League General Discussion Thread


  • I.e. formula needs a K-factor but it does not have to vary with numbers of games played.


  • Might it be viable to implement a fixed game value for matches against new players (like 25) UNTIL his or her first win? With the first win the new player is considered to have started that game with his/her opponent’s ELO

    Example (with an overall K-factor of 50)
    A new guy looses two games, then wins the third against a 1500 ELO player. His ELO from here would be
    1500 +25 (for last win) -25*2 = 1475

    It is just an idea I just got. Only if it appears of interest I want to elaborate on it.

  • 2024 2023 '22 '15 '11 '10 Official Q&A Moderator

    I also considered the timing of games starting and ending, but thought that it’s OK to just let it come out on average.
    If someone delays a resignation because they want to wait for somebody else’s result to come in first, I think it’s just another minor inherent flaw of the ELO system being used for our super long brainy and dicey games.
    But I admit I didn’t think about it very long and welcome discussion.
    I think the vast majority of players will continue to resign when they feel suffocated, and over all on average that works out.

    I remind myself that this is not a science - we are not measuring the temperature of the air, where there is only one correct number. Every system will have weaknesses, we just want one that’s going to do a really good job at reporting to us the information that we want.

    I don’t like the idea where a new guy doesn’t get a deduction from his 1500 ELO until after he wins one…

    I love the idea of looking harder at the k factors and not having more different ranges than necessary. I confess I should have invested more thought energy into that sooner, but I’ll do that now.
    As Mr Roboto said, those factors are definitely not set in stone. He put them out as a starting point, and I tweaked them a couple times and then quit.

  • 2023 '20

    @gamerman01

    I agree with @pacifiersboard in the request for the ELO to be based on the beginning status of the 2 players, not just for the effect of timing on the gain/loss on purpose. The effect of timing for not on purpose.

    Not so hypothetical: A game last 6 months. We judge ELO affect based on when games ends.

    Player Axis simultaneously plays and finishes 10 games while player Allies only finishes 4 other games, the ELO affect for both players on the game is largely different than when they started. For those of you trying to improve your ELO, you are probably, just as before, going to choose you opponents and the attention given to games for maximum effect. Or at least give it some weight. Hard to do that when the ELO maybe very different when its scored than the current information would predict.

    By only scoring at the end, when the game itself can go for months while ELO scores on both sides are changing, seems a bit like making a bet on a football game when the spread is allowed to change between the time the bet is placed and the end of the game.

    I am no mathematician, but I think that the variable K factor would compound this problem for experienced established players going up against an unknown new to league player.

    I have no skin in this game. I am unsure if I will ever go for play-offs, where this matters, as my life quite often prevents me from playing consistently, and that’s not fair to my opponents or the league at large at the play off level. But I do have this weird thing for general fairness. Which, to me, also includes knowing as much as is reasonable to predict what you sign up for when you sign for it.

    My 2 cents. Well, that’s probably 5 with current inflation.


  • Wait, BombsAway is back in 2023 after last game in league before was November 10, 2015?! That’s newsworthy

    I’m doing some tests on BombsAway and Booper who are new to BM this year and played 5 or 6 games to see if the sensitivity is enough to apparently give them a fair shot at a fair seed in the playoffs.

  • 2024 2023 '22 '15 '11 '10 Official Q&A Moderator

    I’ll read Mainah’s post in a minute. I’m in the middle of looking at k factors

    110 for games 1-3
    90 for games 4-6

    We currently have 3 game minimum for OOB and PTV, and 6 for BM, is the reason for these ranges.

    Looking at Booper and Bombs Away, who had 5 or 6 games of BM completed this year, their ending ELO looks appropriate to me based on who they played.

    Interestingly, BOTH played our beloved Dawg, TWICE, FIRST. Getting the rust off, getting the feel of the game, it would seem.

