Question

数据间隔为15分钟：

Time               Value
2010-01-01 00:15   3
2010-01-01 00:30   2
2010-01-01 00:45   4
2010-01-01 01:00   5
2010-01-01 01:15   1
2010-01-01 01:30   3
2010-01-01 01:45   4
2010-01-01 02:00   12
2010-01-01 02:15   13
2010-01-01 02:30   12
2010-01-01 02:45   14
2010-01-01 03:00   15
2010-01-01 03:15   3
2010-01-01 03:30   2
2010-01-01 03:45   3
2010-01-01 04:00   5
..........
..........
..........
2010-01-02 00:00

通常会有96分。

根据这些值，我们可能会注意到00:15到01:45之间的值彼此接近，从02:00到03:00它们彼此接近，从03:15到04:00他们彼此接近。

基于“彼此接近”规则，我希望将数据“分组”为3个部分：

00:15至01:45
02:00至03:00
03:15至04:00

请注意，数据可以是随机的，并且可以根据上面定义的规则分组为3个以上，但最大值不应超过10个。分组必须遵守时间顺序，例如，您不能将00：15/02：30/04：45放入1组，因为这3个点不是连续的。

请介绍一下如何在t-sql中实现它。

更新价值可能是：

Time               Value
2010-01-01 00:15   3
2010-01-01 00:30   2
2010-01-01 00:45   4
2010-01-01 01:00   5
2010-01-01 01:15   1
2010-01-01 01:30   3
2010-01-01 01:45   4
2010-01-01 02:00   12
2010-01-01 02:15   13
2010-01-01 02:30   4  --suddenly decreased
2010-01-01 02:45   14
2010-01-01 03:00   15
2010-01-01 03:15   3
2010-01-01 03:30   2
2010-01-01 03:45   3
2010-01-01 04:00   5
..........
..........
..........
2010-01-02 00:00

对于这种情况，我们不应该单独分组02:30，因为我们希望组大小必须至少为3分，我们将把这一点（02:30）放到上一组（从02开始）：00至03:00）。

Answer 1

声明并填充testdata：

set nocount on
declare @result table(mintime datetime, maxtime datetime)
declare @t table(time datetime, value int)

-- variation is how much difference will be allowed from one row to the next
declare @variation int
set @variation = 5     

insert @t values('2010-01-01 00:15',3)
insert @t values('2010-01-01 00:30',2)
insert @t values('2010-01-01 00:45',4)
insert @t values('2010-01-01 01:00',5)
insert @t values('2010-01-01 01:15',1)
insert @t values('2010-01-01 01:30',3)
insert @t values('2010-01-01 01:45',4)
insert @t values('2010-01-01 02:00',12)
insert @t values('2010-01-01 02:15',13)
insert @t values('2010-01-01 02:30',12)
insert @t values('2010-01-01 02:45',14)
insert @t values('2010-01-01 03:00',15)
insert @t values('2010-01-01 03:15',3)
insert @t values('2010-01-01 03:30',2)
insert @t values('2010-01-01 03:45',3)
insert @t values('2010-01-01 04:00',5)

代码：

a:

;with t as
( -- add a rownumber
select *, rn = row_number() over(order by time) from @t
), a as
(-- increase group if current row's value varies more than @variation from last row's value
select time, value, rn, 0 grp from t where rn = 1
union all
select t.time, t.value, t.rn, case when t.value between 
       a.value - @variation and a.value +@variation 
       then grp else grp+1 end 
from t join a on 
t.rn = a.rn +1
)
insert @result
select min(time), max(time) from a group by grp


if @@rowcount > 10 
begin 
    -- this will activate if more than 10 groups of numbers are found
    -- start over with higher tolerance for variation
    set @variation=@variation + 1 
    delete @result
    goto a 
end

select convert(char(5), mintime,114) + ' to ' + convert(char(5), maxtime,114)
from @result

结果在这里： http://data.stackexchange.com/stackoverflow/q/110891/declare-and-populate-testdata

Answer 2

由于您的问题发生了很大变化，这里是新问题的新答案，我只包含了代码部分。

declare @t table(time datetime, value int)
declare @variation float
set @variation = 2
set nocount on

insert @t values('2010-01-01 00:15',3)
insert @t values('2010-01-01 00:30',2)
insert @t values('2010-01-01 00:45',4)
insert @t values('2010-01-01 01:00',5)
insert @t values('2010-01-01 01:15',1)
insert @t values('2010-01-01 01:30',3)
insert @t values('2010-01-01 01:45',4)
insert @t values('2010-01-01 02:00',52)
insert @t values('2010-01-01 02:15',5)
insert @t values('2010-01-01 02:30',52)
insert @t values('2010-01-01 02:45',54)
insert @t values('2010-01-01 03:00',55)
insert @t values('2010-01-01 03:15',3)
insert @t values('2010-01-01 03:30',2)
insert @t values('2010-01-01 03:45',3)
insert @t values('2010-01-01 04:00',5)


declare @result table(mintime datetime, maxtime datetime)
a:
delete @result

;with t as
(
select *, rn = row_number() over(order by time), log(value) lv from @t where datediff(day, time, '2010-01-01') = 0
), a as
(
select time, lv, rn, 0 grp from t where rn = 1
union all
select t1.time, a.lv, t1.rn, 
case when exists (select 1 from t t2 where t1.rn between rn + 1 and rn + 3 and 
lv between t1.lv - @variation and t1.lv +@variation) then grp else grp + 1 end
from t t1 join a on 
t1.rn = a.rn +1
)
insert @result
select min(time), max(time) from a group by grp

if @@rowcount > 10 
begin 
    set @variation=@variation + .5 
    goto a 
end

select * from @result

结果：

mintime                     maxtime
2010-01-01 00:15:00.000     2010-01-01 01:45:00.000
2010-01-01 02:00:00.000     2010-01-01 03:00:00.000
2010-01-01 03:15:00.000     2010-01-01 04:00:00.000

根据sql server中的相关值进行数据分组

2 个答案: