Recursive SQL statement (PostgreSQL 9.1.4)

Question

PostgreSQL 9.1

Business situation

Every month, there is a new batch of accounts given to a specific process. Every batch can be described by mon

Answer 1

        --
        -- rank the dates.
        -- , also fetch the the fields that seem to depend on them.
        -- (this should have been done in the data model)
        --
CREATE VIEW date_rank AS (
        SELECT uniq.granularity,uniq.entry_accounts,uniq.entry_amount
        , row_number() OVER(ORDER BY 0) AS zrank
        FROM ( SELECT DISTINCT granularity, entry_accounts, entry_amount FROM vintage_data)
             AS uniq
        );

-- SELECT * FROM date_rank ORDER BY granularity;
        --
        -- transform to an x*y matrix, avoiding the date key and the slack columns
        --
CREATE VIEW matrix_data AS (
        SELECT vd.distance_in_months AS xxx
        , dr.zrank AS yyy
        , vd.recovery_amount AS val
        FROM vintage_data vd
        JOIN date_rank dr ON dr.granularity = vd.granularity
        );
-- SELECT * FROM matrix_data;

        --
        -- In order to perform the reversed transformation:
        -- make the view insertable.
        -- INSERTS to matrix_data will percolate back into the vintage_data table
        -- (don't try this at home ;-)
        --
CREATE RULE magic_by_the_plasser AS
        ON INSERT TO matrix_data
        DO INSTEAD (
        INSERT INTO vintage_data (granularity,distance_in_months,entry_accounts,entry_amount,recovery_amount)
        SELECT dr.granularity, new.xxx, dr.entry_accounts, dr.entry_amount, new.val
        FROM date_rank dr
        WHERE dr.zrank = new.yyy
                ;
        );

        --
        -- This CTE creates the weights for a Pascal-triangle
        --
-- EXPLAIN -- ANALYZE
WITH RECURSIVE pascal AS (
        WITH empty AS (
                --
                -- "cart" is a cathesian product of X*Y
                -- its function is similar to a "calendar table":
                -- filling in the missing X,Y pairs, making the matrix "square".
                -- (well: rectangular, but in the given case nX==nY)
                --
                WITH cart AS (
                        WITH mmx AS (
                                WITH xx AS ( SELECT MIN(xxx) AS x0 , MAX(xxx) AS x1 FROM matrix_data)
                                SELECT generate_series(xx.x0,xx.x1) AS xxx
                                FROM xx
                                )
                        , mmy AS (
                                WITH yy AS ( SELECT MIN(yyy) AS y0 , MAX(yyy) AS y1 FROM matrix_data)
                                SELECT generate_series(yy.y0,yy.y1) AS yyy
                                FROM yy
                                )
                        SELECT * FROM mmx
                        JOIN mmy ON (1=1) -- Carthesian product here!
                        )
                --
                -- The (x,y) pairs that are not present in the current matrix
                --
                SELECT * FROM cart ca
                WHERE NOT EXISTS (
                        SELECT *
                        FROM matrix_data nx
                        WHERE nx.xxx = ca.xxx
                        AND nx.yyy = ca.yyy
                        )
                )
        SELECT md.yyy AS src_y
                , md.xxx AS src_x
                , md.yyy AS dst_y
                , md.xxx AS dst_x
                -- The filled-in matrix cells have weight 1
                , 1::numeric AS weight
        FROM matrix_data md
        UNION ALL
        SELECT pa.src_y AS src_y
                , pa.src_x AS src_x
                , em.yyy AS dst_y
                , em.xxx AS dst_x
                -- the derived matrix cells inherit weight/2 from both their parents
                , (pa.weight/2) AS weight
        FROM pascal pa
        JOIN empty em
                ON ( em.yyy = pa.dst_y+1 AND em.xxx = pa.dst_x)
                OR ( em.yyy = pa.dst_y AND em.xxx = pa.dst_x+1 )
        )
INSERT INTO matrix_data(yyy,xxx,val)
SELECT pa.dst_y,pa.dst_x
        ,SUM(ma.val*pa.weight)
FROM pascal pa
JOIN matrix_data ma ON pa.src_y = ma.yyy AND pa.src_x = ma.xxx
        -- avoid the filled-in matrix cells (which map to themselves)
WHERE NOT (pa.src_y = pa.dst_y AND pa.src_x = pa.dst_x)
GROUP BY pa.dst_y,pa.dst_x
        ;

