Postgres Pro
Downloads
Home   >   mailing lists

Re: recursive sql - Mailing list pgsql-sql

From Igor Andriychuk
Subject Re: recursive sql
Date
Msg-id 2F5CAE4D-76DA-44D8-AFCA-CABF00DA9422@gmail.com
Whole thread Raw
In response to Re: recursive sql  (ml@ft-c.de)
List pgsql-sql
Tree view
Oh, yes, in this case you need a recursion. This is something that came on my mind in short observation:


with recursive r as(              
select ts, c, row_id from rnk where rnk.row_id = 1
union 
select rnk.ts, rnk.c*0.33 + r.c*0.76, rnk.row_id
from
r
join
rnk
on
r.row_id = rnk.row_id - 1
),
rnk as(
select *, row_number() over(order by ts) row_id from tt
)
select ts, c from r order by ts;

Tested it :-)


On Aug 9, 2020, at 6:25 AM, ml@ft-c.de wrote:

Hello,

sorry for my short explanation. It was not enough to understand the my task/target.

These are the basic computation for an exponential moving average (ema)
an statistic indicator for trading data.

The components of trading data are
timestamp, High, Low, Open and Close value
For this indicator I need the timestamp and the close value, not more.

For the current day (period) the formula is

EMA = Close(t) * SF  + ( (1-SF) * EMA(t-1) )

where Smoothing Factor SF = 2 / (n+1)

The best way is, to explain it with an example:
day close    SF    close  1-SF  EMA(t-1)  = part_of_result
1   105,5
2   104     0.33 * 104 + 0.76 * 105,5    = 105.005
3   103.5   0.33 * 103 + 0.76 * 105.005  = 104.508
4   102     0.33 * 102 + 0.76 * 104.508  = 103.680
5   101     0.33 * 101 + 0.76 * 103.680  = 102.795
6   100     0.33 * 100 + 0.76 * 102.795  = 101.872

0.33 and 0.67 are the SF
You see, the result of one line is a component of the next line.
The result for day 6 is 101.872

I need the close value of the current day and
the the close value of the previous day. But before, it must be calculated.

I believe, the best way is, to do it with
"with recursive"

Franz


On 8/9/20 2:08 PM, Samed YILDIRIM wrote:
Hi Frank,
It seems I need to read more carefully :)
With window functions;
pgsql-sql=# select *,sum(c) over (order by ts) from tt;
ts | c | sum
---------------------+---+-----
2019-12-31 00:00:00 | 1 | 1
2020-01-01 00:00:00 | 2 | 3
2020-07-02 00:00:00 | 3 | 6
2020-07-06 00:00:00 | 4 | 10
2020-07-07 00:00:00 | 5 | 15
2020-07-08 00:00:00 | 6 | 21
(6 rows)
With recursive query:
pgsql-sql=# with recursive rc as (
select * from (select ts,c,c as c2 from tt order by ts asc limit 1) sq1
union
select * from (select tt.ts,tt.c,tt.c+rc.c2 as c2 from tt, lateral (select * from rc order by ts desc limit 1) rc where tt.ts > rc.ts order by tt.ts asc limit 1) sq2
)
select * from rc;
ts | c | c2
---------------------+---+----
2019-12-31 00:00:00 | 1 | 1
2020-01-01 00:00:00 | 2 | 3
2020-07-02 00:00:00 | 3 | 6
2020-07-06 00:00:00 | 4 | 10
2020-07-07 00:00:00 | 5 | 15
2020-07-08 00:00:00 | 6 | 21
(6 rows)
Best regards.
Samed YILDIRIM
09.08.2020, 14:57, "ml@ft-c.de" <ml@ft-c.de>:
   Hallo,
   with the window function lag there is a shift of one or more rows. Every
   row connects to the previous row := lag(column,1).
   What I am looking for:
   ts c c2
   .. 1 1 -- or null in the first row
   .. 2 3 -- it is the result of 1 + 2
   .. 3 6 -- it is the result of 3 + 3
   .. 4 10 -- it is the result of 6 + 4
   Franz
   On 8/9/20 12:38 PM, Samed YILDIRIM wrote:
         Hi Franz,
         Simply you can use window functions[1][2].
         pgsql-sql=# select *, lag(c) over (order by ts) as c2 from tt;
         ts | c | c2
         ---------------------+---+----
         2019-12-31 00:00:00 | 1 |
         2020-01-01 00:00:00 | 2 | 1
         2020-07-02 00:00:00 | 3 | 2
         2020-07-06 00:00:00 | 4 | 3
         2020-07-07 00:00:00 | 5 | 4
         2020-07-08 00:00:00 | 6 | 5
         (6 rows)
         I personally prefer to use window functions due to their
       simplicity. If
         you still want to use recursive query: [3]
         pgsql-sql=# with recursive rc as (
         select * from (select ts,c,null::numeric as c2 from tt order
       by ts asc
         limit 1) k1
         union
         select * from (select tt.ts,tt.c,rc.c as c2 from tt, lateral
       (select *
         from rc) rc where tt.ts > rc.ts order by tt.ts asc limit 1) k2
         )
         select * from rc;
         ts | c | c2
         ---------------------+---+----
         2019-12-31 00:00:00 | 1 |
         2020-01-01 00:00:00 | 2 | 1
         2020-07-02 00:00:00 | 3 | 2
         2020-07-06 00:00:00 | 4 | 3
         2020-07-07 00:00:00 | 5 | 4
         2020-07-08 00:00:00 | 6 | 5
         (6 rows)
         [1]: https://www.postgresql.org/docs/12/functions-window.html
         [2]: https://www.postgresql.org/docs/12/tutorial-window.html
         [3]: https://www.postgresql.org/docs/12/queries-with.html
         Best regards.
         Samed YILDIRIM
         09.08.2020, 09:29, "ml@ft-c.de <mailto:ml@ft-c.de>"
       <ml@ft-c.de <mailto:ml@ft-c.de>>:
             Hello,
             the table
             create table tt (
                 ts timestamp,
                 c numeric) ;
             insert into tt values
                ('2019-12-31',1), ('2020-01-01',2),
                ('2020-07-02',3), ('2020-07-06',4),
                ('2020-07-07',5), ('2020-07-08',6);
             My question: It is possible to get an
                 additional column (named c2)
                 with
                 ( c from current row ) + ( c2 from the previous row )
       as c2
             the result:
             ts c c2
             .. 1 1 -- or null in the first row
             .. 2 3
             .. 3 6
             .. 4 10
             ...
             with recursive ema as ()
             select ts, c,
                 -- many many computed_rows
                 -- <code> as c2
             from tt -- <- I need tt on this place
             thank you for help
             Franz

pgsql-sql by date:

Previous
From: ml@ft-c.de
Date:
Subject: Re: recursive sql
Next
From: Igor Andriychuk
Date:
Subject: Re: recursive sql