双精度Postgres模糊搜索

Question

I would like to fuzzy search on decimal numbers instead of strings. So the idea is searching for 100 should bring a range of 100, 90, 95, 105, 108, 120 number values from rows in database.

I have tried like keyword too but it doesnt work as i wants. How can i do a fuzzy search on decimals. thank you

Answer 1

Use between . The function is an example:

create or replace function fuzzy_match_numeric
    (number numeric, value numeric, deviation numeric)
returns boolean language sql as $$
    select number between value- value* deviation and value+ value* deviation
$$;

Check for matching the value 100 with deviation of 5%:

select
    fuzzy_match_numeric(94, 100, .05) r1,
    fuzzy_match_numeric(95, 100, .05) r2,
    fuzzy_match_numeric(105, 100, .05) r3,
    fuzzy_match_numeric(106, 100, .05) r4

 r1 | r2 | r3 | r4 
----+----+----+----
 f  | t  | t  | f
(1 row)     

Answer 2

I would suggest computing deviation for set of your lookup values and choosing the best candidate. Below is a sample based on integers, but numeric types will work analogously.

Sample data set: search_table

postgres=# select * from search_table order by 1;
 value
-------
    90
    95
   100
   101
   103
   105
   108
   120

Sample lookup values set: search_condition

postgres=# select * from search_condition order by 1;
 value
-------
   100
   103
   105

Look for best candidate:

select 
  distinct on (value) 
  value, 
  lookup_value as best_candidate
from ( 
  select 
    st.value, 
    sc.value as lookup_value, 
    abs(1 - st.value*1.0/sc.value) as deviation 
  from search_table st 
  cross join search_condition sc 
  ) t 
order by value, deviation, best_candidate;

Result:

 value | best_candidate
-------+----------------
    90 |            100
    95 |            100
   100 |            100
   101 |            100
   103 |            103
   105 |            105
   108 |            105
   120 |            105

In case of ties a lower candidate will be chosen. This can be amended by adding DESC to the best_candidate column within ORDER BY clause, to take the highest candidate.