Mvp + rank information

@ugh said in Mvp + rank information:

Okay, now you've lost me. If it is a 'poor predictor', isn't that equivalent with a 50/50 chance for either party and isn't that your professed goal is, i.e. making games fairer instead of skewed?

Oh I know I lost you... you were lost long ago on the Steam forums. Unpredictable based on us having NO information is not balance, it's ignorance. Unpredictable based on as much information as possible is the elimination of bias. That this has to be explained to you tells me you're going to remain lost.

@ugh said in Mvp + rank information:

If TVPlus can already predict that a game favours one party over another, why would you allow choosing a match that is clearly in favour of/biased towards one party?

We have matchmaking create matches with the lowest rating difference based on a rating that can best predict the outcome of the match... that creates the least predictable matches based on the most information. I'm not sure how this confuses you, or anyone.

Lets use boxing as an example. If we see from past fights that higher body weight tends to result in wins, then to create a more balanced match we try to pair up fighters of similar body weight. If we see from past fights that longer reach tends to result in wins, we try to pair up fighters of similar reach to create balanced matches. Through the miracle of math we can take that data and find how much weight a superior reach is worth, or how much reach a higher weight is worth, and we can create balance through a combination of the two, and so on.

What doesn't create balance at the match level is picking people completely at random.

@ugh said in Mvp + rank information:

It might also be a psychological problem of acceptance.

I think that IS one of your major problems, which is why you're forever arguing from incredulity. People's ignorance is not an argument against facts, it's an argument for people to shut up and learn the facts until they stop being ignorant.

@ugh said in Mvp + rank information:

I mean, I wasn't suggesting to use games-played-difference solely, but using it as an additional (weighted) criterion to minimize. How can that be worse than TVPlus only? It would just mean that if you have a choice between two similar TVPlus differences, you would look at the games-played-difference as well, so when you have two bad matchups, at least choose the one that people can accept more easily.

Cool, go do the math and come back with the results from the existing data supporting your belief. I've done the statistical regression and it says games played is not a significant contributor.

@ugh said in Mvp + rank information:

Can you give me an example how a concession can cause a bad matchmaking? I can imagine the other way, but not that. Or maybe you could elaborate how a correlation between something that can only happen after another thing can be confused with a 'causation'. Or is it that a lot of concessions can lead to worse matchmakings? How?

Ice cream sales correlate with murder rates. According to your usual garbage logic, one must cause the other... so tell me, which one do you think is the cause and which one is the effect?

What I did point out is that differences in performance rates more strongly correlate with concession rates, meaning that people were conceding more often when faced with better performing coaches than them, than they were when facing teams with higher TVs than them. You seem to have completely ignored that part of the story in favour of something you imagine you CAN argue.

@ugh said in Mvp + rank information:

Two statements that have no correlation/connection to me at all. You have no control now and you wouldn't have control then either. You would just exclude more unfair matchups, both for you and for others. In a competitive environment, isn't that what you strive for?

How do you not have more control over your matches if you can refuse them or if you can limit the types of matches you get?

@voodoomike said in Mvp + rank information:

We have matchmaking create matches with the lowest rating difference based on a rating that can best predict the outcome of the match... that creates the least predictable matches based on the most information. I'm not sure how this confuses you, or anyone.

That's the problem. You have a good predictor of clearly bad matchups and then you choose that part of the predictor where you're (still) ignorant. If you'd use the predictor to find those matches which are clearly bad matches, using that predictor would make sense. Yes, choosing this method does exclude bad matchups that you know about. But following from that that the matchup that are chosen by this exclusion-method are better is utter hogwash. Any one of them can still be a bad matchup. You're just confusing a necessary and a satisfying condition.

@voodoomike said in Mvp + rank information:

I think that IS one of your major problems, which is why you're forever arguing from incredulity. People's ignorance is not an argument against facts, it's an argument for people to shut up and learn the facts until they stop being ignorant.

It has nothing to do with ignorance or facts, but with feelings. People are not robots (well, most aren't) and they act according to their feelings as well as their ratio. That might be hard for you to accept, but I'm sure you'll find that is true if you do the research. And if feelings play a part in decisions and actions involving a game, maybe those feelings shouldn't be ignored, but incorporated into decision making about the game. Otherwise, you end up with lots of people with bad feelings and that will probably not contribute to the fun aspect that the game should have for those involved (and I'd say, which should be maximized in total, not just for some at the expense of others).

@voodoomike said in Mvp + rank information:

Ice cream sales correlate with murder rates. According to your usual garbage logic, one must cause the other... so tell me, which one do you think is the cause and which one is the effect?

Again, you show your poor skills at analogies. Are icecream sales and murder rates something intrinsically entwined like a match and a concession? Can a concession precede a matchup in a game? Can a concession in a game cause the bad matchup of that game? The answer to all of these questions is: No.

What I did point out is that differences in performance rates more strongly correlate with concession rates, meaning that people were conceding more often when faced with better performing coaches than them, than they were when facing teams with higher TVs than them. You seem to have completely ignored that part of the story in favour of something you imagine you CAN argue.

Maybe you should formulate things in more understandable terms when people ask you instead of dancing around the question, not answering it and just trying to show off your general knowledge of all things statistic.