    @BombsAway, a veteran of the game but maybe not with BM, rips off 3 more wins and loses to MrRoboto. 2 of the wins were very impressive, Me1945 and Wizmark.
    His ELO today is a little lower than Me1945 and Wizmark. Appropriate because he’s only played 6 games, and he did lose to MrRoboto at the end.
    He’ll get a seat at the table, the chance to win the 2023 championship game, if 2-3 more players above him don’t participate. Let that sink in. He can enter the top playoff, albeit a low seed, after 2 impressive wins. Seems the ELO k factors are definitely sufficiently sensitive. IMO he shouldn’t be as high as Wizmark and Me1945 even though he beat them. He only played 6 games and he lost one. 2 of the wins were against good ol’ Dawg.

    @Booper , After warming up TWICE on good ol’ Dawg (our grizzled, veteran trainer), beat Simon twice and learned a lesson or two from @GeneralDisarray. 5 games is not enough to qualify for the playoffs, so a 6th would give more information. (And this is a good example of why 6 are required) He has an ELO today of 1536, slightly above average. Seems right to me.

    I eliminated range 6-10 games and just made 6+ to 50 sensitivity as it was for 11+ before. Little change, just everyone in the top 25 that I looked at dropped a couple points and there were a couple minor position changes for players who were already close, as would be expected. So I’m thinking 50 k factor for games 7+, about the right size that you get a noticeable bump up or down for a win or loss, but not too much.

    These factors can be changed in future years. We’re just rolling it out now, I really don’t think we need to fine-tune k factors perfectly right now, it can be done later too, all the data is in there.

  • 2023 '22 '21 '20 '19 '18 '17 '16 '15

    I think elo should be based on ranking when game result is posted. Using the start date of the game is just an extra complicating factor that in most cases will have very little impact. I could write much more but this is the bottom line.

    However, I only see one problem and that is with the case of mass forfeits. How is that handled? One player decides to forfeit six (or many) games in the same day. In this situation the players ranking should count for all forfeits, right? @MrRoboto @gamerman01


  • @oysteilo said in League General Discussion Thread:

    I think elo should be based on ranking when game result is posted. Using the start date of the game is just an extra complicating factor that in most cases will have very little impact. I could write much more but this is the bottom line.

    However, I only see one problem and that is with the case of mass forfeits. How is that handled? One player decides to forfeit six (or many) games in the same day. In this situation the players ranking should count for all forfeits, right? @MrRoboto @gamerman01

    I knew someone could say it better than me, thank you for that.

    I am also a little bit concerned with game results that come in on the same day. I don’t know which the system calculates first, and that’s one for programmer MrRoboto

    Players ranking counts for all forfeits, same as before, if I understand you correctly. Issue is the order of the calculations, AFAIK


  • @gamerman01 said in League General Discussion Thread:

    Wait, BombsAway is back in 2023 after last game in league before was November 10, 2015?! That’s newsworthy

    I’m doing some tests on BombsAway and Booper who are new to BM this year and played 5 or 6 games to see if the sensitivity is enough to apparently give them a fair shot at a fair seed in the playoffs.

    Unfortunately I’m not going to be able to participate in playoffs this year. Been busier than anticipated, and have not been able to play much this latter part of the year.


  • @mainah a lot of interesting thoughts and points. I’ll just leave one with you, off the top of my head.

    Especially if you’re not talking about veteran players, your opponent may actually be improving over the 6 months that you are playing him. He is learning things from his other games, and you may be actually playing an increasingly good player over the months that you are playing him! So there is a weakness to counting ELO’s at game start, as well.

    This is not intended as a complete answer, but one thought that I think may be another thought to consider. I like your post.


  • @gamerman01 said in League General Discussion Thread:

    @oysteilo said in League General Discussion Thread:

    I think elo should be based on ranking when game result is posted. Using the start date of the game is just an extra complicating factor that in most cases will have very little impact. I could write much more but this is the bottom line.

    However, I only see one problem and that is with the case of mass forfeits. How is that handled? One player decides to forfeit six (or many) games in the same day. In this situation the players ranking should count for all forfeits, right? @MrRoboto @gamerman01

    I knew someone could say it better than me, thank you for that.

    I am also a little bit concerned with game results that come in on the same day. I don’t know which the system calculates first, and that’s one for programmer MrRoboto

    Players ranking counts for all forfeits, same as before, if I understand you correctly. Issue is the order of the calculations, AFAIK

    Yes, its the order of calculation i am refering too when multiple forfeits are posted the same day

  • 2023 '20

    @gamerman01
    Thank you. I understand the principle may or may not pan out in real life, or even in the formula. One of those basic intuition things that may or may not be true when applied to actual events
    But I’d like to point out it would solve the question of how does multiple forfeits in a single day effect the score, as that becomes irrelevant when the score is based on the start, not the finish.

  • 2024 2023 '22 '15 '11 '10 Official Q&A Moderator

    Actually would be the same issue, with starting on the same day as ending on the same day, I’m pretty sure.

    Would probably solve it if we entered a time, but probably not worth the trouble.
    Someone resigning all games at once because their cat died is still an issue, as all those games would all be decided at the exact same time.
    Actually, you’d have the same problem if 2 players started 2 games with each other at precisely the same time.

    Who woulda thunk? We have a lot of very analytical and precise people around here! Keep it coming!

    Keep in mind several of these issues are not set in stone for all time. We’re ushering in an big overhaul of the basic system that’s been used since G40 was invented.

  • 2023 '20

    @gamerman01 said in League General Discussion Thread:

    @mainah a lot of interesting thoughts and points. I’ll just leave one with you, off the top of my head.

    Especially if you’re not talking about veteran players, your opponent may actually be improving over the 6 months that you are playing him. He is learning things from his other games, and you may be actually playing an increasingly good player over the months that you are playing him! So there is a weakness to counting ELO’s at game start, as well.

    This is not intended as a complete answer, but one thought that I think may be another thought to consider. I like your post.

    And another thing - as the player is getting better (and we all want to) I don’t think that it should help up the eli affect on the first game, as the first few rounds are utterly crucial, and would be hard to make up for with your new knowledge.

    Player A elo 1300 vs b elo 1500. At start. Player A gets better over the 5 months, when the game ends is at 1400. Player B, having played flawlessly since G1, doesn’t get the full elo bump they earned by playing good from the beginning. Or maybe I have that backwards as I don’t know how the formula really works. Maybe it ends up reverse, player A gets an unfair change.

    And all of this, when applied to the “new to league unranked” player is really variable when they start at 1500 . And I understand that that is a sword that cuts both ways, wether we score at the beginning or end of the match.

    Again I don’t understand the mathematics behind the formula. I use carpenter in the field math without a calculator, not algorithms. So I don’t know how this plays out on the spreadsheet. I am trying to understand, but probably won’t. I do trust that both you and @MrRoboto will get it right.

  • 2023 '20

    @gamerman01
    Wrote my latest while you posted yours.

    Not sure - but I think 2 games started by the same players on the same game happens less than mass forfeits.
    And look at the effect of ghosts. Simon33 has gone mia (I hope he’s doing well, always enjoyed playing against him and learning from his prolific postings), and players have been slowly calling the game by the timing rules. Which means the elo affect of the person calling the game a month a go is different than the one calling it today before the end of the year.


  • These are some of the issues when going from a system that takes averages during the year and adjusts retroactively, to ELO. You do lose some of the benefits we’ve enjoyed, but overall is a significant improvement.

  • 2023 '20

    @gamerman01
    Agreed.
    Pardon the rapid fire posts. Wife and kids are off overnight enjoying a new year celebration while I am stuck at home as it was the only weekend I could tend to the smoke house.

    Coldsmoking hams for preservation take 48 hours, and the fire needs tending every hour or so. So I have 57 minutes to fill in an empty house 48 times over.

  • '19 '18

    Alright, I can finally chime in, but before I do:

    Happy new year everybody! It’s 10 am in Central Europe so I think most players on this community are in 2024 right, or do we have any Hawaiians here?


    There are 3 issues I want to address:

    1) How to handle new players.

