How to use conditional group by aggregations correctly

Question

I want to be able to count the total type of apples (organic only) from each continent, broken down by countries; including the total count if they're mixed.

For example, food item B1 is organic golden apples from the USA. Thus there should be a count of "1" golden_bag and "1" for organic. Now, A1 is also organic from Argentina - however, it has both granny and red delicious apples - thus it is counted as "1" mixed_bag and "1" for granny_bag and "1" for red_bag as well.

Finally, E1 and F1 are both fuji apples from laos, but one is organic the other isn't; so total count is 2 fuji_bag and it should have a total count of 1 for organic_fd.

Table X:
food_item | food_area | food_loc   | food_exp
A1          lxgs        argentina   1/1/20
B1          iyan        usa         5/31/21
C1          lxgs        peru        4/1/20
D1          wa8e        norway      10/1/19
E1          894a        laos        5/1/19
F1          894a        laos        9/17/19


Table Y:
food_item | organic
A1          Y
B1          Y
C1          N
D1          N
E1          Y
F1          N

Table Z:
food_item | food_type
A1          189
A1          190
B1          191
C1          189
D1          192
E1          193
F1          193

SELECT continent, country,
      SUM(organic)  AS organic_fd, SUM(Granny) AS granny_bag,
      SUM(Red_delc) AS red_bag,    SUM(Golden) AS golden_bag,
      SUM(Gala)     AS gala_bag,   SUM(Fuji)   AS fuji_bag,
      SUM(CASE WHEN Granny + Red_delc + Golden + Gala + Fuji > 1 THEN 1  ELSE 0 END) AS mixed_bag     
FROM (SELECT (CASE SUBSTR (x.food_area, 4, 1)
              WHEN 's' THEN 'SA' WHEN 'n' THEN 'NA'
              WHEN 'e' THEN 'EU' WHEN 'a' THEN 'AS' ELSE NULL END) continent,
          x.food_loc country, COUNT(y.organic) AS Organic
          COUNT(CASE WHEN z.food_type = '189' THEN 1 END) AS Granny,
          COUNT(CASE WHEN z.food_type = '190' THEN 1 END) AS Red_delc,
          COUNT(CASE WHEN z.food_type = '191' THEN 1 END) AS Golden,
          COUNT(CASE WHEN z.food_type = '192' THEN 1 END) AS Gala,
          COUNT(CASE WHEN z.food_type = '193' THEN 1 END) AS Fuji      
    FROM x LEFT JOIN z ON x.food_item = z.food_item
           LEFT JOIN y on x.food_item = y.food_item and y.organic = 'Y'    
               WHERE  x.exp_date > sysdate
    GROUP BY SUBSTR (x.food_area, 4, 1), x.food_loc, y.organic) h
GROUP BY h.continent, h.country, h.organic

I'm not getting the correct output, since for example, Laos will show TWICE to account for the organic count and non-organic count. So it will show 1 organic_fd and 0 organic_fd and 1 fuji_bag and the other line will be another 1 fuji_bag. I would like the TOTAL count. (Also, if I add more food items, my mixed_bag shows mostly "1" count for each record/lines).

Below is the desired output:

| continent | country   |organic_fd | granny_bag| red_bag| golden_bag| gala_bag|fuji_bag | mixed_bag
| SA        | argentina |    1      | 1         |   1    | 0         | 0       | 0       | 1
| SA        | peru      |    0      | 1         |   0    | 0         | 0       | 0       | 0
| NA        | usa       |    1      | 0         |   0    | 1         | 0       | 0       | 0
| EU        | norway    |    0      | 0         |   0    | 0         | 1       | 0       | 0
| AS        | laos      |    1      | 0         |   0    | 0         | 0       | 2       | 0

So, say I want to add another food item, G1 from Norway and it has 3 types of organic apples: fuji, red, granny... then Norway will now have a count of 1 for the following columns: mixed_bag, organic_fd, fuji_bag, red_bag ,granny_bag (in addition to the previous count of 1 gala_bag). If you add H1, which is exactly the same as G1, then it will now have a total count of 2 for the following: mixed_bag, organic_fd, fuji_bag,red_bag, granny_bag

score 1 · Answer 1 · answered Feb 07 '19 at 12:19

The query:

