Best way to group data and show details of only last result in group

Question

Hi All! First off, just letting everyone know that this is a great place to explore and learn Oracle - I've learned more here than I have in some classes. Just exploring the forums and searching for an answer leads me to functions that I hadn't otherwise known existed.
Here's what I'm looking to accomplish now... let's say I have a table that holds family information - if two or more persons are related in a family (determined by a separate table) then it should return the person identification and then the details of the group.
For instance, the following data is contained in two tables, the current result follow and then the result I'm looking for...
PERSONS TABLE
PERSON         PERSON_ID           ADDRESS
John Smith     101                     1 Oracle Drive
Jane Smith     102                     1 Oracle Drive
RELATIONSHIPS TABLE
PERSON_ID      RELATEDPERSON_ID
101                 102
102                 101A simple query would result in the following:
WITH PERSONS AS
  SELECT 'John Smith' AS person, 101 AS person_id, '101 Oracle Drive' AS address FROM dual union all
  SELECT 'Jane Smith', 102, '101 Oracle Drive' FROM dual
,    RELATIONSHIPS AS
  SELECT 101 AS person_id, 102 AS relatedperson_id FROM dual union all
  SELECT 102, 101 FROM dual
SELECT
    person
  , address
FROM  PERSONS p
JOIN  RELATIONSHIPS r ON r.person_id = p.person_idRESULT
PERSON      ADDRESS
John Smith      101 Oracle Drive
Jane Smith      101 Oracle DriveI'm looking to generate the following result but I'm not sure how to accomplish this... I'm confident it's something simple.
DESIRED RESULT
PERSON      ADDRESS
John Smith     
Jane Smith      101 Oracle DriveNotice that the address for the related family members is not displayed until the last family member is returned. It would repeat this process for each family.
Thanks everyone for any help you can provide! 11g
Edited by: nage62587 on Oct 16, 2012 8:20 PM

Answer 1

Hi,
nage62587 wrote:
I've done a lot of forum searching and revised my question a bit to hopefully make things a bit clearer... I've taken a query Frank wrote and revised to meet my criteria; Searching the forum (and other places on the web) is great! Not only do you find things yourself, but people on this forum are more likely to help you when they see that you're doing all you can.
If you find something that you try to adapt, post a link to it. Seeing the correct way to adapt it can be very instructive.
essentially, if I can determine what relatives a person has, then I can create a unique "FAMILY_ID" for them - once I have that, it would appear as though I could then use the FAMILY_ID to group their addresses and other info together.
The problem I'm having is that if a RELATEDPERSON_ID is linked to a PERSON_ID (PERSON_ID linked to RELATEDPERSON_ID works great), it assigns them a new FAMILY_ID rather than including them into the correct family.
Here's my sql
WITH PERSONS AS
SELECT 'John Smith' AS person, 101 AS person_id, '1 Oracle Drive' AS address FROM dual union all
SELECT 'Jane Smith', 102, '1 Oracle Drive' FROM dual union all
SELECT 'Jack Smith', 103, '8 Oracle Drive' FROM dual union all
SELECT 'John Doe', 104, '10 Oracle Drive' FROM dual union all
SELECT 'Jane Doe', 105, '10 Oracle Drive' FROM dual union all
SELECT 'Pete Smith', 106, '1 Oracle Drive' FROM dual
,    RELATIONSHIPS AS
SELECT 101 AS person_id, 102 AS relatedperson_id FROM dual union all
SELECT 102, 101 FROM dual union all
SELECT 104, 105 FROM dual union all
SELECT 105, 104 FROM dual union all
SELECT 106, 101 FROM dual
, table_x
AS
SELECT   person_id         AS col1
,        relatedperson_id  AS col2
FROM     relationships
,     got_relatives     AS
     SELECT     col1
     ,     CONNECT_BY_ROOT col2     AS relative
     FROM     table_x
     CONNECT BY NOCYCLE     col1     =  col2
OR  col2  =  col1
SELECT       col1
,       DENSE_RANK () OVER ( ORDER BY  MIN (relative)
) AS family_id
FROM      got_relatives
GROUP BY  col1The result I expect is:
COL1   FAMILY_ID
102     1
106     1
101     1
105     2
104     2
I suspect that whatever you copied originally had the PRIOR keyword somewhere in the CONNECT BY clause.
Here's one way to do what you want:
WITH    all_relationships     AS
     SELECT  person_id
     ,     relatedperson_id
     FROM     relationships
    UNION
     SELECT  relatedperson_id     AS person_id
     ,     person_id          AS relatedperson_id
     FROM     relationships
,     got_relatives     AS
     SELECT     CONNECT_BY_ROOT person_id     AS person_id
     ,     relatedperson_id
     FROM     all_relationships
     CONNECT BY NOCYCLE     person_id       = PRIOR relatedperson_id
             OR          relatedperson_id  = PRIOR person_id
,     got_family_id     AS
     SELECT       person_id
     ,       MIN (relatedperson_id)     AS family_id
     ,       ROW_NUMBER () OVER ( PARTITION BY  MIN (relatedperson_id)
                                  ORDER BY          person_id     DESC
                                )            AS r_num
     FROM       got_relatives
     GROUP BY  person_id
SELECT       p.person
,       CASE
           WHEN  f.r_num  = 1
           THEN  p.address
       END          AS address
,       p.person_id
,       f.family_id     
FROM       got_family_id  f
JOIN       persons      p  ON  p.person_id  = f.person_id
ORDER BY  family_id
,            person_id
;Output:
PERSON     ADDRESS          PERSON_ID  FAMILY_ID
John Smith                        101        101
Jane Smith                        102        101
Pete Smith 1 Oracle Drive         106        101
John Doe                          104        104
Jane Doe   10 Oracle Drive        105        104Obviously, you don't have have to display all the columns I displayed above In your first message, you said you wanted only person and address. In your last message, you said you only wanted person_id and family_id. Change the main SELECT clause any way you wish.
I used the lowest person_id in each family as the family_id. You could use DENSE_RANK if you really wanted to have the families numbered 1, 2, 3, ..., but I suspect you don't really care what family_id is, as long as all the members of the family have the same value.
You relationship table has some symmetrical relationships, such as
SELECT 104, 105 FROM dual union all
SELECT 105, 104 FROM dual union alland some asymmetrical relationships. For example, the only relationship involving person_id=106 is
SELECT 106, 101 FROM dualthat is, there is no mirror-image row:
-- SELECT 101, 106 FROM dual union all   -- THIS IS NOT IN THE SAMPLE DATAI assume there is no significance to that. As long as 101 and 106 appear on the same row, they are in the same family, regardless of which is the person_id and which is the relatedperson_id.
So the first thing I did above was make sure all the mirror-image rows were represented. That's what all_relationships does.
The next sub-query, got_relatives, is probably what you meant to adapt, but you left off the PRIOR operators.
Got_family_id actually does the grouping, and also computes r_num to determine which is the last member of the family. Only that family member's address will be displayed in the main query.
You could combine got_family_id and the main query; you don't really need a sub-query for that. It would be a little less coding, but I wrote it this way because I think it's a little easier to understand and maintain.