Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[WanderingWinder]

... is terrible. For one thing, I just won a game against the number 12 guy on the list, and gained a whopping 0 points. But ok, this appears to be a bug - it's not normally THAT bad. But there are a couple things. One, the rating it shows you is some form of rating - 2* uncertainty. Well, I don't like that so much in general - I didn't like it abut iso - and it basically strikes me as a gimmick to get people to play more. But the bigger issue I see with it is I will play someone who is rated, I don't know 5000, and I will 'gain 8 points'. Well, at least some of that is just my uncertainty lowering. Indeed, it seems to me that it wouldn't take someone much lower than that for my rating (i.e. the actual skill bit, not the number they show you) to actually *decrease* with game play. Maybe this is already happening there, it's hard to tell, since they only give you the number that they do. And while there is a big problem with the system in that people with the same rating can have a wide variety of skills, well, even someone who is pretty bad is going to have more than a 1% chance, just because of the inherent randomness. Indeed, it seems to me that the expectation function might actually be *linear*, which would be truly terrible, and that certainly there comes a point where the expectation is 100%, and after that, you are actually expected to win *more* games than you are playing, which is of course absurd.

So the other thing is that people even in the 5000s and 6000s seem to be, well, pretty bad. That might not be a problem with the ratings, so much as a weak player pool, but on the other hand, the system stops the few actually pretty good players there from rising above them significantly. I mean, there is a competitive aspect to ratings, but one of their main features is to actually get you paired up against people of reasonably similar skill levels, which should make the game more fun. And this system is failing at that.

So, question for someone who knows, or who has any kind of experience, how does the system deal with multiplayer? Because I've only played 1v1s.

[Polk5440]

So the other thing is that people even in the 5000s and 6000s seem to be, well, pretty bad. That might not be a problem with the ratings, so much as a weak player pool, but on the other hand, the system stops the few actually pretty good players there from rising above them significantly.

I think part of the problem is that a LOT of players are new or "provisional" (platform is new and new people keep joining). Because there is a lot of variation in skill in provisional players and so many people are provisional, the rating system isn't sorting as well as Iso, yet. I do not get the impression it's a fault of the actual rating system, though.

I was about 35 on Iso before leaving and have been hovering around 6000 or so on Goko (I like to think I've improved a bit since leaving Iso), and I am not finding an unreasonable skill disparity with people similarly ranked (say +/- 500 points) if they have 100+ games under their belt.

@WW: I saw you on Thursday waiting (and waiting) for people 6000+ to play against. Sorry I couldn't join, but I was playing against some friends; hopefully, you eventually found some good matches. But based on this post, maybe you didn't? 

[Watno]

You can loose insane amounts of points losing a multiplayer game.

[Schlippy]

You can loose insane amounts of points losing a multiplayer game.

As a matter of fact, you can also gain quite a few points losing a multiplayer game.

[jsh357]

I find the system very disheartening just on a visceral level.  sometimes I beat players I know are good and gain nothing, but losing any game drops me near or over a hundred points.  really makes you angry when you drop like that sometimes if you're at all competitive.

[Schlippy]

For me the most disheartening thing about the rating system is that when I play 10 games against my brothers, win 8 of them, lose 1 of them and tie in the last one I lose more points than I win and drop 10-80 ranks.

[michaeljb]

Back in January, I got up to the #1 spot for a while after around 30 games I think.

The rating system is definitely frustrating, even more so because of one question we've asked each of the Goko's Q&A things they've done here--"what is the actual formula used to determine the ratings?" And we never got an actual formula answer (that I'm aware of anyway).

[Rabid]

Back in January, I got up to the #1 spot for a while after around 30 games I think.

The rating system is definitely frustrating, even more so because of one question we've asked each of the Goko's Q&A things they've done here--"what is the actual formula used to determine the ratings?" And we never got an actual formula answer (that I'm aware of anyway).

Yep after trying several times to get an answer to this the best we got was:
A bit like ELO, followed by a description of what I guess to a TrueSkill variant.

