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

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

1026

Dominion General Discussion / Re: How do you guys not get bored?

« on: May 11, 2013, 01:44:52 pm »

Casual or pro? I've found point swings on casual mode bigger than on pro mode. If pro mode:

I'm guessing this is because you play fewer casual games, so your rating uncertainty is larger.

1027

Dominion General Discussion / Re: How do you guys not get bored?

« on: May 08, 2013, 01:07:04 pm »

I played ooksoo yesterday.  It was our first game, I believe, despite this puzzling claim: http://forum.dominionstrategy.com/index.php?topic=8017.msg235659#msg235659.  I immediately regretted playing him and I'll avoid him in the future.

While it's hard to be morally outraged that someone would scam an online board game, it does irk me that he's pretending nothing ever happened.  No apology, no self-justification... the guy didn't even bother to change his Goko username.

Anyway, he played the early game well and snapped up some early Provinces and the Tournament prizes.  But then tried to piledrive Provinces, ignoring Colonies, Platinums, and Grand Markets, and ground to a halt after Province 5 or 6.

would have got highest on pro long time ago if plat/colony is available, actually this is how i see it
casual = plat/colony with shelter,

 

1028

Goko Dominion Online / Re: 6000 rating

« on: May 08, 2013, 04:03:24 am »

I was around 5.6k until I decide to get to the top,

I like this quote a lot. 

1029

Goko Dominion Online / Re: 6000 rating

« on: May 07, 2013, 03:17:02 pm »

And actually, if you look at what they are saying, they have to pervert their system to not allow WE of over 100%/less than 0% in some games, which is prima facie quite bad.

Yikes, that sounds awful.  Okay, I just found the Goko wiki entry and the Funsockets Q&A, but they don't really say anything useful about their system, except that it's a TrueSkill variant.  Can you link me to where they talk about perverting their system?

Okay, the 10% sure win but-then-change-that-based-on-how-big-a-rating-gap-there-is is really wanting a different curve. Which is fine, of course, though finding exactly which one is best is EXTREMELY tricky.

Ah.  That makes sense.  Ok, so there's no reason that Elo with a non-Gaussian WE has to have the bias I'm thinking about.

1030

Goko Dominion Online / Re: 6000 rating

« on: May 07, 2013, 05:44:10 am »

I don't think it's so easy.  There is no automatic win in Dominion, someone who has never played this game will not win 100% of the games against you when they open 5/2 vs 4/3 on a Mountebank/Chapel board.  They probably will even lose most of the games.

I agree.  That sort of formula couldn't apply for all ratings differences.  If a 10% adjustment were appropriate for a 1500-pt difference, then you'd need a smaller one for a 3000-pt difference.  But that's clumsy and inelegant, and I'm not seriously proposing it as a modification of Elo... it's just an example of how using Elo for Dominion might generate a bias in favor of the lower-rated player.

Personally, I find it impossible to break even against anyone more than 1500 points lower than me.  I've also jumped about 300 points since I changed my match criteria from 5000+ to 5500+.  But maybe that's just noise, or maybe it's the variance in the quality of my play that's higher than Goko's rating system expects.

How much variance a kingdom has depends on the board, but I don't really see a good way on how to get this into a model.  It also depends on the player skills, e.g. say Horse Traders/Duke is dominant on a board without other support, that's more or less a 50/50+first player advantage if both players know the strategy, but if not the board might be much more complicated.

Actually, I really like this idea.  I doubt you could map kingdoms to variances, but modifying Elo so that the variance is drawn from some random distribution might capture the key difference between a game like chess and a game like Dominion.  It feels like it might even be mathematically tractable.

I'm less confident about modifying the variance based on the players' skill, but maybe that could work too.

1031

Goko Dominion Online / Re: Goko log parser - now for Chrome!

« on: May 07, 2013, 03:57:39 am »

Possession freezes the whole game, not sure about Silk road.

Silk Road freezes the whole game too.

Setting aside the debugging of these particular issues, shouldn't the log viewer catch and surpress any errors it generates?  Or is this a difficult thing to do in JS?

