implement standard inference

[mastercyb]

We discussed for a while one property of the ranking mechanism which rise huge concerns into the answering capabilities of cyber: lets call it true-false problem

true-false problem

• let us imagine that true particle have cyberank 10, and false particle have cyberank 9
• the environment allow to link any particle with any
• that means that for any questions which cyberlinked to true and false the winning answer will always be true
• of course such behavior does not feels like something superintelligent
• we always emphasize that cyberank is a core ranking factor, but not the only one which is possible on top the cyber architecture
• in order to solve true-false problem we have to compute weights of links in a context

in this post i introduce an algo i will call standard inference

• does not introduce additional complexity
• requires zero additional information
• follow occam razor principle, that is it is as dumb and simple as possible

standard inference

• get all particles cyberlinked with asking particle sorted by rank
• for each particle get all cyberlinks
  • for each cyberlink get address
  • for each address
    • get [[V]] balance
    • get number of cyberlinks
    • compute average [[V]] per cyberlink
  • sum (average V) votes for every particle
• multiplay rank of each particle by sum (average V)

cyber project without consensus around standard inference on top of cybergraph which solves true false problem seems very blurry to me

[SaveTheAles]

@Snedashkovsky

Check the query, there are probably some ways to improve or correct it

WITH filtered_particles AS (
    SELECT 
        multiIf (
            particleFrom != 'QmZPGMtHBaJwURxUrfXdDswhWZ9zL1mPvABcdxSf1dJz1R', particleFrom,
            particleTo != 'QmZPGMtHBaJwURxUrfXdDswhWZ9zL1mPvABcdxSf1dJz1R', particleTo,
            ''
        ) AS particle FROM spacebox.cyberlinks
    WHERE particleFrom = 'QmZPGMtHBaJwURxUrfXdDswhWZ9zL1mPvABcdxSf1dJz1R' OR particleTo = 'QmZPGMtHBaJwURxUrfXdDswhWZ9zL1mPvABcdxSf1dJz1R'
), linked_cyberlinks AS (
    SELECT 
        particle,
        particleFrom,
        particleTo,
        neuron
    FROM filtered_particles
    LEFT JOIN 
        spacebox.cyberlinks AS c
    ON 
        filtered_particles.particle = c.particleFrom OR filtered_particles.particle = c.particleTo
), account_debs AS (
    SELECT * FROM spacebox.debs_and_creds
), sum_with_filter as (
    SELECT 
            address,
            height,
            amount,
            sum(amount) over (partition by address, denom order by height) as saldo, 
            denom
    FROM
        account_debs
    WHERE denom = 'millivolt'
), ranked_saldo AS (
    SELECT
        address,
        denom,
        saldo,
        height,
        ROW_NUMBER() OVER (PARTITION BY address, denom
    ORDER BY
        height DESC) AS rn
    FROM sum_with_filter
), filtered_saldo AS (
    SELECT
        address,
        height,
        saldo,
        denom
    FROM
        ranked_saldo
    WHERE
        rn = 1 and saldo != 0
), joined_balances AS (
    SELECT 
        particle,
        particleFrom,
        particleTo,
        neuron,
        saldo as balance
    FROM linked_cyberlinks
    LEFT JOIN filtered_saldo ON linked_cyberlinks.neuron = filtered_saldo.address
), average_vote AS (
    SELECT 
        particle,
        particleFrom,
        particleTo,
        neuron,
        balance / 1000 as balance,
        links,
        balance / links as avg_vote
    FROM joined_balances
    LEFT JOIN (SELECT neuron, count(neuron) as links FROM spacebox.cyberlinks GROUP by neuron) as t ON joined_balances.neuron = t.neuron
), united_particles AS (
    SELECT particleFrom as particle, avg_vote
    FROM average_vote
    UNION ALL
    SELECT particleTo as particle, avg_vote
    FROM average_vote
) 
SELECT particle, sum(avg_vote) as sum_avg_vote 
FROM united_particles
GROUP BY particle
ORDER BY sum_avg_vote DESC

cyberlinks and debs_and_creds tables are required

[Snedashkovsky]

@SaveTheAles LGTM

[mastercyb]

what is LGTM?

[happylolonly]

API not deployed for now