[rrenaud]

Goko's ranking system sucking is a good reason to just go and make your own.

[DG]

As far as I can tell, when you win a multiplayer game you get 75-90% of the rating change you might from beating your opponents in a series of two player games. For me the worst part about the ratings is losing quite a lot of points for a tied game against a lower ranked player. So if you come joint first in a four player game you could still lose rating overall even though that's a good result in most circumstances.

In beta I tried to monitor the rating changes and ultimately had no confidence that they were calculated correctly, or at least be consistent to a player's eyes. It would probably need an exclusive series of matches between a few players who were recording their ratings to properly give me confidence in the ratings system.

[hsiale]

Goko's ranking system sucking is a good reason to just go and make your own.

That's a good idea. To make it possible, I think they need to add to the log:
- info on if a game is professional play, casual, unranked or adventure,
- in case of pro games, sets used to generate the kingdom.

Anything else I missed?

It would probably need an exclusive series of matches between a few players who were recording their ratings to properly give me confidence in the ratings system.

I'm playing Lord Bottington in pro play mode quite often. What data should I log next time I'll be playing quite many games in a row?

[Watno]

The problem is that it still won't affect matchmaking (assuming it's implemented at some point)

[heron]

I just played two games against Defender Bot and won both, but the second time my rating went up twice as much. That makes no sense at all.

[Mic Qsenoch]

I just played two games against Defender Bot and won both, but the second time my rating went up twice as much. That makes no sense at all.

Well, that does seem strange, but it might be explained by a whole bunch of other people losing to that Bot causing its rating to increase between your two games.

[heron]

Its rating seemed to remain the same, 3945 or something.

[WanderingWinder]

So, I'm pretty sure that a straight implementation of the Elo system would do better. Actually, I have most of the math for my own system already (there are a couple of parameters that need empirical fitting; well, one really, but then you have to deal with provisional-ness, which is something I haven't looked into too much), but there are a couple issues. One is getting the data, because you need all of it, and the bots are going to potentially complicate it by playing multiple games simultaneously. I guess you could ignore them. But definitely you also only want 'pro' games. The other issue is, even if I can keep my own ratings... so what? Unless I publish them all in some place that everyone sees, it's not going to help most of the problems - the only thing this would do is tell the few people who have access it's the best. Mostly this is just me, though probably I could find someone (the CR people) to host the data in a  place where we see it, if they can work out getting the data. Anyways, the vast majority of people are just going to see the official thing, and so there's very little upside. Not to mention there may be some issues with the site allowing things, Id have to check the ToS. Though probably if I asked, they wouldn't care - they're pretty chill on such things.

[Kirian]

Goko's ranking system sucking is a good reason to just go and make your own.

<shameless plug>This is what tournaments are for, of course.</shameless plug>

I wonder if there would be some way to create a "rolling" tournament.

[philosophyguy]

So, it should be possible to write an extension that extends the click-on-player behavior to get the rating from [whoever is hosting WW's algorithm] in addition to the ratings already there. I wasn't able to dig through the JS enough to determine if you could change so that the WW-rating is displayed by default in the player list, but I would imagine so.

[hsiale]

I wonder if there would be some way to create a "rolling" tournament.

You mean a tournament going on  indefinitely with past rounds getting gradually less important? I think it's definitely possible. Play Swiss, I guess one round per week is realistic, latest round counts 100%, one before that 90% and so on, you cannot play any of your let's say 8 latest opponents, once every 2 months (or twice every half year, or some other solution that lets people have holiday but still makes them play regularly) you can take a week off freezing your ranking. Players joining need some provisional ranking for pairing reasons.

