Question

我有以下数据集：

Ranking Segment Month
1   1   201501
2   1   201501
3   1   201501
4   1   201501
5   1   201501
6   1   201501
7   1   201501
…   1   201501
567 1   201501
1   2   201501
2   2   201501
3   2   201501
4   2   201501
….. 2   201501
456 2   201501
1   1   201502
2   1   201502
3   1   201502
4   1   201502
5   1   201502
6   1   201502
7   1   201502
…   1   201502
326 1   201502
1   2   201502
2   2   201502
3   2   201502
4   2   201502
…   2   201502
562 2   201502
...........

我需要将每个细分受众群分成5％的细分。由于每个细分市场每个月的销售数量不同，您能否告诉我如何将每个细分市场分成20个组，其中5％的排序按排名分类？

谢谢！

Answer 1

一种方法是使用count(*)作为窗口函数。目前还不清楚每组的样本类型。

以下是第一个＆＃34; n＆＃34;样本，假设ranking没有间隙或重复：

select t.*,
       floor( 20 * (ranking - 1) / cnt) as grp
from (select t.*, count(*) over (partition by segment, month) as cnt
      from t
     ) t;

如果存在空白或重复，您可以使用row_number()来获取正确的枚举。

Answer 2

根据您的输入数据，您不会说出您的预期输出。也许ntile()就是你所追求的，例如：

WITH sample_data AS (SELECT LEVEL ranking, 1 SEGMENT, 201501 mnth FROM dual CONNECT BY LEVEL <= 21 UNION ALL
                     SELECT LEVEL ranking, 2 SEGMENT, 201501 mnth FROM dual CONNECT BY LEVEL <= 40 UNION ALL
                     SELECT LEVEL ranking, 1 SEGMENT, 201502 mnth FROM dual CONNECT BY LEVEL <= 60 UNION ALL
                     SELECT LEVEL ranking, 2 SEGMENT, 201502 mnth FROM dual CONNECT BY LEVEL <= 80)
SELECT ranking,
       segment,
       mnth,
       NTILE(20) OVER (PARTITION BY mnth, SEGMENT ORDER BY ranking) grp
FROM   sample_data;

   RANKING    SEGMENT       MNTH        GRP
---------- ---------- ---------- ----------
         1          1     201501          1
         2          1     201501          1
         3          1     201501          2
         4          1     201501          3
        ...
        19          1     201501         18
        20          1     201501         19
        21          1     201501         20
         1          2     201501          1
         2          2     201501          1
         3          2     201501          2
         4          2     201501          2
         5          2     201501          3
        ...
        36          2     201501         18
        37          2     201501         19
        38          2     201501         19
        39          2     201501         20
        40          2     201501         20
         1          1     201502          1
         2          1     201502          1
         3          1     201502          1
         4          1     201502          2
         5          1     201502          2
         6          1     201502          2
         7          1     201502          3
        ...
        54          1     201502         18
        55          1     201502         19
        56          1     201502         19
        57          1     201502         19
        58          1     201502         20
        59          1     201502         20
        60          1     201502         20
         1          2     201502          1
         2          2     201502          1
         3          2     201502          1
         4          2     201502          1
         5          2     201502          2
         6          2     201502          2
         7          2     201502          2
         8          2     201502          2
         9          2     201502          3
        ...
        72          2     201502         18
        73          2     201502         19
        74          2     201502         19
        75          2     201502         19
        76          2     201502         19
        77          2     201502         20
        78          2     201502         20
        79          2     201502         20
        80          2     201502         20

N.B。 ntile将首先填充早期的桶，例如对于上面我的例子中的段1和月201501，grp = 1有2行，而所有其他行都有1.如果你想要它以便后面的桶被填满更多，你可以简单地反转ntile的顺序，然后从桶的数量减去+ 1，所以在你的情况下，那将是：

21 - NTILE(20) OVER (PARTITION BY mnth, SEGMENT ORDER BY ranking DESC)

Answer 3

您正在寻找NTILE(20)，它将数据分成5％的块。

select ranking, segment, month
from
(
  select
    ranking, segment, month,
    ntile(20) over (partition by segment, month order by ranking) as tile
  from mytable
)
where tile = 1
order by month, segment, ranking;

在PL SQL（Oracle）中按％的分区进行分区

3 个答案: