有条件的 ROW_NUMBER() 跳过仍然计算它们的空值答案

【问题标题】：Conditional ROW_NUMBER() to skip nulls still counting them有条件的 ROW_NUMBER() 跳过仍然计算它们的空值
【发布时间】：2018-11-19 15:33:47
【问题描述】：

我正在尝试将 ROW_NUMBER 列添加到遵循 [Withdrawal then Fee; 顺序的数据集； Withdrawal then Fee] 基于NETWORKCD 列是否为空。但是，我的 row_number 公式仍在计算 Null，即使它没有显示数字，如下所示：

+--------+-------------+------------+-----------+-------------+
| Acctid | Transaction |  PostDate  | NetworkCd | PeriodCount |
+--------+-------------+------------+-----------+-------------+
|  12345 | Withdrawal  | 10/4/2018  | FRGN      |           1 |
|  12345 | Fee         | 10/4/2018  |           |             |
|  12345 | Withdrawal  | 10/11/2018 | FRGN      |           3 |
|  12345 | Fee         | 10/11/2018 |           |             |
|  12345 | Withdrawal  | 10/22/2018 | FRGN      |           5 |
|  12345 | Fee         | 10/22/2018 |           |             |
+--------+-------------+------------+-----------+-------------+

我对@987654323@使用以下公式

(case when networkcd is not null 
      then row_number() over (partition by acctid order by postdate) 
 end) PeriodCount

我希望 PeriodCount 计数 [1,2,3] 而不是 [1,3,5]，如下所示：

+--------+-------------+------------+-----------+-------------+
| Acctid | Transaction |  PostDate  | NetworkCd | PeriodCount |
+--------+-------------+------------+-----------+-------------+
|  12345 | Withdrawal  | 10/4/2018  | FRGN      |           1 |
|  12345 | Fee         | 10/4/2018  |           |             |
|  12345 | Withdrawal  | 10/11/2018 | FRGN      |           2 |
|  12345 | Fee         | 10/11/2018 |           |             |
|  12345 | Withdrawal  | 10/22/2018 | FRGN      |           3 |
|  12345 | Fee         | 10/22/2018 |           |             |
+--------+-------------+------------+-----------+-------------+

我错过了什么？

【问题讨论】：

row_number() 正是在这种情况下，行号，无论您选择显示它还是在 case 语句中用 null 替换它，都不会改变它是哪个行号。
@Andrew 这是我的第一个想法，但 networkcd 的 case 语句不是为 null 调整的吗？
别担心，Alex 发布的 Dense_Rank 是前进的方向，其工作方式的细微差别

标签： sql plsql oracle11g

【解决方案1】：

它仍然在第 1、第 3 和第 5 行显示一个值 - 因此遇到这些行中的每一行时的行数都是正确的。它是整个结果集中的行号，而不是非空值中的行号。您的 case 表达式决定是否显示该值，而不影响如何找到它。

您可以改用dense_rank() 获得您想要的结果；

with your_table (Acctid, Transaction, PostDate, NetworkCd) as (
          select 12345, 'Withdrawal', to_date('10/4/2018', 'MM/DD/YYYY'), 'FRGN' from dual
union all select 12345, 'Fee', to_date('10/4/2018', 'MM/DD/YYYY'), null from dual
union all select 12345, 'Withdrawal', to_date('10/11/2018', 'MM/DD/YYYY'), 'FRGN' from dual
union all select 12345, 'Fee', to_date('10/11/2018', 'MM/DD/YYYY'), null from dual
union all select 12345, 'Withdrawal', to_date('10/22/2018', 'MM/DD/YYYY'), 'FRGN' from dual
union all select 12345, 'Fee', to_date('10/22/2018', 'MM/DD/YYYY'), null from dual
)
select Acctid, Transaction, PostDate, NetworkCd,
(case when networkcd is not null then dense_rank() over (partition by acctid order by postdate) end) as periodcount
from your_table;

    ACCTID TRANSACTIO POSTDATE   NETW PERIODCOUNT
---------- ---------- ---------- ---- -----------
     12345 Withdrawal 2018-10-04 FRGN           1
     12345 Fee        2018-10-04                 
     12345 Withdrawal 2018-10-11 FRGN           2
     12345 Fee        2018-10-11                 
     12345 Withdrawal 2018-10-22 FRGN           3
     12345 Fee        2018-10-22

... 因为这确实抑制了生成值中的差距。 From the docs:

DENSE_RANK 计算行在有序行组中的排名，并将排名作为NUMBER 返回。秩是从 1 开始的连续整数。 ...

