34

I have a table like this:

 Column  | Type | Modifiers 
---------+------+-----------
 country | text | 
 food_id | int  | 
 eaten   | date | 

And for each country, I want to get the food that is eaten most often. The best I can think of (I'm using postgres) is:

CREATE TEMP TABLE counts AS 
   SELECT country, food_id, count(*) as count FROM munch GROUP BY country, food_id;

CREATE TEMP TABLE max_counts AS 
   SELECT country, max(count) as max_count FROM counts GROUP BY country;

SELECT country, max(food_id) FROM counts 
   WHERE (country, count) IN (SELECT * from max_counts) GROUP BY country;

In that last statement, the GROUP BY and max() are needed to break ties, where two different foods have the same count.

This seems like a lot of work for something conceptually simple. Is there a more straight forward way to do it?

Ciro Santilli OurBigBook.com
  • 347,512
  • 102
  • 1,199
  • 985
Martin C. Martin
  • 341
  • 1
  • 3
  • 5

9 Answers9

25

It is now even simpler: PostgreSQL 9.4 introduced the mode() function:

select mode() within group (order by food_id)
from munch
group by country

returns (like user2247323's example):

country | mode
--------------
GB      | 3
US      | 1

See documentation here: https://wiki.postgresql.org/wiki/Aggregate_Mode

https://www.postgresql.org/docs/current/static/functions-aggregate.html#FUNCTIONS-ORDEREDSET-TABLE

jrouquie
  • 4,315
  • 4
  • 27
  • 43
20

PostgreSQL introduced support for window functions in 8.4, the year after this question was asked. It's worth noting that it might be solved today as follows:

SELECT country, food_id
  FROM (SELECT country, food_id, ROW_NUMBER() OVER (PARTITION BY country ORDER BY freq DESC) AS rn
          FROM (  SELECT country, food_id, COUNT('x') AS freq
                    FROM country_foods
                GROUP BY 1, 2) food_freq) ranked_food_req
 WHERE rn = 1;

The above will break ties. If you don't want to break ties, you could use DENSE_RANK() instead.

pilcrow
  • 56,591
  • 13
  • 94
  • 135
8
SELECT DISTINCT
"F1"."food",
"F1"."country"
FROM "foo" "F1"
WHERE
"F1"."food" =
    (SELECT "food" FROM
        (
            SELECT "food", COUNT(*) AS "count"
            FROM "foo" "F2" 
            WHERE "F2"."country" = "F1"."country" 
            GROUP BY "F2"."food" 
            ORDER BY "count" DESC
        ) AS "F5"
        LIMIT 1
    )

Well, I wrote this in a hurry and didn't check it really well. The sub-select might be pretty slow, but this is shortest and most simple SQL statement that I could think of. I'll probably tell more when I'm less drunk.

PS: Oh well, "foo" is the name of my table, "food" contains the name of the food and "country" the name of the country. Sample output:

   food    |  country   
-----------+------------
 Bratwurst | Germany
 Fisch     | Frankreich
jkramer
  • 15,440
  • 5
  • 47
  • 48
6

try this:

Select Country, Food_id
From Munch T1
Where Food_id= 
    (Select Food_id
     from Munch T2
     where T1.Country= T2.Country
     group by Food_id
     order by count(Food_id) desc
      limit 1)
group by Country, Food_id
Jamal Hansen
  • 954
  • 7
  • 16
3

Try something like this

select country, food_id, count(*) cnt 
into #tempTbl 
from mytable 
group by country, food_id

select country, food_id
from  #tempTbl as x
where cnt = 
  (select max(cnt) 
  from mytable 
  where country=x.country 
  and food_id=x.food_id)

This could be put all into a single select, but I don't have time to muck around with it right now.

Good luck.

John MacIntyre
  • 12,910
  • 13
  • 67
  • 106
3

Here's how to do it without any temp tables:

Edit: simplified

select nf.country, nf.food_id as most_frequent_food_id
from national_foods nf
group by country, food_id 
having
  (country,count(*)) in (  
                        select country, max(cnt)
                        from
                          (
                          select country, food_id, count(*) as cnt
                          from national_foods nf1
                          group by country, food_id
                          )
                        group by country
                        having country = nf.country
                        )
JosephStyons
  • 57,317
  • 63
  • 160
  • 234
  • I'd be interested in seeing the plan for this execution vs. the temp table -- those "having" clauses are evaluated _after_ select retrieves matching rows, right? Seems like there may be a boatload of extra IO. – Ken Gentle Dec 05 '08 at 18:32
  • There are a couple of full table scans in the plan, yes. – JosephStyons Dec 05 '08 at 20:12
3
SELECT country, MAX( food_id )
  FROM( SELECT m1.country, m1.food_id
          FROM munch m1
         INNER JOIN ( SELECT country
                           , food_id
                           , COUNT(*) as food_counts
                        FROM munch m2
                    GROUP BY country, food_id ) as m3
                 ON m1.country = m3.country
         GROUP BY m1.country, m1.food_id 
        HAVING COUNT(*) / COUNT(DISTINCT m3.food_id) = MAX(food_counts) ) AS max_foods
  GROUP BY country

I don't like the MAX(.) GROUP BY to break ties... There's gotta be a way to incorporate eaten date into the JOIN in some way to arbitrarily select the most recent one...

I'm interested on the query plan for this thing if you run it on your live data!

Matt Rogish
  • 24,435
  • 11
  • 76
  • 92
3
select country,food_id, count(*) ne  
from   food f1  
group by country,food_id    
having count(*) = (select max(count(*))  
                   from   food f2  
                   where  country = f1.country  
                   group by food_id)  
Theo
  • 452
  • 3
  • 3
3

Here is a statement which I believe gives you what you want and is simple and concise:

select distinct on (country) country, food_id
from munch
group by country, food_id
order by country, count(*) desc

Please let me know what you think.

BTW, the distinct on feature is only available in Postgres.

Example, source data:

country | food_id | eaten
US        1         2017-1-1
US        1         2017-1-1
US        2         2017-1-1
US        3         2017-1-1
GB        3         2017-1-1
GB        3         2017-1-1
GB        2         2017-1-1

output:

country | food_id
US        1
GB        3
JCF
  • 651
  • 5
  • 9
  • 1
    If you are going to propose a new answer after all this time, I recommend trying it on a sample table, and posting the results you get. Also, please mention which database server you are using (mysql or whatever). – ToolmakerSteve Jan 12 '17 at 01:24
  • 3
    The _distinct on_ feature is only available in Postgres so, I'm not sure how you would do something like this in another database. OP is using Postgres so it seems appropriate. I wrote this using the database table suggested by op called _munch_ which has three fields: country (text), food_id (int), and eaten (date) – JCF Jan 12 '17 at 21:52