Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Transpose rows and columns (a.k.a. pivot) only with a minimum COUNT()?

Here's my table 'tab_test':

year    animal  price
2000    kittens 79
2000    kittens 93
2000    kittens 100
2000    puppies 15
2000    puppies 32
2001    kittens 31
2001    kittens 17
2001    puppies 65
2001    puppies 48
2002    kittens 84
2002    kittens 86
2002    puppies 15
2002    puppies 95
2003    kittens 62
2003    kittens 24
2003    puppies 36
2003    puppies 41
2004    kittens 65
2004    kittens 85
2004    puppies 58
2004    puppies 95
2005    kittens 45
2005    kittens 25
2005    puppies 15
2005    puppies 35
2006    kittens 50
2006    kittens 80
2006    puppies 95
2006    puppies 49
2007    kittens 40
2007    kittens 19
2007    puppies 81
2007    puppies 38
2008    kittens 37
2008    kittens 51
2008    puppies 29
2008    puppies 72
2009    kittens 84
2009    kittens 26
2009    puppies 49
2009    puppies 34
2010    kittens 75
2010    kittens 96
2010    puppies 18
2010    puppies 26
2011    kittens 35
2011    kittens 21
2011    puppies 90
2011    puppies 18
2012    kittens 12
2012    kittens 23
2012    puppies 74
2012    puppies 79

Here's some code that transposes the rows and columns so I get an average for 'kittens' and 'puppies':

SELECT
    year,
    AVG(CASE WHEN animal = 'kittens' THEN price END) AS "kittens",
    AVG(CASE WHEN animal = 'puppies' THEN price END) AS "puppies"
FROM tab_test
GROUP BY year
ORDER BY year;

The output for the code above is:

    year    kittens puppies
    2000    90.6666666666667    23.5
    2001    24.0    56.5
    2002    85.0    55.0
    2003    43.0    38.5
    2004    75.0    76.5
    2005    35.0    25.0
    2006    65.0    72.0
    2007    29.5    59.5
    2008    44.0    50.5
    2009    55.0    41.5
    2010    85.5    22.0
    2011    28.0    54.0
    2012    17.5    76.5

What I'd like is a table like the second one, but it would only contain items which had a COUNT() of at least 3 in the first table. In other words, the goal is to have this as output:

year    kittens
2000    90.6666666666667

There were at least 3 instances of 'kitten' in the first table.
Is this possible in PostgreSQL?

like image 853
user1626730 Avatar asked Oct 31 '12 21:10

user1626730


People also ask

How do I transpose rows to columns in pivot table?

Activate the pivot table. Double-click a table and click the Transpose control on the Edit mode Properties tab. Click the x control to exit pivot table edit mode.

How do I switch rows and columns in a pivot table?

In the PivotTable, right-click the row or column label or the item in a label, point to Move, and then use one of the commands on the Move menu to move the item to another location.

How do I transpose rows to columns in MySQL?

If you want to transpose only select row values as columns, you can add WHERE clause in your 1st select GROUP_CONCAT statement. If you want to filter rows in your final pivot table, you can add the WHERE clause in your SET statement.


2 Answers

CASE

If your case is as simple as demonstrated, a CASE statement will do:

SELECT year
     , sum(CASE WHEN animal = 'kittens' THEN price END) AS kittens
     , sum(CASE WHEN animal = 'puppies' THEN price END) AS puppies
FROM  (
   SELECT year, animal, avg(price) AS price
   FROM   tab_test
   GROUP  BY year, animal
   HAVING count(*) > 2
   ) t
GROUP  BY year
ORDER  BY year;

Doesn't matter whether you use sum(), max() or min() as aggregate function in the outer query. They all result in the same value in this case.

SQL Fiddle

crosstab()

With more categories it will be simpler with a crosstab() query. This should also be faster for bigger tables.

You need to install the additional module tablefunc (once per database). Since Postgres 9.1 that's as simple as:

CREATE EXTENSION tablefunc;

Details in this related answer:

  • PostgreSQL Crosstab Query
SELECT * FROM crosstab(
      'SELECT year, animal, avg(price) AS price
       FROM   tab_test
       GROUP  BY animal, year
       HAVING count(*) > 2
       ORDER  BY 1,2'

      ,$$VALUES ('kittens'::text), ('puppies')$$)
AS ct ("year" text, "kittens" numeric, "puppies" numeric);

No sqlfiddle for this one because the site doesn't allow additional modules.

Benchmark

To verify my claims I ran a quick benchmark with close to real data in my small test database. PostgreSQL 9.1.6. Test with EXPLAIN ANALYZE, best of 10:

Test setup with 10020 rows:

CREATE TABLE tab_test (year int, animal text, price numeric);

-- years with lots of rows
INSERT INTO tab_test
SELECT 2000 + ((g + random() * 300))::int/1000 
     , CASE WHEN (g + (random() * 1.5)::int) %2 = 0 THEN 'kittens' ELSE 'puppies' END
     , (random() * 200)::numeric
FROM   generate_series(1,10000) g;

-- .. and some years with only few rows to include cases with count < 3
INSERT INTO tab_test
SELECT 2010 + ((g + random() * 10))::int/2
     , CASE WHEN (g + (random() * 1.5)::int) %2 = 0 THEN 'kittens' ELSE 'puppies' END
     , (random() * 200)::numeric
FROM   generate_series(1,20) g;

Results:

@bluefeet
Total runtime: 95.401 ms

@wildplasser (different results, includes rows with count <= 3)
Total runtime: 64.497 ms

@Andreiy (+ ORDER BY)
& @Erwin1 - CASE (both perform about the same)
Total runtime: 39.105 ms

@Erwin2 - crosstab()
Total runtime: 17.644 ms

Largely proportional (but irrelevant) results with only 20 rows. Only @wildplasser's CTE has more overhead and spikes a little.

With more than a handful of rows, crosstab() quickly takes lead. @Andreiy's query performs about the same as my simplified version, aggregate function in outer SELECT (min(), max(), sum()) makes no measurable difference (just two rows per group).

Everything as expected, no surprises, take my setup and try it @home.

like image 121
Erwin Brandstetter Avatar answered Nov 15 '22 22:11

Erwin Brandstetter


Here's an alternative to @bluefeet's suggestion, which is somewhat similar but avoids the join (instead, the upper level grouping is applied to the already grouped result set):

SELECT
  year,
  MAX(CASE animal WHEN 'kittens' THEN avg_price END) AS "kittens",
  MAX(CASE animal WHEN 'puppies' THEN avg_price END) AS "puppies"
FROM (
  SELECT
    animal,
    year,
    COUNT(*) AS cnt,
    AVG(Price) AS avg_price
  FROM tab_test
  GROUP BY
    animal,
    year
) s
WHERE cnt >= 3
GROUP BY
  year
;
like image 34
Andriy M Avatar answered Nov 15 '22 21:11

Andriy M