WITH
  t AS (
    SELECT
      CASE SUBSTR(X.food_area, LENGTH(X.food_area), 1)
        WHEN 's' THEN 'SA'
        WHEN 'n' THEN 'NA'
        WHEN 'e' THEN 'EU'
        WHEN 'a' THEN 'AS'
        ELSE NULL
      END AS continent,
      x.food_loc AS country,
      COUNT(DISTINCT CASE Y.organic WHEN 'Y' THEN X.food_item END) OVER (
        PARTITION BY x.food_loc
      ) AS organic_fd,
      CASE
        WHEN MIN(Z.food_type) OVER (
               PARTITION BY x.food_loc, X.food_item
             ) = Z.food_type AND
             MAX(Z.food_type) OVER (
               PARTITION BY x.food_loc, X.food_item
             ) > Z.food_type THEN 1 END AS mixed,
      Z.food_type
    FROM X
    JOIN Y ON X.food_item = Y.food_item
    JOIN Z ON Y.food_item = Z.food_item
  )
SELECT
  continent, country, organic_fd,
  COUNT(CASE WHEN food_type = '189' THEN 1 END) AS Granny,
  COUNT(CASE WHEN food_type = '190' THEN 1 END) AS Red_delc,
  COUNT(CASE WHEN food_type = '191' THEN 1 END) AS Golden,
  COUNT(CASE WHEN food_type = '192' THEN 1 END) AS Gala,
  COUNT(CASE WHEN food_type = '193' THEN 1 END) AS Fuji,
  COUNT(mixed) AS mixed_bag
FROM t
GROUP BY continent, country, organic_fd

You can try this query here: https://rextester.com/TSSH87409.

This works, but I need to learn more about "partition..." it gets way complicated when I add more columns — BFF, Feb 12 '19 at 05:21
@BFF, what troubles did you encounter? I would like to offer you this link as a starting point https://stackoverflow.com/a/561884/3350428. — Andrei Odegov, Feb 12 '19 at 07:49

score 0 · Accepted Answer · answered Feb 07 '19 at 11:24

You have one to many relationship between x and z, and join may produce many rows for each row in x, like in case of A1. So you have to number rows in x at first, this is what my subquery t1 do, except of mapping values. Then group them taking max() for each counted column (granny, organic etc.), like in subquery t2. Finally sum values.

^{dbfiddle demo}

with
  t1 as (
    select rn, food_item, food_area, food_loc country, food_exp, food_type,
           decode(substr(food_area, 4, 1), 's', 'SA', 'n', 'NA', 'e', 'EU', 'a', 'AS') continent,
           case organic when 'Y' then 1 else 0 end org,
           case when food_type = '189' then 1 else 0 end gra,
           case when food_type = '190' then 1 else 0 end red,
           case when food_type = '191' then 1 else 0 end gol,
           case when food_type = '192' then 1 else 0 end gal,
           case when food_type = '193' then 1 else 0 end fuj 
      from (select rownum rn, x.* from x) x join y using (food_item) join z using (food_item)
      where food_exp > sysdate),
  t2 as (
    select rn, country, continent, max(org) org, max(gra) gra, 
           max(red) red, max(gol) gol, max(gal) gal, max(fuj) fuj,
           case when max(gra) + max(red) + max(gol) + max(gal) + max(fuj) > 1 
                then 1 else 0 
            end mix
       from t1 group by rn, country, continent)
select continent, country, sum(org) organic_fd, sum(gra) granny, sum(red) red_delc, 
       sum(gol) golden_bag, sum(gal) gala_bag, sum(fuj) fuji_bag, sum(mix) mixed_bag 
  from t2 
  group by continent, country

Above query gave expected output, please test it and adjust if needed. I noticed you use left joins. If there is possibility that for some rows in X there is no data in Y or Z you may have to add nvl()s in calculations. Maybe you should also put mapped, hardcoded values into tables. Hardcoding them is not good practice. Hope this helps :)

Not sure why but "using" does not work for me. I appreciate the explanation, though. Thank you! — BFF, Feb 12 '19 at 05:20

How to use conditional group by aggregations correctly

2 Answers2