You are viewing a single comment's thread from:
RE: Top posts for July - rank calculated by minnow 1, dolphin 2, whale 3, megawhale 4
Its very possible im just an idiot, but i did not get what you were doing earlier.
For reference, in case there are others as thicks as I am , @jl777 is doing this:
(minnow likes1) + (dolphin likes2) +(whale likes3 )+(MW likes4)
^ which is to say (minnow likes TIMES 1)+ (dolphin likes TIMES 2) etc etc....
Which is what i said in the post above, but the UI thinks the asterisk that i used as a times sign is markup.
yes, the actual code is:
for (i=0; i<=STEEMIT_MEGAWHALE; i++)
{
metric += post->tally[i] * i;
}
I made it so MINNOWS have index of 1, DOLPHINS 2, ...
there is still no getting around the whale's influence. Maybe what I need to do is DIVIDE the score by the whale count? that would undo (or more than undo) the effect of whale created visibility.
Clearly once a post gets a whale vote its visibility is enhanced, so finding the elapsed time of the first whale/megawhale vote and then diluting the votes that come after it by the number of then current whalevotes. would be a lot more calculation intensive, but it would factor out the whale visibility effect, maybe even too much.
really what is needed is the estimated total unique pageviews as that is the best metric and it works with or without whale created visibility
A couple math points...
I feel like for a model like this, the exponential scale should be constant.
So for example, if you start at 100KV your cohorts should be 100KV times 10^1 times 10^2 times 10^3. Im not sure if there was a reason to give dolphins double the exponential range. Especially since the $ value of their vests is in a point in the curve where most people would argue there is a very high marginal dollar value.
In the same vein, if youre going to use exponential cohorts, you ought to use exponential multipliers, they don't have to be 10 (obviously since that wouldnt tell you anything) but they should be exponential.
That said... but you could probably take the total number of Kv then exponentially transform the number to normalize the disparity in voting power. Not sure how easy this would be (and certainly don't expect you to do it) but it would give you
I feel like theres a name and a formula for this that i just cant think of..... but im talking like this
where Z is a level of correction, with z=10 as no correction (linear relationship between vests and voting power), and Z=1 as one man one vote.
As for correcting for the sticky effect and schooling... Schooling is probably impossible to correct for... the sticky effect.... the easiest way to correct, if you had TPTB in on it, would be just to change the default view to "new" for a week or so and see how it effected minnow vote distribution on high paid posts.
Failing that, i think there might be a way to do it based on the info on the blockchain... Voters get credit on the curation reward based on how quickly they vote... so a post with early whale voters and minnow followers due to pageviews is going to have lower minnow curation rewards, where a post that has gets minnow and dolphin groundswell which attracts whales is going to have a porportion of high-reward minnow curation rewards
But im not 100% on how curation rewards are handled, so i don't know how good they would be to mitigate the exposure effect.
there is also a linear penalty for early voting. 30 minutes is full credit, 15 minutes is half credit, etc.
Keep in mind I just parsed and put this together to get something/anything working. Clearly mapping ranges to a specific integer will have edge effects. One way is to just use log10 of VESTS, or kiloVESTS, or sqrt, or cbrt, etc. It is just a matter of deciding how to map the different account levels.
Maybe cbrt of Mv would be interesting. anybody with less than 1 Mv will get increase wt, probably need a cap at 4 and also a floor to filter out the plankton to remove the botspam.
Since we are trying to correct for what is a large dynamic range, I dont think we need to follow "normal" rules