Oracle
 sql >> Base de données >  >> RDS >> Oracle

Requête SQL pour compter les changements de valeurs dans une colonne

Depuis Oracle 12, vous pouvez utiliser MATCH_RECOGNIZE :

SELECT cat,
       month,
       COUNT(*)
FROM   (
  SELECT t.*,
         TRUNC( "DATE", 'MM' ) AS month
  FROM   table_name t
)
MATCH_RECOGNIZE(
  PARTITION BY cat, month
  ORDER BY "DATE", version
  ONE ROW PER MATCH
  AFTER MATCH SKIP TO LAST change_code
  PATTERN ( strt change_code )
  DEFINE
    change_code AS change_code.some_code <> strt.some_code
)
GROUP BY cat, month

Qui, pour les exemples de données :

CREATE TABLE table_name ( CAT, NR, "DATE", VERSION, SOME_CODE ) AS
SELECT 'ABC',   123,    TIMESTAMP '2009-02-19 00:00:00 UTC',    1,  'OPP' FROM DUAL UNION ALL
SELECT 'ABC',   456,    TIMESTAMP '2009-03-18 00:00:00 UTC',    1,  'ZUM' FROM DUAL UNION ALL
SELECT 'ABC',   444,    TIMESTAMP '2009-03-18 00:00:00 UTC',    1,  'ZUM' FROM DUAL UNION ALL
SELECT 'ABC',   444,    TIMESTAMP '2009-03-18 00:00:00 UTC',    2,  'MUZ' FROM DUAL UNION ALL
SELECT 'ABC',   456,    TIMESTAMP '2009-04-18 00:00:00 UTC',    2,  'XXX' FROM DUAL UNION ALL
SELECT 'ABC',   456,    TIMESTAMP '2009-04-18 00:00:00 UTC',    3,  'XXX' FROM DUAL UNION ALL
SELECT 'ABC',   456,    TIMESTAMP '2009-04-18 00:00:00 UTC',    4,  'UIO' FROM DUAL UNION ALL
SELECT 'ABC',   456,    TIMESTAMP '2009-05-18 00:00:00 UTC',    5,  'RQA' FROM DUAL UNION ALL
SELECT 'DEF',   637,    TIMESTAMP '2018-02-16 00:00:00 UTC',    1,  'FAW' FROM DUAL UNION ALL
SELECT 'DEF',   789,    TIMESTAMP '2018-02-17 00:00:00 UTC',    1,  'WER' FROM DUAL UNION ALL
SELECT 'GHI',   248,    TIMESTAMP '2018-02-17 00:00:00 UTC',    1,  'QWE' FROM DUAL UNION ALL
SELECT 'GHI',   248,    TIMESTAMP '2019-02-17 00:00:00 UTC',    2,  'PPP' FROM DUAL UNION ALL
SELECT 'GHI',   357,    TIMESTAMP '2020-02-16 00:00:00 UTC',    1,  'FFF' FROM DUAL UNION ALL
SELECT 'GHI',   420,    TIMESTAMP '2020-02-16 00:00:00 UTC',    1,  'QDS' FROM DUAL UNION ALL
SELECT 'GHI',   357,    TIMESTAMP '2020-02-16 00:00:00 UTC',    2,  'GGG' FROM DUAL UNION ALL
SELECT 'GHI',   357,    TIMESTAMP '2020-02-16 00:00:00 UTC',    3,  'LLL' FROM DUAL UNION ALL
SELECT 'GHI',   357,    TIMESTAMP '2020-02-16 00:00:00 UTC',    4,  'LLL' FROM DUAL UNION ALL
SELECT 'GHI',   357,    TIMESTAMP '2020-08-16 00:00:00 UTC',    4,  'FFF' FROM DUAL UNION ALL
SELECT 'GHI',   357,    TIMESTAMP '2020-10-16 00:00:00 UTC',    5,  'ZZZ' FROM DUAL

Sorties :

Si vous souhaitez voir les modifications, vous pouvez utiliser :

SELECT *
FROM   (
  SELECT t.*,
         TRUNC( "DATE", 'MM' ) AS month
  FROM   table_name t
)
MATCH_RECOGNIZE(
  PARTITION BY cat, month
  ORDER BY "DATE", version
  MEASURES
    MATCH_NUMBER()     AS mn,
    FIRST( some_code ) AS change_from,
    LAST( some_code )  AS change_to
  ONE ROW PER MATCH
  AFTER MATCH SKIP TO LAST change_code
  PATTERN ( strt change_code )
  DEFINE
    change_code AS change_code.some_code <> strt.some_code
)

Qui sort :

db<>violon ici

Si votre exigence pour "dans un mois" est que vous voulez des changements où il y a au plus un mois de différence entre la ligne précédente et la ligne modifiée, même si les lignes sont dans deux mois calendaires différents (plutôt que seulement les changements qui se produisent dans le même mois calendaire), vous pouvez alors utiliser :

SELECT cat,
       TRUNC( change_date, 'MM' ) AS month,
       COUNT(*)
FROM   table_name
MATCH_RECOGNIZE(
  PARTITION BY cat
  ORDER BY "DATE", version
  MEASURES
    LAST( "DATE" ) AS change_date
  ONE ROW PER MATCH
  AFTER MATCH SKIP TO LAST change_code
  PATTERN ( strt change_code )
  DEFINE
    change_code AS (
      change_code.some_code <> strt.some_code
      AND MONTHS_BETWEEN( change_code."DATE", strt."DATE" ) <= 1
    )
)
GROUP BY cat, TRUNC( change_date, 'MM' )

Qui sort :

db<>violon ici