1

Cascade rounding is an algorithm to round an array of floats while preserving their sum. How does one implements this algorithm in PostgreSQL?

ido klein
  • 137
  • 2
  • 11

1 Answers1

2

You can implement this function in plpgsql:

create or replace function cascade_rounding(float[])
returns int[] immutable language plpgsql as $$
declare
    fp_total float = 0;
    int_total int = 0;
    fp_value float;
    int_value int;
    result int[];
begin
    foreach fp_value in array $1 loop
        int_value := round(fp_value + fp_total) - int_total;
        fp_total := fp_total + fp_value;
        int_total := int_total + int_value;
        result := result || int_value;
    end loop;
    return result;
end $$;

select cascade_rounding(array[1.1, 1.2, 1.4, 1.2, 1.3, 1.4, 1.4])

 cascade_rounding
------------------
 {1,1,2,1,1,2,1}
(1 row) 

Try the function in Db<>fiddle.

Update. You can apply the function to a column. Exemplary table:

create table my_table(id serial primary key, float_number float);
insert into my_table (float_number) 
select unnest(array[1.1, 1.2, 1.4, 1.2, 1.3, 1.4, 1.4])

Query:

select 
    unnest(array_agg(id order by id)) as id,
    unnest(array_agg(float_number order by id)) as float_number,
    unnest(cascade_rounding(array_agg(float_number order by id))) as int_number
from my_table;

However, this is not a perfect solution. The query is quite complex and suboptimal.

In Postgres, you can create a custom aggregate with the intention of using it as a window function. It's not particularly difficult but requires some knowledge, see User-Defined Aggregates in the documentation.

create type cr_type as (int_value int, fp_total float, int_total int);

create or replace function cr_state(state cr_type, fp_value float)
returns cr_type language plpgsql as $$
begin
    state.int_value := round(fp_value + state.fp_total) - state.int_total;
    state.fp_total := state.fp_total + fp_value;
    state.int_total := state.int_total + state.int_value;
    return state;
end $$;

create or replace function cr_final(state cr_type)
returns int language plpgsql as $$
declare
begin
    return state.int_value;
end $$;

create aggregate cascade_rounding_window(float) (
    sfunc = cr_state,
    stype = cr_type,
    finalfunc = cr_final,
    initcond = '(0, 0, 0)'
);

Use the aggregate as a window function:

select 
    id, 
    float_number, 
    cascade_rounding_window(float_number) over (order by id) as int_number
from my_table;

 id | float_number | int_number
----+--------------+------------
  1 |          1.1 |          1
  2 |          1.2 |          1
  3 |          1.4 |          2
  4 |          1.2 |          1
  5 |          1.3 |          1
  6 |          1.4 |          2
  7 |          1.4 |          1
(7 rows)    

Db<>fiddle.

klin
  • 112,967
  • 15
  • 204
  • 232
  • thanks, however - I would like to implement this on a column, not on an array. I've searched a lot on how to do this, and haven't found a solution. running this solution on a mock table returned an error: ERROR: column "A" does not exist LINE 1: SELECT "A", although this command was never passed. – ido klein Apr 25 '20 at 18:12
  • 1
    The question was too broad. It should contain at least data (table) definition, sample data, and expected result. Please be more precise in future. – klin Apr 25 '20 at 19:44