The K-factor, if it is the same across all games, makes no difference, barring rounding concerns if the K is too small. K value just means “how much does this game matter in relation to others”. If the K value is halved for all games and the starting rating is 1500, a player who was 1700 would be a 1600 and a player who is 1300 would be a 1400. Changing it wouldn’t make any difference with regard to how players are ranked or how much incentive there is to play or not play.
Glicko is good because it has a factor that concerns length between games. Basically, the idea is that if you don’t have a lot of games in recently, (or if the league just started) your rating will change more than if you’ve played a lot recently, This is nice because it makes peoples ratings settle at their ‘actual’ rating faster, and presumably, makes people who are highly ranked want to keep playing so that they don’t have a lot riding on each individual game. Unfortunately, it could have the opposite effect, where someone gets a really high rating, doesn’t play for a week, and feels that they can’t ever play again without risking a ton of points. Also, if you reach your actual rating quickly and like it, you might want to stop playing even faster than if you were using ELO. The best solution might be to use Glicko for 1-month leagues.
Anyway, the way I see it is that we have 3 main ways to run it:
(1) the same.
(2) using a different site than challongeboards that allows us to use a different rating system.
(3) using the open-source nature of challongeboards to make a league there using a different rating system that someone else implements, (no idea if someone has already done this somewhere or how it can be done if not).
The K-factor effectively says “how much information does one more game give”. An ELO rating completely refreshes in approximately 800/k games, so having a lower k-factor reduces variance. The fact that more games are needed shouldn’t be a problem since the top players are all pretty active anyway. It is not true that it would have no effect on rankings.
I agree that Glicko is better, but reducing the k-factor is a much easier adjustment to make for future leagues.
I was thinking that glicko would be so much better than elo because it’d let us punish the “parking the bus” pattern that ELO encouraged. We can just say the top cut will be the N highest rated players with RDs below 100 (or whatever RD is appropriate).
Got it, that makes sense then. Now we have to figure out how to actually do it. The thing is, challengeboards works quite well for what we need it to do and everyone is used to it. Can anyone build Glicko into it or find another place that we can use that supports Glicko?
Yeah - like I said, I have a guy. Should be able to meet up with him tomorrow, give me a week to be on the safe side. When exactly would we want to launch the next league? Would “in 5-6 weeks” be ok?
In terms of functionality, I’m guessing rudimentary like on challongeboards, with possible gradual implementation of Netrunner-specific stuff?
I don’t think it’s a good idea to hold off on SHL3 to wait for some dude to finish a project. Such projects have a tendency to take a long time to finish. Maybe SHL3 runs with challongeboards and then hopefully this new thing is good to go for SHL4?
Yeah, people seem to be much happier OCTGNing when there is a league going. If he can get it done in 5-6 weeks for sure, we can potentially launch SHL3 as a one-month league. It seemed like a few of the top players didn’t play very much league in the second month anyway, (though it did give a few others time to get more games in).
If it sounds like something that might take longer, we can just do same old thing it and do another 2 month on challengeboards with ELO.
Oh yeah, also, I vote for 6 weeks. Both 2 month leagues felt like they were dragging on towards the end, but with just 1 month I’d feel very rushed to get games in, no matter the rating system.
I have played both in 2 and 1 month leagues. 1 month feels like a tourney. You have 2 decks, do not change them a lot and hope for the best. In this league I started horribly, tuned the decks in the process and actually played other ids sometime to fight the meta.
For me two months was very nice. For others maybe not. Maybe 6 weeks is the sweet spot and we cover an additional data pack release
I would love to play another league - this time from the very beginning. And since I see the wisdom in the suggestion that every participant should be registered here, I just registered an account.
6 weeks sounds good.
Thanks to all people who spend time to organize this!
One thing I would advocate is a ‘SH ranked game’ ongoing between leagues, one that doesn’t reset. Just having a number that goes up if you get better compared to your peers would interest me. For this system, ELO is probably fine.
[quote=“groober, post:2, topic:2804, full:true”]
1: Challenge board is open source. If someone codes up something else, that could help.[/quote]
Just a note here. It’s not enough to just come up with something and code it. That won’t change the website that is there. You need to then host it elsewhere and have your own database, server, etc…
True, a year is the maximum length challenge board. I didn’t realise that. I’ve sent the challengeboards.net guy a message on BGG to see if he would change that restriction to have a perpetual board.
I like playing complete games, but agree that a separate ranking number for corp and runner would be interesting to have. I am very aware that my corp play is much stronger than my runner play. I don’t think this is just because corps are better than runner at the moment.
I’ve talked with my guy, and bad news: he’s changing jobs, so there’s virtually no free time to be expected until, say, April or so. Guess a different solution is in order for now.
Working on someone else’s project is always a drag and hard to do. I’m diddling with the algorithm based on what’s on wiki, but it’s particularly complex. No promises of anything ever, but its an interesting puzzle. If I ever get it right I’ll post up the javascript. Someone would still need to hook it up to something like a user interface & database if I got that far (big if). I will definitely never care enough (or have enough time) to run a website for the whole thing or maintain a website.
And at that point you just need to figure out how and where to take advantage of it. You’re going to need to find a person willing to host it regardless.
There’s quite a few implementations out there, some are rather buggy and give ratings that are completely off, though. I remember going through all of them back when I was starting our league, and eventually settling on Java Ranking System, because it’s both a reference implementation and actually rather tidy. Others (there was a js version and a node.js version I was testing, at the very least) gave results that were plain out wrong.
When the time comes to test, I have about seven league seasons’ worth of data we can verify an algorithm against, to make sure it’s working correctly.