【讨论】：

完美运行，谢谢。我会花一些时间研究 Dense_rank()。

【解决方案2】：

尝试将partition by acctid 更改为partition by acctid, NetworkCd IS NOT NULL

【讨论】：

【解决方案3】：

Alex Poole 的答案有效，但取决于您的空行与您要计算的非空行具有相同的 POSTDATE 的事实。如果您将日期更改为彼此不同，您会看到 dense_rank 失败：

with your_table (Acctid, Transaction, PostDate, NetworkCd) as (
              select 12345, 'Withdrawal', to_date('10/1/2018', 'MM/DD/YYYY'), 'FRGN'
    union all select 12345, 'Fee', to_date('10/2/2018', 'MM/DD/YYYY'), null
    union all select 12345, 'Withdrawal', to_date('10/3/2018', 'MM/DD/YYYY'), 'FRGN'
    union all select 12345, 'Fee', to_date('10/4/2018', 'MM/DD/YYYY'), null
    union all select 12345, 'Withdrawal', to_date('10/5/2018', 'MM/DD/YYYY'), 'FRGN'
    union all select 12345, 'Fee', to_date('10/6/2018', 'MM/DD/YYYY'), null
    )

select
    Acctid,
    Transaction,
    PostDate,
    NetworkCd,
    case when networkcd is not null then dense_rank() over (partition by Acctid order by PostDate) end as PeriodCount,
    dense_rank() over (partition by Acctid order by PostDate) as DenseRank
from your_table
order by Acctid, PostDate, Transaction
;

产量

acctid  transaction  postdate    networkcd  periodcount  denserank
------  -----------  ----------  ---------  -----------  ---------
12345   Withdrawal   2018-10-01  FRGN       1            1
12345   Fee          2018-10-02                          2
12345   Withdrawal   2018-10-03  FRGN       3            3
12345   Fee          2018-10-04                          4
12345   Withdrawal   2018-10-05  FRGN       5            5
12345   Fee          2018-10-06                          6

（在右侧，显示不带 null 条件的 dense_rank 的结果）因此它适用于您的情况，但总体而言并不可靠。

Kevin Siemons 的回答效果更好：它分别计算空行和非空行，您可以简单地隐藏对空行进行的计数：

with your_table (Acctid, Transaction, PostDate, NetworkCd) as (
              select 12345, 'Withdrawal', to_date('10/1/2018', 'MM/DD/YYYY'), 'FRGN'
    union all select 12345, 'Fee', to_date('10/2/2018', 'MM/DD/YYYY'), null
    union all select 12345, 'Withdrawal', to_date('10/3/2018', 'MM/DD/YYYY'), 'FRGN'
    union all select 12345, 'Fee', to_date('10/4/2018', 'MM/DD/YYYY'), null
    union all select 12345, 'Withdrawal', to_date('10/5/2018', 'MM/DD/YYYY'), 'FRGN'
    union all select 12345, 'Fee', to_date('10/6/2018', 'MM/DD/YYYY'), null
    )

select
    Acctid,
    Transaction,
    PostDate,
    NetworkCd,
    case when networkcd is not null then row_number() over (partition by Acctid, NetworkCd is not null order by PostDate) end as PeriodCount,
    row_number() over (partition by Acctid, NetworkCd is not null order by PostDate) as RowNumber
from your_table
order by Acctid, PostDate, Transaction
;

产量

acctid  transaction  postdate    networkcd  periodcount  rownumber
------  -----------  ----------  ---------  -----------  ---------
12345   Withdrawal   2018-10-01  FRGN       1            1
12345   Fee          2018-10-02                          1
12345   Withdrawal   2018-10-03  FRGN       2            2
12345   Fee          2018-10-04                          2
12345   Withdrawal   2018-10-05  FRGN       3            3
12345   Fee          2018-10-06                          3

我建议使用 case when networkcd is not null then row_number() over (partition by Acctid, NetworkCd is not null order by PostDate) end as PeriodCount。

【讨论】：

【解决方案4】：

ROW_NUMBER() 函数无法按预期工作，但您可以改为：

select t.*,
       (select count(*)
        from table t1
        where t1.acctid = t.acctid and t1.PostDate <= t.PostDate and
              t1.networkcd is not null 
       ) as PeriodCount
from table t;

【讨论】：

【解决方案5】：

看起来它不是 NULL，而是一个空字符串。 SQL 将在代码中以不同的方式看待这一点。

试试：

case when networkcd = '' then row_number()....

【讨论】：

Oracle Database treats a character value with a length of zero as null.。将任何内容与= '' 进行比较的结果与= null 相同，即“未知”。