GROUP BY 整数值区间

Question

我需要总结线性对象的长度，并以几年为间隔对它们进行分组。我有一张桌子，像这样存储我的对象：

- gid serial NOT NULL, 
- year INTEGER, 
- the_geom geometry(MULTILINESTRING) ;

我需要这样的结果：

period          | length
----------------+-----------
 2005 - 2014    | 18.6
 1995 - 2004    | 16.1
 1985 - 1994    | 7.6
 1975 - 1984    | 19.0
 1965 - 1974    | 28.2
 1945 - 1964    | 10.2
 before 1945    | 0.1 

我无法在网上找到如何执行此操作，除非使用不同的查询生成每一行并使用 UNION ALL 将它们合并在一起，这不太好......

Answer 1

您需要一个判别函数，它可以通过多种方式创建。出于您的目的，案例陈述只是一张票，例如：

select case
         when t.year >= 2015 then '2015-present'
         when t.year >= 2005 then '2005-2014'
         when t.year >= 1995 then '1995-2004'
         when t.year >= 1985 then '1985-1994'
         when t.year >= 1975 then '1975-1984'
         when t.year >= 1965 then '1965-1974'
         when t.year >= 1955 then '1955-1964'
         when t.year >= 1945 then '1945-1954'
         when t.year <  1945 then 'before 1945'
         else                     'no year given'
       end as period ,
       sum( compute_length_from_geometry( t.geometry) ) as length
from some_table t
where .
      .
      .
group by case
           when t.year >= 2015 then '2015-present'
           when t.year >= 2005 then '2005-2014'
           when t.year >= 1995 then '1995-2004'
           when t.year >= 1985 then '1985-1994'
           when t.year >= 1975 then '1975-1984'
           when t.year >= 1965 then '1965-1974'
           when t.year >= 1955 then '1955-1964'
           when t.year >= 1945 then '1945-1954'
           when t.year <  1945 then 'before 1945'
           else                     'no year given'
         end as period
order by case
           when t.year >= 2015 then  1
           when t.year >= 2005 then  2
           when t.year >= 1995 then  3
           when t.year >= 1985 then  4
           when t.year >= 1975 then  5
           when t.year >= 1965 then  6
           when t.year >= 1955 then  7
           when t.year >= 1945 then  8
           when t.year <  1945 then  9
           else                     10
         end as period

您也可以只考虑一个包围表，无论是永久的还是临时的，例如：

create table report_period
(
  period_id          int         not null ,
  year_from          int         not null ,
  year_thru          int         not null ,
  period_description varchar(32) not null ,

  primary key clustered ( period_id ) ,
  unique nonclustered ( year_from , year_thru ) ,

)
insert report_period values ( 1 , 2015 , 9999 , '2015-present' )
insert report_period values ( 2 , 2005 , 2014 , '2005-2014'    )
insert report_period values ( 3 , 1995 , 2004 , '1995-2004'    )
insert report_period values ( 4 , 1985 , 1994 , '1985-1994'    )
insert report_period values ( 5 , 1975 , 1984 , '1975-1984'    )
insert report_period values ( 6 , 1965 , 1974 , '1965-1974'    )
insert report_period values ( 7 , 1955 , 1964 , '1955-1964'    )
insert report_period values ( 8 , 1945 , 1954 , '1945-1954'    )
insert report_period values ( 9 , 0000 , 1944 , 'pre-1945'     )

然后你的查询就变成了

select p.period_description as period ,
       sum( compute_length_from_geometry( t.geometry ) ) as length
from report_period p
join some_table    t on t.year between p.year_from and p.year_thru
group by p.period_id ,
         p.period_description
order by p.period_id

你甚至可以使用派生表来获得同样的效果

select p.period_description as period ,
       sum( compute_length_from_geometry( t.geometry ) ) as length
from (           select 1 as period_id , 2015 as year_from , 9999 as year_thru , '2015-present' as period_description
       UNION ALL select 2 as period_id , 2005 as year_from , 2014 as year_thru , '2005-2014' as period_description
       UNION ALL select 3 as period_id , 1995 as year_from , 2004 as year_thru , '1995-2004' as period_description
       ...
     ) p
join some_table    t on t.year between p.year_from and p.year_thru
group by p.period_id ,
         p.period_description
order by p.period_id

或者，您也可以简单地进行整数除法，例如

period_id = ( 2014 - t.year ) / 10

这将为您提供域的期间标识符

• > 0：2015 年或以后
• 0: 2005-2014
• -1: 1995-2004
• -2: 1985-1994
• -3: 1975-1984
• -4: 1965-1974
• -5: 1955-1964
• -6: 1945-1954

然后只需添加/减去一个适当的偏移量来移动零点（或以年为单位更改计算偏移量）。

然而，这通常会否定使用列year 上的任何索引，因为它现在是一个表达式。

Answer 2

假设以下得到长度：

select year, length(the_geo) as len
from table

那么你的问题是找到具有相同长度的连续年份。我喜欢以下查找此类序列的技巧，假设您每年都有一个值：

select min(year) || '-' || max(year), len
from  (select year, length(the_geo) as len,
              (row_number() over (order by year) -
               row_number() over (partition by length(the_geo) order by year)
              ) as grp
       from table
      ) t
group by grp, len
order by 1;

Answer 3

按整数除法的结果分组，有效地截断为除数的倍数 - 在您的情况下为 10。 Shift 5 来​​实现你的分区：

SELECT min(year)::text || ' - ' || max(year) AS period
     , sum(length(the_geom)) AS "length"
FROM   tbl
GROUP BY (year + 5) / 10
ORDER  BY min(year) DESC;

Per documentation:

/除法（整数除法截断结果）

Answer 4

括号表report_period的解决方案非常好，对我来说是最简单的。非常感谢尼古拉斯（以及所有）！