
2016-10-21 19:53:49.983  NULL
2016-11-12 03:25:21.753  b
2016-11-24 07:43:24.483  NULL
2016-11-28 16:06:23.090  b
2016-11-28 16:09:07.200  c
2016-12-10 11:21:55.807  c

我希望有一个计算非空值的排名列：

Date                     TheThing  DesiredTotal
-----------------------  --------  ------------
2016-03-09 08:17:29.867  a         1
2016-03-09 08:18:33.327  a         2
2016-03-09 14:32:01.240  NULL      2 <---notice it's still 2 (good)
2016-10-21 19:53:49.983  NULL      2 <---notice it's still 2 (good)
2016-11-12 03:25:21.753  b         3
2016-11-24 07:43:24.483  NULL      3 <---notice it's still 3 (good)
2016-11-28 16:06:23.090  b         4
2016-11-28 16:09:07.200  c         5
2016-12-10 11:21:55.807  c         6

但等级计算为空：

Date                     TheThing  Total
-----------------------  --------  -----
2016-03-09 08:17:29.867  a         1
2016-03-09 08:18:33.327  a         2
2016-03-09 14:32:01.240  NULL      3 <--- notice it is 3 (bad)
2016-10-21 19:53:49.983  NULL      4 <--- notice it is 4 (bad)
2016-11-12 03:25:21.753  b         5 <--- and all the rest are wrong (bad)
2016-11-24 07:43:24.483  NULL      7
2016-11-28 16:06:23.090  b         8
2016-11-28 16:09:07.200  c         9
2016-12-10 11:21:55.807  c         10

Date                     TheThing  Total
-----------------------  --------  -----
2016-03-09 08:17:29.867  a         1
2016-03-09 08:18:33.327  a         2
2016-03-09 14:32:01.240  NULL      1 <--- reset to 1?
2016-10-21 19:53:49.983  NULL      2 <--- why go up?
2016-11-12 03:25:21.753  b         3 
2016-11-24 07:43:24.483  NULL      3 <--- didn't reset?
2016-11-28 16:06:23.090  b         4 
2016-11-28 16:09:07.200  c         5
2016-12-10 11:21:55.807  c         6

但等级计算为空：

Date                     TheThing  Total
-----------------------  --------  -----
2016-03-09 08:17:29.867  a         1
2016-03-09 08:18:33.327  a         2
2016-03-09 14:32:01.240  NULL      3 <--- notice it is 3 (bad)
2016-10-21 19:53:49.983  NULL      4 <--- notice it is 4 (bad)
2016-11-12 03:25:21.753  b         5 <--- and all the rest are wrong (bad)
2016-11-24 07:43:24.483  NULL      7
2016-11-28 16:06:23.090  b         8
2016-11-28 16:09:07.200  c         9
2016-12-10 11:21:55.807  c         10

Date                     TheThing  Total
-----------------------  --------  -----
2016-03-09 08:17:29.867  a         1
2016-03-09 08:18:33.327  a         2
2016-03-09 14:32:01.240  NULL      1 <--- reset to 1?
2016-10-21 19:53:49.983  NULL      2 <--- why go up?
2016-11-12 03:25:21.753  b         3 
2016-11-24 07:43:24.483  NULL      3 <--- didn't reset?
2016-11-28 16:06:23.090  b         4 
2016-11-28 16:09:07.200  c         5
2016-12-10 11:21:55.807  c         6

编辑：有了所有的答案，我花了很多次迭代才找到我不想要的所有边缘案例。最后，为了数数，我在概念上想要的一切都结束了。除了等级和密集等级之外，我不知道什么都适用

额外阅读

---

我的蜥蜴脑把我带到这里。。。秩和

Select *
       ,NewCol = sum(sign(TheThing)) over (Order by Date)
       ,OrEven = sum(TheThing/TheThing) over (Order by Date)  
 From  MyData

返回

我将使用子查询：

试试这个：

declare @tbl table (dt datetime, col int);
insert into @tbl values
('2016-03-09 08:17:29.867', 1),
('2016-03-09 08:18:33.327', 1),
('2016-03-09 14:32:01.240', NULL),
('2016-10-21 19:53:49.983', NULL),
('2016-11-12 03:25:21.753', 1),
('2016-11-24 07:43:24.483', NULL),
('2016-11-28 16:06:23.090', 1),
('2016-11-28 16:09:07.200', 1),
('2016-12-10 11:21:55.807', 1);

select dt,
       col,
       sum(case when col is null then 0 else 1 end) over (order by dt) rnk
from @tbl

其思想非常简单：如果将1指定给非空值，并在列为空时将其指定为零，则按日期排序的累积和将与不包括空值的秩完全相同

另一种方法是将秩与行号结合使用，这将尊重日期列中的关系，并与秩与空值的作用完全相同：

---

我首先使用CTE获得正确的日期，然后将排名应用于修改的日期：

CREATE TABLE #tmp(dt datetime, TheThing int)

INSERT INTO #tmp VALUES('2016-03-09 08:17:29.867',  1)
INSERT INTO #tmp VALUES('2016-03-09 08:18:33.327',  1)
INSERT INTO #tmp VALUES('2016-03-09 14:32:01.240',  NULL)
INSERT INTO #tmp VALUES('2016-10-21 19:53:49.983',  NULL)
INSERT INTO #tmp VALUES('2016-11-12 03:25:21.753',  1)
INSERT INTO #tmp VALUES('2016-11-24 07:43:24.483',  NULL)
INSERT INTO #tmp VALUES('2016-11-28 16:06:23.090',  1)
INSERT INTO #tmp VALUES('2016-11-28 16:09:07.200',  1)
INSERT INTO #tmp VALUES('2016-12-10 11:21:55.807',  1)


