Ben gets full points for winning but X3 only gets four points… The points of both matches should be taken into account for both players instead of only the last one. Ben get more reward out of the tie, were X3 had the lower rating and deserved more points.
Oh yeah… I thought the nonsymmetric elo change was suspicious but I didn’t notice @x3r0h0ur’s wins were reported as ties. They should be wins of course… and there’s another such tie still pending.
If they can’t be changed retroactively, maybe @x3r0h0ur should just go ahead and report something like… 2 more wins and another tie against me? To approximate what the elo change of 3 wins should have been?
This time around I’m playing janky decks and not at all particular about my ELO so I’m fine with whatever solution.
I did not change anything that could impact official rating calculations (approved matches) or match results (pending matches). All I did was replaying the backlog on top of the standings ad-hoc in memory without any database commit for the purposes of “unofficial” button. If I dropped the ball somewhere it would be reflected only in messed up unofficial standings.
In @bblum’s history of resolved matches I see two ties, which both look fine for me:
There are laso two ties in the backlog, also seem fine:
I think that @x3r0h0ur simply reported ties by mistake?
[quote=“rankade, post:69, topic:2804”]
As you said, rankade algorithm is now not in the public domain (like Glicko was for years, until open-source release).
As for Glicko (but in a deeper way), rankade’s ree is an Elo algorithm improvement, with major differences (even if less evaluable in a plain 1-vs-1 structure like the one in your Android: Netrunner matches), a completely different way to manage playing frequency dynamics, and many features that you don’t find elsewhere.
Some of these may be useful in your league (playing frequency dynamics, different weight for matches, group characteristic modelization, late users dynamic in groups, direct players match adding, admin editing and deleting powers, graphs, reports, shareability, and more) while some others, that may be interesting in other boardgames, maybe do not for Netrunners (multiplayer, multifactions, cooperative matches, asymmetrical factions, detailed result with specific ties, partial rankings, belts, and more). [/quote]
If interested, here you can find a comparison at a glance between some ranking systems (Elo, Glicko, TrueSkill, Ree). As we said in previous post, not every rankade ‘plus’ feature is useful for your league, but some of these (see quoted text) may be.
We remain at your disposal for any clarification, and a thank you in advance for the feedback that you possibly will let us have.
On the one hand, Rankade’s rating decay over time disincentivizes parking the bus, which I like.
While it also penalizes people who don’t get to play much – like me – people who don’t get to play much probably won’t be top 4/8 in the league anyway. Still, if that turns out to be a problem, we might need to do a shorter league with Rankade.
Given that there’s no real evidence anyone in the top parked their bus, seen that total games played for those players was quite high, I’m not quite sure if this is necessary. The majority of the people with few games are (understandably) sitting in the middle of the pack. Any activity “encouragement” is more likely to penalize people who come looking after they saw a SHL tag or those who don’t need a rating but just want some serious games now and then, in stead of people using negative strats in the top of the league (which gets a lot more talk than it actually happens). I think you are going to hit the bulk of players for little competative gain.
And if the ranking works nicely with a gradual slope for losses and gains depending on the skill of your opponent, even when we make a longer league than we have now (which didn’t exactly settle which is why a race for the top 4 cut was even possible), parking the bus still isn’t a functional strategy as your expected gains would just attempt to keep you at your “perfect elo”.
tl;dr. I don’t think we should worry too much about activity.
What do you plan on choosing for the constants? (I’m looking at the maths on the wikipedia article.) Just to hazard a guess maybe something like rating period = 1 day or 1 week, and c = sqrt((350^2 - 100^2) / (twice the length of the league))?
IOW, I’d expect maybe half the players to play frequently enough to get RD<100, and a player having to not play for two full leagues before we knew nothing about their rating anymore. But I can easily see myself being off by a factor of two either way.