    I’d like to keep it as is. @pacifiersboard said that right now, new players “should be advised to play low rated League players first.”
    This is generally a good idea, no matter the system! Think about it, it would be odd to see newcomers challenge the top of top or even reigning champions in other sports too, wouldn’t it?
    And even concerning their ELO ranking: While some strategic choosing of opponents might lead to a higher ELO in the short run, long-term ELO can’t be “gamed”. You will end up where you belong, eventually.


    2) K-factor (sensitivity)

    @gamerman01 said in League General Discussion Thread:

    I eliminated range 6-10 games and just made 6+ to 50 sensitivity as it was for 11+ before.

    I’m actually not very fond of that change.
    Right now, game 6 has a K-factor of 90 while game 7 has a K-factor of 50. Just as an explanation: Against an equal opponent, game 6 would award ± 45 points, while game 7 gives ±25. That’s a HUGE jump. Game 6 is almost double worth as game 7.

    It feels horrible to lose game 6 before winning game 7.

    One more reason: This gives the impression that ELO has more or less found the accurate place in the ranking after 6 (or 7) games. While I do trust the system to work for most new players after 6-7 games, any upsets or unusual results within these first couple of games would distort the ranking quite a lot. Imagine an upset (in any direction) happening in game 6, it would take longer for the system to “correct” the ranking afterwards.

    I’m MUCH more in favour of a gradual K-factor decline. In fact, even the 4 values we had before
    d765e0fe-a770-4a25-9f32-23d57c1f5093-image.png
    were not gradual enough for my liking.

    I’d prefer the following table:
    5d861ed5-a57a-40c5-a117-2e89bf72e16e-image.png

    The only reason I haven’t implemented it yet is because I have to adapt the formula but I will probably do it later today.


    3) Counting at the beginning or the end

    Originally I was completely with @gamerman01 on this. Yes, games can go on for quite a while and your opponent might (and probably will) change ELO in that timeframe: But so is their skill. And yes, early rounds have a bigger impact, althought an argument can be made that the decisive battles occur later in the game. In any way, the effect is negligible in my opinion.
    @mainah 's example is actually a great argument for that ;-)
    Player B with 1500 ELO is expected to win but with Player A at 1400 in the end, player B will be awarded a higher ELO change as opposed to counting at the beginning (A started at 1300). Which is fair, since apparently A got better during the play.
    Now player A loses more at 1400 of course, but on the other hand, with start-ELO-counting all of these values would be different anyway. Because those other games that ended during the game between A and B, that resulted in A’s ELO going from 1300 to 1400 - well they must have started some time. Some probably before starting the game against B, so the values wouldn’t be 1300 anyways.
    TLDR: The difference is negligible and it doesn’t really make a difference to count at start or beginning.

    HOWEVER!

    You convinced me with the mass forfeits.
    These do happen from time to time and even more often we have ghosts like @simon33 recently. I myself went AWOL in 2019, with multiple games against me being called as wins for my opponents.
    Right now the results are entered in order of posting. so first results are being calculated first.
    In @simon33 's case: The opponent who called the win first (that was @avner on Nov24) had the biggest advantage. After that call, simons Rating dropped from 1353 to 1330 so subsequents calls awarded less and less points - first @Sovietishcat on Dec 10, than myself on Dec 17 and finally @Adam514 on Dec 29.

    The points awarded were
    Avner - 23
    Sovietishcat - 19
    MrRoboto - 13
    Adam514 - 1

    Simon33 final rating: 1297

    Now had we called the game in a different order (adam first, then me, then sovietishcat and avner last), the points awarded would have been:

    Adam514 - 1
    MrRoboto - 15
    Sovietishcat - 20
    Avner - 21

    Simon33 final rating: 1296

    So the order of calling benefitted the first callers over the last callers (except Adam because he is so far from simon, that game is always worthless).


    HOWEVER HOWEVER

    As you can see, the differences in points are marginal at best! We arrived at a point in the discussion, where we are trying to tweak the absolute nuances. Counting at beginning or end won’t shake up the rankings, nor will it lead to any substantial rating changes.
    There are arguments for counting at the end (better reflects the improvements opponents make, especially when not absolute veterans with 100+games under their belt), there are arguments for counting at the beginning (fairer for mass forfeits and ghosts).
    My nod goes to counting at the beginning, if that’s also fine with @gamerman01 and some others.


    One last note

    I enjoy the eagerness of everybody to find loopholes in the system and to close any doors that some might try to exploit.
    But remember: You might be able to “game” the system and thus artifically increase your ELO higher than it should be by maybe 30-50 points (if at all!), but that would only be a momentary snapshot! A higher ELO will subsequently decrease the worth of future wins and also INCREASE the worth of future losses.
    You WILL end up exactly where your skill level is eventually, there is no way around that. It’s like the algorithms of social media feeds. You can tell yourself you’re not interested in this or that, but the algorithm can’t be tricked and knows the truth ;-)

    The ONLY way to climb the rankings sustainable beyond a short-term burst is to actually get better.

  • 2023 '20

    @MrRoboto
    Well that covers it Thank you.

    Stay in school and learn maths kids! All my attention and deep thinking and pondering of what if’s in the end mattered a 0.1% temporary difference to 2 players.


  • This is another MrRoboto post that I appreciate very much.

    @MrRoboto said in League General Discussion Thread:

    While some strategic choosing of opponents might lead to a higher ELO in the short run, long-term ELO can’t be “gamed”. You will end up where you belong, eventually.

    Yes.


    2) K-factor (sensitivity)

    @gamerman01 said in League General Discussion Thread:

    I eliminated range 6-10 games and just made 6+ to 50 sensitivity as it was for 11+ before.

    I’m actually not very fond of that change.
    Right now, game 6 has a K-factor of 90 while game 7 has a K-factor of 50. Just as an explanation: Against an equal opponent, game 6 would award ± 45 points, while game 7 gives ±25. That’s a HUGE jump. Game 6 is almost double worth as game 7.

    Yep. Playing around before the new year starts. K factors can be tinkered with for at least a month, if not more, into 2024 in my opinion. It’s not like we’re going to change them next December until the ratings are desirable for us. We’ll settle on some scale before long. Higher k factors early, declining over # of games played, and nothing crazy, is what I want to see. No arguments otherwise.

    I’m MUCH more in favour of a gradual K-factor decline.

    That’s cool

    The only reason I haven’t implemented it yet is because I have to adapt the formula but I will probably do it later today.

    Awesome


    3) Counting at the beginning or the end

    TLDR: The difference is negligible and it doesn’t really make a difference to count at start or beginning.

    This.
    It’s easier to record the resignation date when the result is posted. If we record the start date for a game when the game ends, then the rankings will jump around for everyone. If you record it as the resignation date, then only the ratings for the 2 players will change. If I’m understanding this correctly, then I really think we should continue recording the game end date.

    HOWEVER!

    You convinced me with the mass forfeits.

    So the order of calling benefitted the first callers over the last callers

    Of course. Because each game result recorded is a loss. Chalk it up to the uniqueness of our game (you don’t resign 10 tennis matches at once). The ELO system is designed for sequential game results, not simultaneous. Right?


    HOWEVER HOWEVER

    As you can see, the differences in points are marginal at best! We arrived at a point in the discussion, where we are trying to tweak the absolute nuances. Counting at beginning or end won’t shake up the rankings, nor will it lead to any substantial rating changes.

    This.

    My nod goes to counting at the beginning, if that’s also fine with @gamerman01 and some others.

    Please let me know what you think after considering my 2 points above.

    One last note

    I enjoy the eagerness of everybody to find loopholes in the system and to close any doors that some might try to exploit.

    :)

    You WILL end up exactly where your skill level is eventually, there is no way around that.

    You can run but you can’t hide. And most of us can’t quit playing.

Suggested Topics

  • 24
  • 51
  • 44
  • 47
  • 102
  • 141
  • 165
  • 102
Axis & Allies Boardgaming Custom Painted Miniatures

47

Online

17.4k

Users

39.9k

Topics

1.7m

Posts