Reputation: 13
I am trying to calculate streaks across a result database based on win loss and draw criteria.
Goal: Get longest streak of wins/non-wins grouped by team
I have tried different SQL query suggestions from other threads, but I am either missing out the grouping or the team column and often only takes 2-way option(Win and loss) - I need 3-way option(Win, Los, Draw incl. non-win,non-loss and non-draw)
Looked at this - https://www.sqlteam.com/articles/detecting-runs-or-streaks-in-your-data
But I have no clue how to get the team incl. grouping into the mix
Scheme:
CREATE TABLE teamresults (matchid varchar(255), date DATE, time TIME, team varchar(255), teamresult varchar(255))
Data sample:
INSERT INTO teamresults (matchid,"date","time",team,teamresult) VALUES
('030420181800acfc','2018-04-03','18:00:00','AC Horsens','L')
,('080420181600brac','2018-04-08','16:00:00','AC Horsens','L')
,('150420181400aaac','2018-04-15','14:00:00','AC Horsens','L')
,('180420181800acfc','2018-04-18','18:00:00','AC Horsens','D')
,('210420181600fcac','2018-04-21','16:00:00','AC Horsens','L')
,('270420181900acfc','2018-04-27','19:00:00','AC Horsens','L')
,('040520181900acaa','2018-05-04','19:00:00','AC Horsens','W')
,('110520181900fcac','2018-05-11','19:00:00','AC Horsens','L')
,('180520182000acbr','2018-05-18','20:00:00','AC Horsens','D')
,('210520181800fcac','2018-05-21','18:00:00','AC Horsens','L')
,('120520191200veac','2019-05-12','12:00:00','AC Horsens','W')
,('190520191400acve','2019-05-19','14:00:00','AC Horsens','D')
,('140720191400acfc','2019-07-14','14:00:00','AC Horsens','L')
,('210720191200siac','2019-07-21','12:00:00','AC Horsens','W')
,('270720191730acfc','2019-07-27','17:30:00','AC Horsens','L')
,('040820191600brac','2019-08-04','16:00:00','AC Horsens','W')
,('010420181400hoag','2018-04-01','14:00:00','AGF','W')
,('080420181800agsi','2018-04-08','18:00:00','AGF','W')
,('130420181900agfc','2018-04-13','19:00:00','AGF','W')
,('170420181900fcag','2018-04-17','19:00:00','AGF','L')
,('230420181900agho','2018-04-23','19:00:00','AGF','L')
,('300420181900siag','2018-04-30','19:00:00','AGF','W')
,('060520181200agob','2018-05-06','12:00:00','AGF','W')
,('130520181800obag','2018-05-13','18:00:00','AGF','W')
,('190520181600ags�','2018-05-19','16:00:00','AGF','D')
;
The below query does work, but does only a take single input statement - So I can only get streak of win, loss or draws - Not non-win, non-loss and non-draws.
SELECT
team,
MAX(cnt)
FROM
(
SELECT
team,
COUNT(*) AS cnt
FROM
(
SELECT
team,
date,
teamresult,
SUM(CASE WHEN teamresult <> 'W' THEN 1 else 0 END)
OVER (PARTITION BY team
ORDER BY date
ROWS UNBOUNDED PRECEDING) AS dummy
FROM teamresults
) dt
WHERE teamresult = 'W'
GROUP BY team, dummy
) dt
GROUP BY team;
I also want to be able to find longest non-streaks grouped by team
SQL fiddle is available here: http://sqlfiddle.com/#!18/3a2ac/1
Thanks in advance
Update: Gordon queries are working, but these queries does not work in postgres/cockroach - So now trying to convert them to supported queries via window functions rank()
select team, teamresult, cnt, rank() over (order by cnt desc) from
(SELECT team, teamresult, COUNT(*) as cnt
FROM (SELECT tr.*,
ROW_NUMBER() OVER (PARTITION BY team ORDER BY "date", "time") as seqnum,
ROW_NUMBER() OVER (PARTITION BY team, teamresult ORDER BY "date", "time") as seqnum_r
FROM teamresults tr
) tr
WHERE teamresult = 'W'
GROUP BY team, teamresult, (seqnum - seqnum_r)
ORDER BY ROW_NUMBER() OVER (PARTITION BY team ORDER BY COUNT(*) DESC)) as ranked
This does gives me an output like this(Data sample from my DB):
FC København W 9 1
AaB W 8 2
FC København W 8 2
FC København W 8 2
FC København W 8 2
Brøndby IF W 7 6
FC Midtjylland W 7 6
FC København W 7 6
FC København W 7 6
FC København W 7 6
Esbjerg fB W 6 11
FC Midtjylland W 6 11
AaB W 6 11
Brøndby IF W 6 11
Brøndby IF W 6 11
Expected output:
Team Longest consecutive streak
FC København 9
AaB 8
Brøndby IF 7
FC Midtjylland 7
Esbjerg fB 6
Upvotes: 0
Views: 746
Reputation: 13
Thanks to Gordon, I solved my case with the following queries:
Longest consecutive streak of Wins(replace W with D for draw, and L for loss) grouped by team
select team, max(cnt) longeststreak from (
SELECT team, teamresult, COUNT(*) as cnt
FROM (SELECT tr.*,
RANK() OVER (PARTITION BY team ORDER BY "date", "time") as seqnum,
RANK() OVER (PARTITION BY team, teamresult ORDER BY "date", "time") as seqnum_r
FROM teamresults tr
) tr
WHERE teamresult = 'W'
GROUP BY team, teamresult, (seqnum - seqnum_r)
ORDER BY RANK() OVER (PARTITION BY team ORDER BY COUNT(*) DESC)
)
group by team
order by longeststreak DESC
Longest consecutive streak of non-wins(replace W with D for non-draw, and L for non-loss) grouped by team
select team, max(cnt) longestnonstreak from (
SELECT team,
(CASE WHEN teamresult = 'W' THEN 'W' END) as is_win,
COUNT(*) as cnt
FROM (SELECT tr.*,
RANK() OVER (PARTITION BY team ORDER BY "date", "time") as seqnum,
RANK() OVER (PARTITION BY team, (CASE WHEN teamresult = 'W' THEN 'W' END) ORDER BY "date", "time") as seqnum_r
FROM teamresults tr
) tr
GROUP BY team, (CASE WHEN teamresult = 'W' THEN 'W' END), (seqnum - seqnum_r)
ORDER BY RANK() OVER (PARTITION BY team ORDER BY COUNT(*) DESC)
)
group by team
order by longestnonstreak desc
Thanks to Gordon for assisting in the solution.
Upvotes: 1
Reputation: 1270391
You can get all the streaks using:
SELECT team, teamresult, COUNT(*) as cnt
FROM (SELECT tr.*,
ROW_NUMBER() OVER (PARTITION BY team ORDER BY "date", "time") as seqnum,
ROW_NUMBER() OVER (PARTITION BY team, teamresult ORDER BY "date", "time") as seqnum_r
FROM teamresults tr
) tr
GROUP BY team, teamresult, (seqnum - seqnum_r);
You can modify this to get the longest winning streak per team:
SELECT TOP(1) WITH TIES team, teamresult, COUNT(*) as cnt
FROM (SELECT tr.*,
ROW_NUMBER() OVER (PARTITION BY team ORDER BY "date", "time") as seqnum,
ROW_NUMBER() OVER (PARTITION BY team, teamresult ORDER BY "date", "time") as seqnum_r
FROM teamresults tr
) tr
WHERE teamresult = 'W'
GROUP BY team, teamresult, (seqnum - seqnum_r)
ORDER BY ROW_NUMBER() OVER (PARTITION BY team ORDER BY COUNT(*) DESC);
If you just want the longest streak of any type, remove the WHERE
. If you want the longest for each type per team, then add teamresult
to the PARTITION BY
.
Here is a db<>fiddle.
EDIT:
If you want non-wins, you need to partition by an expression:
SELECT TOP(1) WITH TIES team,
(CASE WHEN teamresult = 'W' THEN 'W' END) as is_win,
COUNT(*) as cnt
FROM (SELECT tr.*,
ROW_NUMBER() OVER (PARTITION BY team ORDER BY "date", "time") as seqnum,
ROW_NUMBER() OVER (PARTITION BY team, (CASE WHEN teamresult = 'W' THEN 'W' END) ORDER BY "date", "time") as seqnum_r
FROM teamresults tr
) tr
-- WHERE teamresult = 'W'
GROUP BY team, (CASE WHEN teamresult = 'W' THEN 'W' END), (seqnum - seqnum_r)
ORDER BY ROW_NUMBER() OVER (PARTITION BY team ORDER BY COUNT(*) DESC)
Upvotes: 1