Dominion Strategy Forum

Please login or register.

Login with username, password and session length
Pages: 1 [2] 3 4 5  All

Author Topic: I don't know what this thread has turned into (was Making it to Level 42 -Lvl 5)  (Read 46135 times)

0 Members and 1 Guest are viewing this topic.

Thisisnotasmile

  • Saboteur
  • *****
  • Offline Offline
  • Posts: 1493
  • Respect: +676
    • View Profile
0

I have no idea because I've not played tat. Similarly, neither have you. Nor has painted_cow or anyone else complaining about him (as far as I can tell). He is obviously a good player because he is at the top of the leader board. Why, exactly, do you feel the need to bitch and moan about a person being at the top of a table on a website? It means nothing other than "he is good at dominion", and you know what... if he wins 76% of his games, he is good. Doesn't matter who his opponents are. There's a lot of luck involved in Dominion and if someone can overcome that luck and win 76% of games, they're good at Dominion.
Logged

Agrisios

  • Chancellor
  • ***
  • Offline Offline
  • Posts: 24
  • Respect: 0
    • View Profile
0

If you think otherwise, then propose matches directly with tat, and beat tat.

I've tried it too and was ignored several times. And I'm not a top player - just lv 32.
Just wanted to add that I don't mind occasional matches with great level differences but for most of the games I will invite players of the same or better level.
But I appreciate your intention. And you are right, it's off topic - sorry. I'm awaiting fp's post too.
Logged

Dave970

  • Ambassador
  • ***
  • Offline Offline
  • Posts: 30
  • Respect: +2
    • View Profile
0

Correct me if I'm misinterpreting you then, but you think tat would easily be the strongest player on isotropic if he played other strong players too, then? If this is not what you believe, can you explain?

If you're asking me, what I believe is that the number is meaningless... for everyone.  Like the game, play the game, win some games, have fun.  There's enough RNG inherent in the game that anyone can play "flawlessly" (per the given board) and still lose.

If you're asking someone else, hopefully they answer you, also.
Logged

Brando Commando

  • Apprentice
  • *****
  • Offline Offline
  • Posts: 255
  • Respect: +112
    • View Profile
0

I've just sent an e-mail to Doug Z requesting info on how the ranking works, so maybe we'll get some clarification on that.
Logged

Superdad

  • Moneylender
  • ****
  • Offline Offline
  • Posts: 161
  • Respect: +2
    • View Profile
0

Do we know how many points he gained/lost in this streak?

Level 8 -> 94:11
Level 18 -> 31:23
Level 9 -> 62:26
Level 12 -> 31:0
Level 4 -> 25:33 (lost)
Level 4 -> 20:5
Level 5 -> 40:18
Level 6 -> 30:10
Level 11 -> 38:33
Level 1 -> 37:18
Level 4 -> 36:18
Level 0 -> 46:19
Level 7 -> 46:29
Level 15 -> 3:4 (lost)
Level 11 -> 40:43 (lost)
Level 8 -> 48:48 (win)


For example, if he wins 1 "point" for the victories, but loses 40 for each of his losses against the level 15, level 4 and level 11. I mean, he lost to a level 4 here.

I know if this were MtG, the above streak of games would have resulted in a net LOSS of rating points. Without a doubt.

For example, I had a rating of roughly 2400-2500 in MtG. If I played against a "level 4", i.e. someone with a rating of, say, 1500, I would either win 1 rating point (if that) or lose upwards of 40. I've gone 12-1 in a tournament and LOST rating points, because it was a casual tournament with poor rated players. My one loss outweighed the 12 wins.

I know in MtG, that you never ever want to play against lower ranked players, because you WILL lose the random game due to land flood/screw and this will destroy your rating. You essentially must go something like 40-1 against them, but every 20 games you'll lose one to unsurmountable bad luck, so over the long haul, you are screwed playing lower ranked players.

It all depends on the numbers.
« Last Edit: July 26, 2011, 10:13:05 am by Superdad »
Logged

Fabian

  • 2012 Swedish Champion
  • *
  • Offline Offline
  • Posts: 666
  • Respect: +542
    • View Profile
0

Thisisnotasmile, "playing weak players does not artificially increase your rating" and "I don't know if tat would be the strongest player if he played other strong players" are not logically consistent statements. Let's leave the discussion there.

Dave, I was asking Thisisnotasmile, who was the one making statements about it. I appreciate your opinion too though :)
Logged

kn1tt3r

  • Minion
  • *****
  • Offline Offline
  • Posts: 585
  • Respect: +278
    • View Profile
0

Just for interest, how can you know against what levels anyone has played, if you don't want to search for every single one in the leaderboard?
Logged

Dave970

  • Ambassador
  • ***
  • Offline Offline
  • Posts: 30
  • Respect: +2
    • View Profile
0

I have no idea because I've not played tat. Similarly, neither have you. Nor has painted_cow or anyone else complaining about him (as far as I can tell). He is obviously a good player because he is at the top of the leader board. Why, exactly, do you feel the need to bitch and moan about a person being at the top of a table on a website? It means nothing other than "he is good at dominion", and you know what... if he wins 76% of his games, he is good. Doesn't matter who his opponents are. There's a lot of luck involved in Dominion and if someone can overcome that luck and win 76% of games, they're good at Dominion.

Essentially this.

I was lucky enough to play against Theory one time, long before isotropic was as popular as it is today (before you couldn't even log on on an Eastern US Friday afternoon), and long before these forums existed.  The early days of the blog.  I lost to him by one point, because somewhere in the middle of the game, I took a silver, or something, instead of an estate, and at the end of the game, he was able to pick up an estate right before the game ended with a 4/4 province split.  His level on isotropic, today, is roughly double mine.

Do I think he's twice the player I am?  No.

Better?  Maybe... we'll never know.  We played one time, and that's not enough to determine anything.

I would like to see the results of matches between you guys and tat, so it's unfortunate that those games are being declined, but I would still submit that even one game against him won't prove anything.  Of course, the reason I'd like to see the results, is to look at the game log, and see how the game played out, and compare that to what my in-game choices would have been.

I don't know what else there really is to say on this, but I don't think it's something we should be getting worked up about.
« Last Edit: July 26, 2011, 10:19:41 am by Dave970 »
Logged

ackack

  • Explorer
  • *****
  • Offline Offline
  • Posts: 302
  • Respect: +19
    • View Profile
0

A few comments:

Re: tat, I'd say refusing to play good players is a lot more damning than playing a bunch of bad players.

I've gone through streaks of matching high rated people and streaks of automatching, which tends to put you up against lower rated players. I had basically rating stasis during the former and have tended to slowly increase during the latter. Thus, I think there is a good chance that lower ranked people are systematically overrated. But I doubt that TrueSkill itself is inherently inflationary in this way. (And on that note, all the people talking about emailing Doug: you can just check the isotropic FAQ and see how he has TrueSkill configured.)

I'm not sure I'd agree with Fabian that matching high rated players and getting takers is that fast. Maybe my experience is skewed because the last time I did that was when I was in the high 20s, but it was a lot clunkier than automatch in terms of getting games fast. I'd also note that it seems like a fair number of high-end players play with restrictions, which one could also argue sullies the leaderboard in some sense.

Logged

ARTjoMS

  • Bishop
  • ****
  • Offline Offline
  • Posts: 109
  • Respect: +6
    • View Profile
0

yeah, now when i'm close to top i usually play only lvl 35+, will try to play more low levels to see how rating changes.

Logged
Quote
When a friend of mine sees a girl he finds attractive, he remarks how he'd like to "Throne Room" or "King's Court" her.
- Axe Knight

''Especially regarding such an iconic (and somewhat infamous) name that is known as ARTjoMS.'' - shark_bait is boosting my ego.

WanderingWinder

  • Adventurer
  • ******
  • Offline Offline
  • Posts: 5275
  • ...doesn't really matter to me
  • Respect: +4381
    • View Profile
    • WanderingWinder YouTube Page
0

Dave, don't think I haven't challenged tat many times. He won't play me.

THIS is the real issue, and one that I've come up against too. But the dude(tte) can do whatever (s)he wants, and I certainly haven't seen him (her) boasting or bragging anywhere, so I don't really see what to be upset about. Sure, I'd like him to play higher rated people too, but I'm not going to lose any sleep over it. And if you think he's not as strong... who cares?

Also, I don't believe that playing weaker players is intrinsically better for your rating than playing stronger ones, but it is a different skillset to be able to really milk the percentages out of the lower guys than to be more proficient at actually just winning in general (against harder comp). Playing overrated players is, of course, a way to gain, but I don't know of a way to do that.

Thisisnotasmile

  • Saboteur
  • *****
  • Offline Offline
  • Posts: 1493
  • Respect: +676
    • View Profile
0

Thisisnotasmile, "playing weak players does not artificially increase your rating" and "I don't know if tat would be the strongest player if he played other strong players" are not logically consistent statements. Let's leave the discussion there.

Dave, I was asking Thisisnotasmile, who was the one making statements about it. I appreciate your opinion too though :)

I've said nothing about "playing weak players". I have only discussed automatching vs playing against "high ranked players". From my experience, playing "high ranked players" inflates your rating more than automatching. I used to play "high ranked players" and during that time I was ranked around a level 33. Now I automatch, and I am currently ranked around level 27.

I also said nothing about whether I think tat would be the "strongest player" if he played "other strong players". I simply stated that I have no experience of playing Dominion with tat, and therefore cannot draw any conclusions about his play style and/or ability. I assume that since TrueSkill, a universally used rating system, ranks him quite high in his Dominion playing abilities, he is at least "quite good" at Dominion. I feel exactly the same way about other people who have high TrueSkill rankings.

How people choose which matches to play makes no difference to me (as long as they don't pick and choose cards to play with). You cannot tell anything about a players skill by the way they choose their opponents, only by their playing abilities. If automatch does "only match you against weak opponents", then that's a problem with automatch, not a problem with tat, me or anyone else who chooses to use automatch.
Logged

ackack

  • Explorer
  • *****
  • Offline Offline
  • Posts: 302
  • Respect: +19
    • View Profile
0

Also, I don't believe that playing weaker players is intrinsically better for your rating than playing stronger ones, but it is a different skillset to be able to really milk the percentages out of the lower guys than to be more proficient at actually just winning in general (against harder comp). Playing overrated players is, of course, a way to gain, but I don't know of a way to do that.

I'd guess one possible difference, even if they are rated correctly, is how good you are at consistently finding good enough strategies vs. finding near optimal strategies. If you always play a B game that's good enough to beat the weaker players, you can gain against weaker players; if you alternate between an A and a C game, there's probably not that much of a disparity in your winrate against players of various skills past a certain point and you don't benefit as much from playing them.
Logged

Fabian

  • 2012 Swedish Champion
  • *
  • Offline Offline
  • Posts: 666
  • Respect: +542
    • View Profile
0

Me: "Do you get "playing exclusively weak players and winning 76% of your matches against them will result in a higher ranking than strong players playing other strong players" then?"

You: "No"

Are you quite sure you haven't said anything about playing weak players?

I know you started out arguing about your own experiences with playing strong players vs automatching. I was arguing that tat's ranking has been raised artificially by choosing to play exclusively weak players while refusing to play strong players, and that this choice has lead to his ranking being higher than what his actual skill reflects. To this, you responded "No" (proof: see above), and the only way to reconcile that with him being by far the highest rated player on the leaderboard is if he was the strongest player, as well. As you point out, you and I don't know him well enough to definitely say he's not, but it's extremely likely he isn't (proof: his refusal to play any high ranked players), and his incredibly high ranking has more to do with the unique method he chooses his opponents (=artifically raising his rating). Even if he actually is the strongest player, we don't know that as you point out, so a resounding "No" to the above question is still not logically consistent. An "I don't know" or "Possibly" might be more reasonable in that case. I do agree that he's almost definitely at least "quite good", fwiw.

Whether all this is important or not is certainly arguable. I'm not losing any sleep over it, to be sure, but I don't think that means I can't point out that I think it's somewhat lame, on a Dominion Message Board.

Finally, based on your last paragraph, you're missing the point about tat and his method of choosing players. It's not the automatching in itself that's the issue; to be honest I don't even know if he automatches. The "issue" is that he plays exclusively weak players, and refuses to play against strong players. I don't think there's any problem with automatching, at least not that I'm aware of.
« Last Edit: July 26, 2011, 11:42:10 am by Fabian »
Logged

DStu

  • Margrave
  • *****
  • Offline Offline
  • Posts: 2627
  • Respect: +1490
    • View Profile
0

How Trueskill works: http://research.microsoft.com/en-us/projects/trueskill/details.aspx
Code for TS on Isotropic: http://github.com/dougz/trueskill

Also interesting for constants etc: http://dominion.isotropic.org/faq/

Me myself I'm also in the mid of the Twenties and feel like automatching I'm loosing rating, like Thisisnotasmile. Seems like there is some gap, when you're really good you can win often enough against noobs so you can gain rating, while if you're somewhere in the middle you will loose too often due to some errors of yours and your rating drops.
« Last Edit: July 26, 2011, 11:49:07 am by DStu »
Logged

rrenaud

  • Administrator
  • *****
  • Offline Offline
  • Posts: 991
  • Uncivilized Barbarian of Statistics
  • Respect: +1197
    • View Profile
    • CouncilRoom
0

FWIW, I think Larry reached number 1 by calculating that he was winning disproportionately more than TrueSkill predicts against lower rated players.  He was #1 without a winning record against anyone in the top 30.  I could easily believe this would be true that tat is very good at beating mediocre dominion players, but not good at beating great ones.

I don't care enough to do it myself (I am more interested in learning about Dominion than learning about Dominion players), but if someone was really concerned and wanted to do more than complain on a forum, you could grab the trueskill code, download the games, and then optimize the trueskill parameters to maximize the log likelihood of its predictions.  After you've tuned it to maximize the log liklihood, I'll have a separate 'accurate' ladder hosted on councilroom with the optimized parameters.  As smart and talented as Doug is, I'd be surprised if he got the trueskill parameters perfect without any tuning on real data 9 months ago.

http://en.wikipedia.org/wiki/Maximum_likelihood
Logged

shark_bait

  • Saboteur
  • *****
  • Offline Offline
  • Posts: 1103
  • Shuffle iT Username: shark_bait
  • Luckyfin and Land of Hinter for iso aliases
  • Respect: +1868
    • View Profile
0

Looking at tat's councilroom stats (http://councilroom.com/player?player=tat), he is 1 - 3 against ARTjoMS.  Assuming this ARTjoMS is Latvia who currently has a ranking of 41, there are at least a few games of a highly ranked player against him.
Logged
Hello.  Name's Bruce.  It's all right.  I understand.  Why trust a shark, right?

Is quite curious - Who is the mystical "Celestial Chameleon"?

Brando Commando

  • Apprentice
  • *****
  • Offline Offline
  • Posts: 255
  • Respect: +112
    • View Profile
0

How Trueskill works: http://research.microsoft.com/en-us/projects/trueskill/details.aspx
Code for TS on Isotropic: http://github.com/dougz/trueskill

Also interesting for constants etc: http://dominion.isotropic.org/faq/

I've looked through these and don't really understand what's going on or how to apply it to this discussion. But I found this bit from the FAQ provided by the Microsoft dudes who developed the TrueSkill ranking that Doug has adapted for Isotropic (this is about an Xbox racing game where the results are that you come in 1st, 2nd, etc.):

Quote
"Roughly speaking, the change in your skill estimate depends on how "surprising" the game outcome is. If you happen to be (among) the player(s) with the highest skill in each of the games you played, then the 25 wins were not surprising and hence none of these games provided a significant increase in your skill estimate. However, if coming 5th was a rather unlikely outcome in the game were you actually did come fifth, then your skill needs to be adapted significantly. Another way of seeing the issue is that TrueSkill does take the strength of the opposition into account. One cannot simply compute the win ratio and equate this with skill; if all wins happen in ... unbalanced games then a win is not really testament to your (even) high(er) skill!"

Rrenaud's note makes sense to me, so sure, maybe the guy/girl is just cherry-picking his opponents (in this case, automatch is a cherry picker). I just haven't found any notes from Doug or the TrueSkill developers that suggest that beating people who are much weaker than you is even a viable way to get to the top of the leaderboard.

And the experience of people who have found their (formerly high) rankings dip after they have started using automatch seems to support the idea that you need to beat similarly ranked people to get ahead. Presumably, these people were randomly matched against weaker opponents -- their loses were worse for them, because they weren't supposed to lose, so they suffered in the rankings.
« Last Edit: July 26, 2011, 12:11:33 pm by Brando Commando »
Logged

DStu

  • Margrave
  • *****
  • Offline Offline
  • Posts: 2627
  • Respect: +1490
    • View Profile
0

I looked on this this mornig, and I think it's a lot depending on the constants. First, they don't give an explicit formula for this nu-Funktion, but prob, one can find it on the internetz.The quotient of nu( (mu_1-mu_2)/c, epsilon/c) and nu( mu_2-mu_1)/c, epsilon/c) must correspond to the probability, that player1 wins (or loose, don't want to think about the sign) against player2, in order to not in- or deflate ratings due to playing lower opponents.

So more or less everything depends on the choice of beta, which directly enters in the c, if it's to high (or low, don't want to think about the sign) you can gain rating by slaughtering lowies, if it's to low (or high....) you will lose rating when playing lowies.
And I think it is reasonable to assume that the "correct" balanced values of beta would not be the same for all levels,which would explain the different experiences of Thisisnotasmile and me in contrast to what top players will see.

Edit: In this case it can well be that the parameters are optimal in the sense that they optimize the overall likelihoods, which does not prevent them from being "wrong" at the top edge (or anywhere else) of the ladder.
« Last Edit: July 26, 2011, 12:25:26 pm by DStu »
Logged

rrenaud

  • Administrator
  • *****
  • Offline Offline
  • Posts: 991
  • Uncivilized Barbarian of Statistics
  • Respect: +1197
    • View Profile
    • CouncilRoom
0

TrueSkill has a parameter that is supposed to measure how random the game outcome is.  You'd expect this to be very high for a game like chutes and ladders or Rock Paper Scissors, and low for Chess.  If the parameter is very high, the model thinks that the games are very random, and then the model thinks that it requires very much skill to do better than 50/50, and so it compensates you by giving you a huge level for winning frequently, even against low rated players ("omg, it's AMAZING that you win this mostly random game so often!").  On the other hand, if the parameter is very low, then the model thinks that even with a small skill advantage, you should almost certainly beat a somewhat lower rated player, and it will penalize you harshly for losing ("it must have been lack of skill, and not luck, that made you win, therefore, you certainly aren't as good as I used to think you were").

I suspect the parameter is not optimally tuned on isotropic.
Logged

rrenaud

  • Administrator
  • *****
  • Offline Offline
  • Posts: 991
  • Uncivilized Barbarian of Statistics
  • Respect: +1197
    • View Profile
    • CouncilRoom
0

First, they don't give an explicit formula for this nu-Funktion, but prob, one can find it on the internetz.The quotient of nu( (mu_1-mu_2)/c, epsilon/c) and nu( mu_2-mu_1)/c, epsilon/c)

<3 Deutschland
Logged

timchen

  • Minion
  • *****
  • Offline Offline
  • Posts: 704
  • Shuffle iT Username: allfail
  • Respect: +234
    • View Profile
0

For me it definitely seems harder to bloat your rank by playing random opponents. My personal high (lv 42) comes in when I played more with higher ranked players. When I play with automatch, like I almost always do nowadays, I stuck at lv 38-39.

It's hard to imagine there is a strategy which benefits more at playing noobs, unless one chooses the setup purposefully. It is nevertheless possible that playing noobs does not distinguish between certain skills and some players can benefit from that. It is hard to think one can just top the chart by doing this though.
Logged

Dave970

  • Ambassador
  • ***
  • Offline Offline
  • Posts: 30
  • Respect: +2
    • View Profile
0

TrueSkill has a parameter that is supposed to measure how random the game outcome is.  You'd expect this to be very high for a game like chutes and ladders or Rock Paper Scissors, and low for Chess.  If the parameter is very high, the model thinks that the games are very random, and then the model thinks that it requires very much skill to do better than 50/50, and so it compensates you by giving you a huge level for winning frequently, even against low rated players ("omg, it's AMAZING that you win this mostly random game so often!").  On the other hand, if the parameter is very low, then the model thinks that even with a small skill advantage, you should almost certainly beat a somewhat lower rated player, and it will penalize you harshly for losing ("it must have been lack of skill, and not luck, that made you win, therefore, you certainly aren't as good as I used to think you were").

I suspect the parameter is not optimally tuned on isotropic.

It appears, then, based on the statements of others saying that tat would not accept their game requests, that the parameter is high, possibly a little too high.  That being said, there really is a fair amount of randomness in any given Dominion game.  Right from possibly only one person hitting the 5/2 start, down through each shuffle of each deck (not considering strategies that reduce shuffle luck, which can be a measure of player skill).  If the obvious path to victory in a given game is Big Money, with no variants, and both players know this and play accordingly, then you bet the inherent randomness of the game is going to drive the victory towards a coin flip, regardless of the players' designated skill levels coming in.
Logged

DStu

  • Margrave
  • *****
  • Offline Offline
  • Posts: 2627
  • Respect: +1490
    • View Profile
0

Yeah. But the thing is, that you only have one parameter to tune the randomness. If you now tune it by "usual" games where people with more or less equal skill meets each other, and than extrapolate this by this Gaussian density function to games with high difference in levels, the assumption about the randomness might be completly wrong. You should even expect this.

It might even get worse. No matter on which set of games you estimate the parameters, if you manually decide to play on a set with completely different properties (like it seems to be the case here), than you the estimator of the randomness might be wrong.

The only exception is if the reality really has this Gaussian behaviour with the same constant for all kind of games, top player against top player, bad player against bad player, top player against bad player... . Where I see no reason why this should be the case. Of course, we could just have the wrong parameter, but than Thisisnotasmile and me should not have the feeling that we loose rating by automatching while the top players seems to win rating when they do.
Logged

rspeer

  • Witch
  • *****
  • Offline Offline
  • Posts: 469
  • Respect: +877
    • View Profile
0

TrueSkill has a parameter that is supposed to measure how random the game outcome is.  You'd expect this to be very high for a game like chutes and ladders or Rock Paper Scissors, and low for Chess.  If the parameter is very high, the model thinks that the games are very random, and then the model thinks that it requires very much skill to do better than 50/50, and so it compensates you by giving you a huge level for winning frequently, even against low rated players ("omg, it's AMAZING that you win this mostly random game so often!").  On the other hand, if the parameter is very low, then the model thinks that even with a small skill advantage, you should almost certainly beat a somewhat lower rated player, and it will penalize you harshly for losing ("it must have been lack of skill, and not luck, that made you win, therefore, you certainly aren't as good as I used to think you were").

I suspect the parameter is not optimally tuned on isotropic.

I suspect it's tuned to the average player, and tat plays in a way that is less random than usual.

Now, it may be that the distribution isn't Gaussian, and that high-level players are always less random. Or it could be something in particular about the way that tat plays.

Regardless, Doug probably shouldn't turn down the randomness factor: if he turned it down to less than the average, then people who care about their rank wouldn't want to play lower-ranked players. (Which they already don't, to some extent, despite tat's counterexample.)
Logged
Pages: 1 [2] 3 4 5  All
 

Page created in 0.056 seconds with 21 queries.