Oddbean new post about login | logout And @ManiMe is polishing up his own independent implementation.
And @cloud fodder has implemented the GrapeRank algo in go https://github.com/Pretty-Good-Freedom-Tech/gvengine
@cloud fodder and @straycat, the number of iteration to reach convergence (~10^-6 tollerance) for pagerank is roughly 100. Why do you perform only 8 iterations?
My version of GrapeRank determines “convergence” without arbitrary iteration limits … by excluding “calculated” scorecards (whose scores have not changed beyond a “precision” threshold) from future iterations. https://github.com/Pretty-Good-Freedom-Tech/graperank-nodejs/blob/main/src/Calculator/index.ts#L255-L261 Number of iterations varies with “depth” of follows calculated (and other parameters) but usually converges around 10 or 20 .
We just picked 8 as a place to start. Still needs work. My most recent implementation is at grapevine-brainstorm.vercel.app (link to repo in the footer) and I set it to iterate until convergence, defined as sum over (delta-score)^2 is less than some threshold, and I think around 12 or 15 iterations is where it stopped. @ManiMe uses a similar method.