Thanks so much for the extension.  It's an huge improvement over the horrible turd of a log viewer that Goko crapped out on the floor.

1032

Goko Dominion Online / Re: 6000 rating

« on: May 07, 2013, 03:48:27 am »

I'm no expert on the math (and I'd be grateful is someone who is would point me in the right direction), but...

It seems to me that ELO is inappropriate for games of chance.  ELO models win rates as
    P(player 1 wins) = P(x1>x2),
where x1 and x2 are Gaussians random variables with means r1 and r2 representing the ratings of the two players.  IMO, a more appropriate model for Dominion is something like
    P(player 1 wins) = 10% + 80%*P(x1>x2),
the same model, but with a 10% chance of automatic victory for each player.  That 10% represents the odds of opening Mountebank-Chapel, or getting all the prizes in a tourney game, or winning the Duchy split in a Duke game, etc.

In my experience, the ELO/Goko expectation that a 2000-pt difference match go 5-1 in favor of the higher rated player just isn't realistic.  The weaker player wins much more often than the rating system expects.  Consequently, weak players improve their ratings by playing against stronger players and strong players benefit by limiting their opponents to other strong players.

If this assessment is correct, then opponent selection may have a major effect on Goko ratings.  A 5500 player who plays mostly with 4000-ish players might actually be just a skilled as a 6000 player only plays against the top 25.

1033

Dominion General Discussion / Re: Homage to the Best Card

« on: May 06, 2013, 11:28:39 am »

wait, that sounds like it would work! why is it a trap?

Too slow. Sometimes I got no Golem in hand, sometimes I got Golem and either Coppersmith or Counting House getting a terminal conflict. Took 19 turns to get 4 Provinces.

Yeah, if it's a Province game, I'm positive that you would be better off replacing the Coppersmith with a Copper... though I believe that's been shown to still be too slow for most boards.

I think the problem with these setups is that if you don't get to $8, there's nothing you can buy to improve your deck.  You non-Golem hands are aren't just weak... they're very nearly wasted.

IMO, an deck like this desperately wants a sifting +action, both to draw the Golems more frequently and to play the terminals that collide with them.  I recently stumbled into a similar setup: 1 Golem, 2 Scheme, 2 Apothecary, and 1 Coppersmith.  To my surprise, that was good for something like 7 provinces in 15 turns despite an unopposed enemy Sea Hag.

I definitely would have added an Arrested Development if it had been in the kingdom.

1034

Dominion General Discussion / Re: The WW Rating System

« on: May 03, 2013, 12:08:15 pm »

WW, without giving away your intellectual property, can you tell us how your algorithm compares to other Bayesian skill-rating systems (e.g. TrueSkill), either theoretically or in simulated competitions?

Short answer, no.
Long answer, how can I do this? I mean, what do you mean compares, exactly? And what data am I going to test it on - simulated competitions really don't make any sense, and it's not like there's a bunch of real data sitting around. Or maybe there is, but neither have I gone through it, nor have I gone through a full implementation of such other systems myself. And what are you measuring by?

I don't think your questions have obvious answers, but you'll want to address them if you're seriously thinking about patenting or publishing your algorithm.  It's probably worth looking at the approaches used to test TrueSkill and the Glicko system.

TrueSkill: http://research.microsoft.com/pubs/67956/NIPS2006_0688.pdf
Glicko: http://tennis-skill-rankings.googlecode.com/hg-history/c977c53a3af2913e780e39666fe1a272cc298319/links/glicko.pdf

Herbrich & Graepel compare TrueSkill to Elo on predictive performance, match quality, win probability, and convergence rates.  They're a bit imprecise about how they calculate their metrics, but their basic criteria are:

• Ratings differences should be a reliable way of predicting the outcome of a match.
• A match between similarly rated players should be (relatively) likely to result in a draw.
• Each player's long-run record against similarly rated players should be close to 50%.
• Ratings should converge quickly

They work with experimental data from the beta testing of Halo 2, which is team-based so not appropriate here.  I'll bet you could get your hands on a large database of online chess matches though.