        --
        -- This will also get rid of the matrix_data view and the rule.
        --
DROP VIEW date_rank CASCADE;
-- SELECT * FROM matrix_data ;

SELECT * FROM vintage_data ORDER BY granularity, distance_in_months;

RESULT:

NOTICE:  CREATE TABLE / PRIMARY KEY will create implicit index "vintage_data_pkey" for table "vintage_data"
CREATE TABLE
NOTICE:  ALTER TABLE / ADD UNIQUE will create implicit index "mx_xy" for table "vintage_data"
ALTER TABLE
INSERT 0 21
VACUUM
CREATE VIEW
CREATE VIEW
CREATE RULE
INSERT 0 15
NOTICE:  drop cascades to view matrix_data
DROP VIEW
 granularity | distance_in_months | entry_accounts | entry_amount |      recovery_amount      
-------------+--------------------+----------------+--------------+---------------------------
 2012-01-31  |                  1 |            200 |       100000 |                      1000
 2012-01-31  |                  2 |            200 |       100000 |                      2000
 2012-01-31  |                  3 |            200 |       100000 |                      3000
 2012-01-31  |                  4 |            200 |       100000 |                      3500
 2012-01-31  |                  5 |            200 |       100000 |                      3400
 2012-01-31  |                  6 |            200 |       100000 |                      3300
 2012-02-28  |                  1 |            250 |       150000 |                      1200
 2012-02-28  |                  2 |            250 |       150000 |                      1600
 2012-02-28  |                  3 |            250 |       150000 |                      1800
 2012-02-28  |                  4 |            250 |       150000 |                      1200
 2012-02-28  |                  5 |            250 |       150000 |                      1600
 2012-02-28  |                  6 |            250 |       150000 | 2381.25000000000000000000
 2012-03-31  |                  1 |            200 |        90000 |                      1300
 2012-03-31  |                  2 |            200 |        90000 |                      1200
 2012-03-31  |                  3 |            200 |        90000 |                      1400
 2012-03-31  |                  4 |            200 |        90000 |                      1000
 2012-03-31  |                  5 |            200 |        90000 | 2200.00000000000000000000
 2012-03-31  |                  6 |            200 |        90000 | 2750.00000000000000000000
 2012-04-30  |                  1 |            300 |       180000 |                      1600
 2012-04-30  |                  2 |            300 |       180000 |                      1500
 2012-04-30  |                  3 |            300 |       180000 |                      4000
 2012-04-30  |                  4 |            300 |       180000 | 2500.00000000000000000000
 2012-04-30  |                  5 |            300 |       180000 | 2350.00000000000000000000
 2012-04-30  |                  6 |            300 |       180000 | 2550.00000000000000000000
 2012-05-31  |                  1 |            400 |       225000 |                      2200
 2012-05-31  |                  2 |            400 |       225000 |                      6000
 2012-05-31  |                  3 |            400 |       225000 | 5000.00000000000000000000
 2012-05-31  |                  4 |            400 |       225000 | 3750.00000000000000000000
 2012-05-31  |                  5 |            400 |       225000 | 3050.00000000000000000000
 2012-05-31  |                  6 |            400 |       225000 | 2800.00000000000000000000
 2012-06-30  |                  1 |            100 |        60000 |                      1000
 2012-06-30  |                  2 |            100 |        60000 | 3500.00000000000000000000
 2012-06-30  |                  3 |            100 |        60000 | 4250.00000000000000000000
 2012-06-30  |                  4 |            100 |        60000 | 4000.00000000000000000000
 2012-06-30  |                  5 |            100 |        60000 | 3525.00000000000000000000
 2012-06-30  |                  6 |            100 |        60000 | 3162.50000000000000000000
(36 rows)