[SCSN]

So, I'm pretty sure that a straight implementation of the Elo system would do better. Actually, I have most of the math for my own system already (there are a couple of parameters that need empirical fitting; well, one really, but then you have to deal with provisional-ness, which is something I haven't looked into too much), but there are a couple issues. One is getting the data, because you need all of it, and the bots are going to potentially complicate it by playing multiple games simultaneously. I guess you could ignore them. But definitely you also only want 'pro' games. The other issue is, even if I can keep my own ratings... so what? Unless I publish them all in some place that everyone sees, it's not going to help most of the problems - the only thing this would do is tell the few people who have access it's the best. Mostly this is just me, though probably I could find someone (the CR people) to host the data in a  place where we see it, if they can work out getting the data. Anyways, the vast majority of people are just going to see the official thing, and so there's very little upside. Not to mention there may be some issues with the site allowing things, Id have to check the ToS. Though probably if I asked, they wouldn't care - they're pretty chill on such things.

What about contacting Goko with your ideas so that yours might become the official rating system?

[blueblimp]

So, I'm pretty sure that a straight implementation of the Elo system would do better. Actually, I have most of the math for my own system already (there are a couple of parameters that need empirical fitting; well, one really, but then you have to deal with provisional-ness, which is something I haven't looked into too much), but there are a couple issues.

Elo sucks, because it doesn't model rating uncertainty, so it needs kludges like provisional ratings. Why would you want to use Elo instead of TrueSkill / Glicko2 / something else non-awful. Isotropic used TrueSkill, and although that is (unfortunately) patented so Goko probably can't use it, Glicko2 works out pretty similarly for 2-player matches.

[Morgrim7]

Ugh. Ive given up Goko

[noey21]

I have given up also.  How do I not show provisional?  I have been disconnected like 3 times and you can't go back in.  No one will play anymore.  NO one chats.  I am glad i only put in minimal time.  I just need to find more players IRL so glad wife and family plays.  Too bad I am on other side of the world with my family.

Hard to teach new players when you have played over a few thousand games.  Unfortunately I get booted on here every game because I am provisiional and being 6-9 hours ahead of the US less player pool.

I loved you dominion.

[hsiale]

You can get a non-provisional Pro rating playing Bots in Pro mode. I think something like 20-30 games should easily be enough. My Pro rating is above 5000 now and I don't think I played more than 10 Pro games against a human opponent.

[Kirian]

Yep... spend 5 games beating up on bots, you'll have a non-provisional rating.  Most people seem to be in the 3500 and up range, sp a 3000 rating should get you games.

Or host games yourself...

[WanderingWinder]

Goko's ranking system sucking is a good reason to just go and make your own.

http://forum.dominionstrategy.com/index.php?topic=7912.0

[Fabian]

Gaining 7-25 points when you win and losing 80-110 points when you lose is really soul crushing. This rating system is confusing me That is all.

[SCSN]

Has anyone actually played Ooksoo? All I ever see him doing is hosting unrated(!?) games while being afk.

[math]

I have played him before.  He seems to be as good as his ratings say he is.

[philosophyguy]

I entered an open seat in Ooksoo's game a couple days ago but he or she never started the game. After waiting for several minutes, I just left on my own.

[Stealth Tomato]

Has anyone actually played Ooksoo? All I ever see him doing is hosting unrated(!?) games while being afk.

His girlfriend plays unrated games on his account because he has all the promos and she doesn't. She's nice, fairly good player as well (although doesn't appear to be nearly as good as him). Not sure about the AFK part.

[snorkelbike]

I played Ooksoo once. I thought I was crushing him badly and on his final turn he pulled off one of the most massive, impressive combos with develop/fortress I've ever witnessed to beat me by a couple points. He thought and thought that final turn, and I ended up very impressed.

[Fabian]

I played a 4 player game. At some point right before the game was supposed to end, one of the players quit. The game immediately ended, no rating change happened.

Is this the way it's intended to work?

[math]

That is definitely NOT how it is supposed to work, but that is what happens on Goko.  This is another thing they need to fix - one person quitting should not end the game.  The rating bug is unfixable unless they do that first.

[Fabian]

This only works in multiplayer I assume? Like, I remember reading that at some point you could just quit in a 2p game and the rating wouldn't change, but I was under the impression that was fixed? But only for 2p?

