使用MySQL聚合重叠事件

时间:2012-04-27 02:57:55

标签: mysql sql events aggregate overlapping

所以我们假设我们的数据看起来很像:

drop table if exists views; 
create table views(id int primary key,start time,end time); 
insert into views values 
(1, '15:01', '15:04'), 
(2, '15:02', '15:09'), 
(3, '15:12', '15:15'), 
(4, '16:11', '16:23'), 
(5, '16:19', '16:25'), 
(6, '17:52', '17:59'), 
(7, '18:18', '18:22'), 
(8, '16:20', '16:22'), 
(9, '18:17', '18:23'); 

像这样轻松可视化

1     |-----| 
2        |-----| 
3                 |--| 
4                       |-----| 
5                          |-----| 
6                                  |---| 
7                                        |---|  
8                           |---| 
9                                       |-----| 

现在我想绘制数据,使其看起来像这样

+---------------------------+
|              x            |
|    x        x xxx     xxx |
|   x xx  xx x     xx  x    |
+---------------------------+

基本上将它们分成X长度的段并总结每个X长度段被触摸的次数。有关如何创建此视图的任何想法?

(如果您必须知道这一点,那么我可以创建视频分析的参与度数据)

我不希望输出为ASCII我希望它最终作为SQL中的查询结果。类似的东西:

Time Start, Time End,  Num_Views
00:00, 00:05, 10
00:06, 00:10, 3
00:11, 00:15, 2
00:16, 00:20, 8

1 个答案:

答案 0 :(得分:3)

使用辅助数字表,您可以执行以下操作:

select
  r.Time_Start,
  r.Time_End,
  sum(v.id is not null) as Num_Views
from (
  select
    cast(from_unixtime((m.minstart + n.n + 0) * 300) as time) as Time_Start,
    cast(from_unixtime((m.minstart + n.n + 1) * 300) as time) as Time_End
  from (
    select
      unix_timestamp(date_format(minstart, '1970-01-01 %T')) div 300 as minstart,
      unix_timestamp(date_format(maxend  , '1970-01-01 %T')) div 300 as maxend
    from (
      select
        min(start) as minstart,
        max(end  ) as maxend
      from views
    ) s
  ) m
    cross join numbers n
  where n.n between 0 and m.maxend - minstart
) r
  left join views v on v.start < r.Time_End and v.end > r.Time_Start
group by
  r.Time_Start,
  r.Time_End
;

对于您的特定示例,此脚本会生成以下输出:

Time_Start  Time_End  Num_Views
----------  --------  ---------
15:00:00    15:05:00  2
15:05:00    15:10:00  1
15:10:00    15:15:00  1
15:15:00    15:20:00  0
15:20:00    15:25:00  0
15:25:00    15:30:00  0
15:30:00    15:35:00  0
15:35:00    15:40:00  0
15:40:00    15:45:00  0
15:45:00    15:50:00  0
15:50:00    15:55:00  0
15:55:00    16:00:00  0
16:00:00    16:05:00  0
16:05:00    16:10:00  0
16:10:00    16:15:00  1
16:15:00    16:20:00  2
16:20:00    16:25:00  3
16:25:00    16:30:00  0
16:30:00    16:35:00  0
16:35:00    16:40:00  0
16:40:00    16:45:00  0
16:45:00    16:50:00  0
16:50:00    16:55:00  0
16:55:00    17:00:00  0
17:00:00    17:05:00  0
17:05:00    17:10:00  0
17:10:00    17:15:00  0
17:15:00    17:20:00  0
17:20:00    17:25:00  0
17:25:00    17:30:00  0
17:30:00    17:35:00  0
17:35:00    17:40:00  0
17:40:00    17:45:00  0
17:45:00    17:50:00  0
17:50:00    17:55:00  1
17:55:00    18:00:00  1
18:00:00    18:05:00  0
18:05:00    18:10:00  0
18:10:00    18:15:00  0
18:15:00    18:20:00  2
18:20:00    18:25:00  2

数字表可能是临时表,但我建议您创建并初始化永久表,因为它可用于许多目的。这是初始化数字表的一种方法:

create table numbers (n int);
insert into numbers (n) select 0;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
insert into numbers (n) select cnt + n from numbers, (select count(*) as cnt from numbers) s;
/* repeat as necessary; every repeated line doubles the number of rows */

可以找到此脚本的“实时”版本on SQL Fiddle

更新(尝试描述所用方法)

上述解决方案实现了以下步骤:

  1. start表中查找最早end时间和最新views时间。

  2. 将两个值都转换为unix timestamps

  3. 将两个时间戳除以300,这基本上为我们提供了相应的5分钟范围(自纪元以来)的索引。

  4. 借助数字表,生成一系列5分钟范围,涵盖startend之间的整体范围。

  5. 将范围列表与views表中的事件时间匹配(使用外部联接,因为我们想要( if 我们想要的)来计算所有范围)

  6. 按范围界限对结果进行分组,并计算组中事件的数量。 (我刚刚注意到上述查询中的sum(v.id is not null)可以用更简洁的方式替换,在这种情况下,更自然count(v.id)。)