7

Having the following data in a table:

ID     Category   Value
1234   Cat01      V001
1234   Cat02      V002
1234   Cat03      V003
1234   Cat03      V004
1234   Cat03      V005

I want to have the following output:

ID     Cat01  Cat02   Cat03
1234   V001   V002    V003
1234   V001   V002    V004
1234   V001   V002    V005

The output I want to achieve is a kind of pivot table where I have all the values vertically in a table and I want to have those values, horizontally, having the category as a column. But there are some categories that have multiples values, in that case, I need to repeat the values of all other categories and create a row per each repeated value

How can it be done in PostgreSQL?

0

3 Answers 3

6

This is a tricky one. crosstab() expects one (or no) value per category for each row_name.

We can work around this restriction like this:

SELECT id
     , COALESCE(cat01, max(cat01) OVER w)
     , COALESCE(cat02, max(cat02) OVER w)
     , COALESCE(cat03, max(cat03) OVER w)
FROM   crosstab(
   'SELECT id::text || row_number() OVER (PARTITION BY id, category ORDER BY value) * -1 AS ext_id
         , id, category, value
    FROM   tbl
    ORDER  BY ext_id, category, value'

   ,$$VALUES ('Cat01'::text), ('Cat02'), ('Cat03')$$
   ) AS ct (xid text, id int, cat01 text, cat02 text, cat03 text)
WINDOW w AS (PARTITION BY id);

Returns your desired result.

How?

  1. Add an extended id: ext_id from the existing id and a row number for each value of the category for the same id. This way we ensure as many rows per id in as there are values for the most common category. We get a derived table like this to build our crosstab() on:

    ext_id   | id   | category | value
    ---------+------+----------+-------
    '1234-1' | 1234 | 'Cat01'  | 'V001'
    '1234-1' | 1234 | 'Cat02'  | 'V002'
    '1234-1' | 1234 | 'Cat03'  | 'V003'
    '1234-2' | 1234 | 'Cat03'  | 'V004'
    '1234-3' | 1234 | 'Cat03'  | 'V005'
    
  2. Now we can feed it to crosstab() using the safe 2-parameter form for missing attributes. Read the basics first if you are not familiar with this:

  1. Your question leaves room for interpretation. My solution pairs the lowest values per category first and keeps filling the following rows until there are no values left. (We could combine multiple values per category any other way, it has not been defined.) If a category is short of values for a given id, the rest is filled in with NULL values.

  2. In the final step I replace those NULL values with the maximum value of each category per id:

     COALESCE(cat01, max(cat01) OVER (PARTITION BY id, category))
    

    which is effectively the same as:

     max(cat01) OVER (PARTITION BY id, category)
    

    I am hoping to make it slightly faster if we only default to the window function if the value is NULL.

1

Take a look here for an example of how to use the CROSSTAB function. Also, take a good look at Erwin Brandstetter's post in the same thread and links within (especially the "Basics for crosstab():" link.

Be careful with NULLs (see the discussion in link).

If you're not using a PostgreSQL version compiled from source, then all you have to do to access the CROSSTAB function is to input

CREATE EXTENSION tablefunc;

On the command line (see EB link).

I'm not sure that I've totally grasped your supplementary information, but maybe the CTE approach I used to cross join status and slots (your equivalent of category and value) might help. If not, please expand on your comment. EB's code might also be of help.

-2

What you look at is a crosstab. Assuming your table is called "Fact_Table", write:

select * from crosstab('select id, category, value from Fact_Table')

Also see http://www.postgresql.org/docs/9.5/static/tablefunc.html if you look for other variants.

1
  • 'tablefunc' is an extension
    – Sahap Asci
    Apr 12, 2016 at 18:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.