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