;WITH CTE as (
SELECT 
CASE WHEN TheThing IS NULL THEN (SELECT MAX(dt) from #tmp OrigTbl where OrigTbl.dt <     SubTbl.dt and OrigTbl.TheThing IS NOT NULL) ELSE dt end dtMod,
SubTbl.dt,SubTbl.TheThing
   from #tmp SubTbl)
SELECT dt, TheThing, DENSE_RANK() over(ORDER BY dtMod) from CTE

---

我认为您正在寻找一个累积计数：

SELECT Date, TheThing, 
       COUNT(theThing) OVER (ORDER BY Date) AS Total
FROM MyData
ORDER BY Date;

---

从秩中减去当前空值的计数怎么样

---

这也应该可以保留重复项和间隙，并且不需要子查询。

我不认为您可以在单个查询中实现这一点。首先，您需要筛选记录中的not Nulls，然后选择前面创建的列组中的all。该项是否有Frob或null以外的内容？这会影响跑步总量吗？请显示所有的边缘事例。当该项为空时，是否按事例排序，然后滞后。。。否则，使用前一行的值将使您到达任何位置？如果出现初始空值，它可能需要一个模糊因子来处理。@HABO或一行中的两个空值，因为LAG必须知道的不是常数，而是最后一个非空值返回的行数。这会起作用，但它对数据的大小非常敏感。。。对较大数据集上的指数级运行时要小心。@AaronBertrand。我知道这不是生成序列的有效方法。这不会像RANK那样处理重复项，尽管它们可能在RANK的错误轨道上。但是，如果存在多个与最低dt相关的值，则会看到差异。等级将分配所有1。这是行不通的。实际上，它会正确地分配秩，但不是1，在两个绑定行的情况下会有2，但对于绑定行也是一样的。@MartinSmith第二个查询正确地处理了关系，但我认为从提供的数据来看，第一个查询可以安全地使用。当引入关系时，它的工作方式与秩或密集秩不同。这和我回答中的第一个问题几乎是一样的：这也指向了我：虽然所有其他答案都是好的，并且对其他试图解决另一个问题的人都有效，但我想要解决的实际问题确实在计算中。我认为排名是计数的方法。事实证明，倒计时是一件新鲜事。

declare @tbl table (dt datetime, col int);
insert into @tbl values
('2016-03-09 08:17:29.867', 1),
('2016-03-09 08:18:33.327', 1),
('2016-03-09 14:32:01.240', NULL),
('2016-10-21 19:53:49.983', NULL),
('2016-11-12 03:25:21.753', 1),
('2016-11-24 07:43:24.483', NULL),
('2016-11-28 16:06:23.090', 1),
('2016-11-28 16:09:07.200', 1),
('2016-12-10 11:21:55.807', 1);

select dt,
       col,
       sum(case when col is null then 0 else 1 end) over (order by dt) rnk
from @tbl

select dt,
       col,
       case when col is not null then 
           rank() over (order by dt)
       else 
           rank() over (order by dt) - row_number() over (partition by rnDiff order by dt)
       end rnk
from (
    select dt,
           col,
           row_number() over (order by dt) -
               row_number() over (partition by coalesce(col, 0) order by dt) rnDiff
    from @tbl
) a
order by dt

CREATE TABLE #tmp(dt datetime, TheThing int)

INSERT INTO #tmp VALUES('2016-03-09 08:17:29.867',  1)
INSERT INTO #tmp VALUES('2016-03-09 08:18:33.327',  1)
INSERT INTO #tmp VALUES('2016-03-09 14:32:01.240',  NULL)
INSERT INTO #tmp VALUES('2016-10-21 19:53:49.983',  NULL)
INSERT INTO #tmp VALUES('2016-11-12 03:25:21.753',  1)
INSERT INTO #tmp VALUES('2016-11-24 07:43:24.483',  NULL)
INSERT INTO #tmp VALUES('2016-11-28 16:06:23.090',  1)
INSERT INTO #tmp VALUES('2016-11-28 16:09:07.200',  1)
INSERT INTO #tmp VALUES('2016-12-10 11:21:55.807',  1)


;WITH CTE as (
SELECT 
CASE WHEN TheThing IS NULL THEN (SELECT MAX(dt) from #tmp OrigTbl where OrigTbl.dt <     SubTbl.dt and OrigTbl.TheThing IS NOT NULL) ELSE dt end dtMod,
SubTbl.dt,SubTbl.TheThing
   from #tmp SubTbl)
SELECT dt, TheThing, DENSE_RANK() over(ORDER BY dtMod) from CTE

SELECT Date, TheThing, 
       COUNT(theThing) OVER (ORDER BY Date) AS Total
FROM MyData
ORDER BY Date;

SELECT date,
       thething,
       rank() OVER (ORDER BY date)
       -
       sum(CASE
             WHEN thething IS NULL THEN
               1
             ELSE
               0
           END) OVER (ORDER BY date) desiredtotal
       FROM mydata;