I could start playing multiplayer and if I'm losing just close the browser and not lose points? Am I understanding this correctly?

[math]

You still lose points, but no one else gains points (because it ends it early, so it's not certain who would have won).  In 2p, you lose points and your opponent gains.

[Fabian]

I see, thanks.

[qmech]

I had an opponent crash out of a 3p game earlier today and gained 4 (casual) points.

[SCSN]

Gaining 7-25 points when you win and losing 80-110 points when you lose is really soul crushing. This rating system is confusing me That is all.

Losing 22 points for tying someone from seat 2 is also pretty discouraging

[dudeabides]

I don't know why all of you are complaining about the rating system; it is clearly awesome:




I've always known that I was better than Lespeutere 

[GendoIkari]

I've always known that I was better than Lespeutere 

Yeah, well, that's just like, your opinion, man.

[lespeutere]

Who's this LESPEUTERE guy anyway? I heard he was no. 1 on iso but then how can he hover around 25 on goko?

[dudeabides]

Who's this LESPEUTERE guy anyway? I heard he was no. 1 on iso but then how can he hover around 25 on goko?

Iso's rating system must have been flawed.  At one point, it even had some clown named -Stef- atop its leaderboard.

[WanderingWinder]

Who's this LESPEUTERE guy anyway? I heard he was no. 1 on iso but then how can he hover around 25 on goko?

Iso's rating system must have been flawed.  At one point, it even had some clown named -Stef- atop its leaderboard.

I do believe that Iso's rating system was flawed. However, it took a couple thousand games for me to feel very confident about that - which ain't all that bad.

[Jeebus]

Can someone explain to me how Goko's rating system works? I know that the details of the formula is hidden, but I'm asking about something a little more specific. When I log on after a while (say, the next day) my rating has usually (always?) changed since I logged off. So there is some sort of automatic adjustment happening. I guess that's the uncertainty going up, which makes the rating go down. However, the number in parenthesis is often positive at this point. I'm not sure, but it seems that I still have gone down though, not up. So what does that number in parenthesis mean?

To clarify, I know what it means after a game. I'm talking about after an automatic adjustment. And how often if that adjustment anyway, once a day?

[Stealth Tomato]

It looks like Goko compares second and third in 3p games, which is just wrong. So coming in 3rd counts as losing to two people, while coming in second counts as beating one and losing to one.

I found this out by losing a hundred and twenty-one points for coming 3rd in a 3p game using a boom-or-bust strategy.

That's just awful.

[Watno]

I wouldn't say it's wrong, I'd say it's debatable.

[Kirian]

That's what Trueskill does for multiplayer matches too, to a first approximation.  You can certainly debate whether that's a perfect or even good system, but it's not uncommon.

[WanderingWinder]

It's a debate we've had before, and one without a clear answer. Do you want to incentivize going for the win, or trying to place as high as possible? You can make realistic arguments for all-to-nothing, totally even spread (first to second is the same as the drop from second to third, etc. etc. for as many players as there are), and virtually anything in between.

But however you want to do it, it's a choice not so much at the rating level as the game level - what is the goal? (Of course, once this is clear, the rating system should be implemented to match - the two things are inherently interconnected, but the one (rating) should be in service of the other).

[Stealth Tomato]

That's what Trueskill does for multiplayer matches too, to a first approximation.  You can certainly debate whether that's a perfect or even good system, but it's not uncommon.

Wasn't Iso's implementation set to compare the losers of multiplayer matches only to the winner? So in a 3p match, #1 beats both #2 and #3, #2 loses to #1 and ties #3, and #3 loses to #1 and ties #2?

[Kirian]

http://atom.research.microsoft.com/trueskill/rankcalculator.aspx

This is approximately what Iso used.

I tried a few things out quickly:  Winning a 3P game on their calculator gives about 1.5x what you get from winning a 2P game, and winning a 4P game gives about 2x... there are diminishing returns for having more players, which makes sense to me.