如何汇总序列中具有相同值的行数？

Question

我有一个查询，返回以下示例中的数据：

SELECT timestamp, atm_id FROM TRANSACTIONS ORDER BY TIMESTAMP ASC;

输出量

TIMESTAMP  | ATM_ID | 
--------------------
2010-01-01 | EP02   |
2010-01-01 | EP02   |
2010-01-28 | EP02   |
2010-02-07 | EP02   |
2010-02-09 | EP11   |
2010-03-19 | EP11   |
2010-03-19 | EP02   |
2010-04-03 | EP05   |
2010-04-30 | EP02   |

我知道如何按ATM_ID分组并将计数放在每个计数的前面

SELECT
   ATM_ID,
   COUNT(*) CNT
FROM
   TRANSACTIONS
GROUP BY
   ATM_ID;

根据上面的示例数据，这将产生类似

 ATM_ID |  CNT  
---------------
 EP02   |  6
 EP11   |  2
 EP05   |  1

但是，我有兴趣在不同级别进行分组。 如果某个ATM_ID在连续的行中重复，则即使输出相同的ATM_ID之后出现在另一个ATM_ID之后，也应在输出中包括顺序具有相同ATM_ID的行数。

期望的输出

 ATM_ID |  CNT  
---------------
 EP02   |  4         --Four rows of ATM_ID EP02
 EP11   |  2         --Followed by 2 rows of ATM_ID EP11
 EP02   |  1         --Followed by 1 row of ATM_ID EP02
 EP05   |  1         --Followed by 1 row of ATM_ID EP05
 EP02   |  1         --Followed by 1 row of ATM_ID EP02

忽略右边的注释，这些只是为了澄清，而不是输出的一部分。 那可能吗？

PS：Syed Aladeen在下面的回答给出的输出具有正确的计数，但顺序错误。 为了方便起见，我创建了一个SQL提琴：

SQL小提琴

Answer 1

尝试这个

select atm_id , count(*) 
from (select TRANSACTIONS.*,
     (row_number() over (order by id) -
      row_number() over (partition by atm_id order by id)
     ) as grp
from TRANSACTIONS
) TRANSACTIONS
group by grp, atm_id order by max(id)

Answer 2

-- Oracle 12c+: pattern matching
with s(dt, atm_id) as (
select to_date('2010-01-01', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-01-01', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-01-28', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-02-07', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-02-09', 'yyyy-mm-dd'), 'EP11' from dual union all
select to_date('2010-03-19', 'yyyy-mm-dd'), 'EP11' from dual union all
select to_date('2010-03-19', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-04-03', 'yyyy-mm-dd'), 'EP05' from dual union all
select to_date('2010-04-30', 'yyyy-mm-dd'), 'EP02' from dual)
select *
from s
match_recognize (
order by dt
measures v.atm_id as atm_id,
count(v.atm_id) as cnt,
first(dt) as min_dt,
last (dt) as max_dt
pattern (v+)
define v as v.atm_id = first(atm_id)
);

ATM_        CNT MIN_DT              MAX_DT
---- ---------- ------------------- -------------------
EP02          4 2010-01-01 00:00:00 2010-02-07 00:00:00
EP11          2 2010-02-09 00:00:00 2010-03-19 00:00:00
EP02          1 2010-03-19 00:00:00 2010-03-19 00:00:00
EP05          1 2010-04-03 00:00:00 2010-04-03 00:00:00
EP02          1 2010-04-30 00:00:00 2010-04-30 00:00:00

Elapsed: 00:00:00.01

-- Oracle 8i+: window sort + window buffer + group by [+ order by]
with s(dt, atm_id) as (
select to_date('2010-01-01', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-01-01', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-01-28', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-02-07', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-02-09', 'yyyy-mm-dd'), 'EP11' from dual union all
select to_date('2010-03-19', 'yyyy-mm-dd'), 'EP11' from dual union all
select to_date('2010-03-19', 'yyyy-mm-dd'), 'EP02' from dual union all
select to_date('2010-04-03', 'yyyy-mm-dd'), 'EP05' from dual union all
select to_date('2010-04-30', 'yyyy-mm-dd'), 'EP02' from dual)
select atm_id, count(*) cnt, min(dt) min_dt, max(dt) as max_dt
from
  (select dt, atm_id, count(lg) over (order by dt) ct, lg
    from
     (select dt, atm_id, decode(atm_id, lag(atm_id) over (order by dt), null, 1) lg
      from s
     )
  )
group by ct, atm_id
order by min_dt;

ATM_        CNT MIN_DT              MAX_DT
---- ---------- ------------------- -------------------
EP02          4 2010-01-01 00:00:00 2010-02-07 00:00:00
EP11          1 2010-02-09 00:00:00 2010-02-09 00:00:00
EP02          1 2010-03-19 00:00:00 2010-03-19 00:00:00
EP11          1 2010-03-19 00:00:00 2010-03-19 00:00:00
EP05          1 2010-04-03 00:00:00 2010-04-03 00:00:00
EP02          1 2010-04-30 00:00:00 2010-04-30 00:00:00

6 rows selected.