Wrapping my head around PIVOT - Arggggg

Question

Ok, I am stumped. What I want seems so simple to me....
======= Table Def
CREATE TABLE "TI" (
"SNUM" NUMBER,
"SYMBOL" VARCHAR2(30),
"OPEN" NUMBER(15, 5),
"LOW" NUMBER(15, 5),
"HIGH" NUMBER(15, 5),
"CLOSE" NUMBER(15, 5),
"ADJ_CLOSE" NUMBER(15, 5),
"VOLUME" NUMBER(15, 5),
"CHANGE_PCT" NUMBER(15, 5),
"CHANGE_AMT" NUMBER(15, 5),
"TIMESTAMP" TIMESTAMP(9) DEFAULT systimestamp ,
"TRADE_DATE" DATE,
"DATA_SOURCE" NUMBER)
============ data
insert into TI (6975 ,'VT', 48.2,48.11, 48.41,48.23,48.23,200500,.009,.43,'01-JAN-11','08.32.57.366984000 PM','03-JAN-11',2)
insert into TI (6995,' AGG',105.4,105.3,105.68,105.63,105.63,739700,-.00113,-.12,'31-JAN-11 08.33.18.137327000 PM','03-JAN-11',2)
insert into TI(7015,'SPY',126.71,125.7,127.6,127.05,127.05,138725200,.01034,1.3,'31-JAN-11 08.33.34.708030000 PM','03-JAN-11',2)
So as you can see these are all on the same date. We currently collect three indices every day, but a user with the click of a mouse could add another or take one away.
What I am trying to do is get output gives a date and the indices contained in a given day and their adj_close as a single return value from a query.
I have spent the day trying to wrap my head around PIVOT or LISTAGG but just can't seem to get a grasp of the damn thing. All the examples I have seen incluse sums and such and I need, nor in fact do I want anything summed. This would seem to come down to grouping by the trade_date and that where something is just not making sense.
More then an answer ( although an answer would be very helpful :) ) It would be great if someone could simply lay out how this is supposed to work. Uhm did I say an answer would be helpful <big cheesy smile >
Thanks in advance...

Answer 1

Hi,
Thanks for posting the CREATE TABLE and INSERT statements. Please test any code before you post it; there seem to be a lot of mistakes in those 4 statements. Don't forget to post the results you want from the sample data, and your version of Oracle.
One thing that is common to all forms of pivoting or string aggregation is that you need to know, for each row of input, to which row and which column(s) (or position(s)) of output it corresponds. Sometimes one or both of these are already in the data, but sometimes they have to be derived from the data. I'll assume you have one of each: that trade_date indicates the output row, and that the column you called timestamp (but which I'll call tmstmp, since timestamp isn't a good column name) can be used to tell the output position: the earliestt tmstmp goes on the left, the 2nd lowest tmstmp goes next, the 3rd lowest tmptmp goes next, and so on. We can use the analytic ROW_NUMBER function to assing consectuive numbers (1 for lowest, 2 for 2nd lowest, 3 for 3rd lowest, and so on) to the different tmstmps for each trade_date. I'll assume you don't know how many of these there will be on any given trade_date, and there may be different numbers of them on different trade_dates in the same output. I'll assume you're using Oracle 10.
If you have an unknown number of items, then string aggregation (concatenating all the values into one big sting column) is a lot easier and more robust than pivoting, so let's do it that way.
Here's one way to do string aggregation:
WITH     got_c_num     AS
     SELECT     adj_close
     ,     trade_date
     ,     ROW_NUMBER () OVER ( PARTITION BY  trade_date
                         ORDER BY        tmstmp
                       )     AS c_num
     FROM     ti
--     WHERE     ...     -- if you need any filtering, put it here
SELECT       trade_date
,       REPLACE ( SYS_CONNECT_BY_PATH ( TO_CHAR (adj_close, '999990.000')
            )          AS indices
FROM       got_c_num
WHERE       CONNECT_BY_ISLEAF     = 1
START WITH     c_num          = 1
CONNECT BY     c_num          = PRIOR c_num + 1
     AND     trade_date     = PRIOR trade_date
ORDER BY  trade_date
;Output from your sample data:
TRADE_DAT INDICES
03-JAN-11       48.230     105.630     127.050CONNECT BY is used for parent-child realtionships of variable length within one table. Here, we have assigned c_num to be 1 for the first input row for each trade_date, 2 for the next, and so on. In the CONNECT BY uery above, we're saying that every parent-child-grandchild-... chain will start with a row numbered c_num=1, and that a row will be considered the child of another row if its c_num is 1 higher, and their trade_dates are the same. (That's what the START WITH and CONNECT BY clauses are doing.)
SYS_CONNECT_BY_PATH gives a delimited list of items, starting from the root (the item with no parent, c_num=1) and showing the lineage down to each item in the original table that is descendedd from some root. That means we get 3 rows of output in this case: one showing just c_num=1, another with that same c_num=1 and c_num=2. and finally one with all 3 c_nums. We're only interested in the last one, the onme with no child, so the WHERE clause discards the others.
By using TO_CHAR in SYS_CONNECT_BY_PATH, we get items that are padded to the same length. That way, when they are displayed, they will appear to line up in neat columns. SYS_CONNECT_BY_PATH reuires that you delimit the output items with some string that cannot possibly occur in the items themselves. I chose '/', but, since we don't want to see those '/'s, I used REPLACE to get rid of them once we were finished doing the SYS_CONNECT_BY_PATH.