Percentage Matching two JSONB columns, - sql

Question

I am trying to compare two JSONB columns in a table, at the moment it is done in the app, however this doesn't allow proper searching, filtering and ordering without loading the whole data set. It would be better if we could do this comparison in the DB.
The following is an example of the data and calculation.
employer = {
"autism": "1",
"social": "1",
"dementia": "0",
"domestic": "1",
}
employers_keys = ["autism","social","domestic"]
candidate = {
"autism": "0",
"social": "1",
"dementia": "0",
"domestic": "1",
}
candidate_keys = ["social","domestic"]
remainder_keys = employer_key - candidate_key = ["autism"]
1-(remainder_keys.length/employer_keys.length) = 1-(1/3) = 2/3 = 66%
This process is all rather trivial in Ruby, jsonb-> array -> select -> calculation
However, I would like to perform this in SQL or a function at the DB level, something like
function compare_json(employer, candidate) returning a decimal.
More specifically
Select candidates.id,
st_distance_sphere(st_makepoint(employer.long, employer.lat), st_makepoint(candidates.long, candidates.lat)) /
1000 / 8 * 5 as distance
from (select * from users where id = 8117) employer,
(select * from users where role_id = 5) candidates
where st_distance_sphere(st_makepoint(employer.long, employer.lat), st_makepoint(candidates.long, candidates.lat)) /
1000 / 8 * 5 < 25
order by distance
The above SQL calculates the distance between a single employer and multiple candidates, the inline queries employer.skills (1 row), candidate.skills (n rows).
So the output should be..
Candidate id, Distance, SkillsMatch(employer.skills, candidates.skills)
As before the edit, any guidance will be welcome.

Answer 1

Here is a pure SQL approach: it works by turning the employer object to a record set, and then performining conditional aggregation:
select 1 - avg( ((d.candidate ->> e.k)::int is distinct from 1)::int ) res
from (values(
'{ "autism": "1", "social": "1", "dementia": "0", "domestic": "1" }'::jsonb,
'{ "autism": "0", "social": "1", "dementia": "0", "domestic": "1" }'::jsonb
)) d(employer, candidate)
cross join lateral jsonb_each_text(d.employer) e(k, v)
where e.v::int = 1
You can easily turn to a function by replacing the literal objects in the values() row constructor with parameters.
Demo on DB Fiddle:
| res |
| ---------------------: |
| 0.66666666666666666667 |

Answer 2

Okay this is what I got to.
CREATE OR REPLACE FUNCTION JSON_COMPARE(employer_json jsonb, candidate_json jsonb, OUT _result numeric)
AS
$$
BEGIN
select 1 - avg(((d.candidate ->> e.k)::int is distinct from 1)::int)
into _result
from (values (employer_json, candidate_json)) d(employer, candidate)
cross join lateral jsonb_each_text(d.employer) e(k, v)
where e.v::int = 1;
RETURN;
END;
$$
LANGUAGE PLPGSQL;
Which is small variation on GMB super fast answer. With a few indexes and correctly limiting the size of the candidate list we get reasonable performance.
I'm new to Stack so my upvote for GMB doesn't show, but thanks again.