Glickman uses simulated data where outcomes are determined by a random draw of performance p_i from a normal distribution N(s_i,b), where players i=1,2 have individual skill levels s_i and shared performance variance b.  Actually, that's the algorithm that Elo is designed for, but Glickman modifies is by assuming that skill levels vary over time, following a Bayesian random walk with standard deviation v.  He evaluates the Glicko based on the average deviation between it's estimates of s_i and v and their actual values.

There's a fair amount of other academic work on the topic, but these seem to be the best-received (or at least most cited).  Their tests aren't gospel, but they're a reasonable place to start.

1035

Dominion General Discussion / Re: How do you guys not get bored?

« on: May 03, 2013, 05:54:55 am »

I downloaded all of ooksoo's logs since the beginning of April, and eh, just looking through his games from yesterday, I'm not really surprised that he gets bored...

http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415512632.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415578711.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415669568.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415731649.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415787994.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415837952.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415898949.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367415952566.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367416000173.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367416052424.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367416111593.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367438710376.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367438772136.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367438826946.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367438880355.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367442288615.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367456252309.txt
http://dom.retrobox.eu/?/20130501/log.5114872ee4b0a7fa8c3a9919.1367456324702.txt

Hattip to Mischief for bringing this up in the first place and to lespeutere for giving me an ooksoo log to extract the hash.

Oho!  And here I thought the accusations were just speculation!  SheCantSayNo's post should be a sticky at the top of the thread.  And for anyone else who somehow missed the punchline, THEY SHOW OOKSOO USING GUEST ACCOUNTS TO THROW GAMES TO HIMSELF!

I'm not 100% sure that Goko would act that quickly to boot him, I think there's a much simpler explanation: he probably changed his name. Anyone want to check the hash of his account?

His hash is 5114872ee4b0a7fa8c3a9919.  That last game played by that hash was ooksoo himself on May 1 (against a guest, who resigned immediately).

I find it entirely plausible that the geniuses at Goko would generate those hashes dynamically from your username, and that changing your username would therefore change your hash.  On the other hand, if Ooksoo had changed his name there would be a new player near the top of the leaderboard, and I don't see one.  Particularly on the casual board... he was over 2000 pts above the legit players.  PityTheFool (who suddenly looks awfully suspicious) was there before.

I guess someone should just ask the Goko reps directly: https://getsatisfaction.com/goko/topics/ooksoo_is_cheating_the_leaderboard

1036

Dominion General Discussion / Re: How do you guys not get bored?

« on: May 03, 2013, 04:47:56 am »

All we know is that his name isn't anywhere near the top of the leaderboard anymore, which could mean alot of things: they reset his rating, they subtracted a large amount of points, they banned him from using the leaderboard, they banned his account altogether, or they restricted his account use to unrated games only (which would be hilarious and the most appropriate punishment).

Maybe he deleted his own account?  I think he had pretty much restricted himself to unrated already.

1037

Dominion General Discussion / Re: The WW Rating System

« on: May 03, 2013, 04:24:29 am »

WW, without giving away your intellectual property, can you tell us how your algorithm compares to other Bayesian skill-rating systems (e.g. TrueSkill), either theoretically or in simulated competitions?

1038

Dominion General Discussion / Re: How do you guys not get bored?

« on: May 03, 2013, 01:02:51 am »

Ooksoo is nowhere to be found on any of the leaderboards. I'm positively surprised Goko took action this quick

Wait... what?  Since when does Goko delete the accounts of suspected cheaters?  Any indication that this is what actually happened?

1039

Dominion General Discussion / Re: Homage to the Best Card

« on: April 29, 2013, 02:46:50 pm »

See serakfalcon's post above mine.

Er... right.  And now I've followed up on ftl's CC response too.   

1040

Dominion General Discussion / Re: Homage to the Best Card

« on: April 29, 2013, 06:39:42 am »

http://www.quickmeme.com/meme/3u38zp/

I can't explain it in a solitaire game... did you have a little help from your opponent?