I mean, you could just have stated: Yes, there is a correlation. It is stronger in COL than in CCL. There is also a strong correlation to matchups with good coaches. Instead you chose to formulate some sort of brain-twister no one is really interested in and also added the derailing bit about why statistics is not an exact science as you never know for sure what you really can take away from it.

@voodoomike said in Mvp + rank information:

How do you not have more control over your matches if you can refuse them or if you can limit the types of matches you get?

The amount of control is comparable. You still can't see who you're refusing, the matching is still random and might favour or disfavour you. Only some extreme (and supposedly rare) cases are excluded. I wouldn't call that control. But sure, if that's control for you, then yes, you have MUCH MUCH MUCH MORE control with that than otherwise.

@ugh said in Mvp + rank information:

That's the problem. You have a good predictor of clearly bad matchups and then you choose that part of the predictor where you're (still) ignorant. If you'd use the predictor to find those matches which are clearly bad matches, using that predictor would make sense. Yes, choosing this method does exclude bad matchups that you know about. But following from that that the matchup that are chosen by this exclusion-method are better is utter hogwash. Any one of them can still be a bad matchup. You're just confusing a necessary and a satisfying condition.

Your incredulity is irrelevant. You can't "not understand" the facts away - we can pretty clearly see from the data that it works the way I say it does, not the way you say it does. There's quite literally no room for argument here.

@ugh said in Mvp + rank information:

It has nothing to do with ignorance or facts, but with feelings.

If your feelings don't match the facts then your feelings are misleading you. If you feel something is true when it is demonstrably not true, then your feelings are wrong. Feelings are not a compass you can use to find the truth while facts are. I'll leave you to worry about people's feelsies... I'm going to stick to the facts.

@ugh said in Mvp + rank information:

Again, you show your poor skills at analogies. Are icecream sales and murder rates something intrinsically entwined like a match and a concession? Can a concession precede a matchup in a game? Can a concession in a game cause the bad matchup of that game? The answer to all of these questions is: No.

Based on what we know about both pairings yes, they are equally "entwined" because both pairs correlate. You're simply dropping back to your usual argument from incredulity because you feel it makes more sense for one pair to be causal than the other. Again you are insisting that one must cause the other, and that because one direction makes no sense to you, it must be the other - that is a false dilemma, just as it would be if you tried to decide which of the ice cream/murder rate pairing was causal.

@ugh said in Mvp + rank information:

Maybe you should formulate things in more understandable terms when people ask you instead of dancing around the question, not answering it and just trying to show off your general knowledge of all things statistic.

I'm answering them clearly, and I'm cautioning you on the logical fallacies you're applying and are clearly about to apply. If you really thought I was lording my superior knowledge of the topic over you it'd mean you understand that I do, in fact, understand the topic better than you do... which, given your unending bickering, you clearly do not.

@ugh said in Mvp + rank information:

I mean, you could just have stated: Yes, there is a correlation. It is stronger in COL than in CCL. There is also a strong correlation to matchups with good coaches.

Except then I'd be saying something completely different than what I actually said. I never said the correlation was stronger in one league than the other, I said it was stronger pre-TVPlus. I didn't say "there is also a strong correlation to matchups with good coaches" I said there is a stronger correlation to concessions with matches to coaches with superior win records. There's also the strongest correlation to the conceding coach's own past concession rate. Conceders gonna concede.

What none of this supports is your idea that we should attend to TV differences, or that we should let people refuse matches especially in CCL. That's dumb beyond belief. If you want to "refuse" matches then pay in COL and concede like a bitch. If you want to play in CCL, deal with the fact that the only way we find the coaches worthy of playing in the champion cup is by making everyone play matches by the same rules, and not letting people cherry-pick their matches.

@ugh said in Mvp + rank information:

The amount of control is comparable.

You're delusional.

@voodoomike said in Mvp + rank information:

" I said there is a stronger correlation to concessions with matches to coaches with superior win records. There's also the strongest correlation to the conceding coach's own past concession rate.

I would have thought high or low concessions rates were highly related to rosters and match-ups (with TV variations).

@voodoomike said in Mvp + rank information:

we can pretty clearly see from the data that it works the way I say it does (...). There's quite literally no room for argument here.

Are the datas available ?

@ungern said in Mvp + rank information:

I would have thought high or low concessions rates were highly related to rosters and match-ups (with TV variations).

There can be any number of reasons, no doubt, but the ones showing the highest predictive relationship with a given match's concession is the conceder's past concession rate, and the relative performance rates... the latter is under TV matching, though. Under TVPlus matching TV difference is very slightly higher than performance rate differences, but both have lost correlation strength with concession rates, each being lower than they used to be. Past concession rate is strong as ever.

You're welcome to build a model to test your theories about concession, of course.

@ungern said in Mvp + rank information:

Are the datas available ?

Mordrek's site has data from season1 on that you can download in sqlite format. If comparing pre-and-post TVPlus is the goal, you'd need to get the dataset from either myself or Dode. That was during the era of 2 minute pooling and no 500 TV max difference, though.

@voodoomike said in Mvp + rank information:

Your incredulity is irrelevant. You can't "not understand" the facts away - we can pretty clearly see from the data that it works the way I say it does, not the way you say it does. There's quite literally no room for argument here.

Your ignorance, however, obviously knows no bounds. I'll give you a simple example for what I mean, since you obviously can't understand it. TVPlus alone cannot predict the outcome of a game with two fresh teams of equal TV, but anybody who thinks about this for half a minute would concede that there are good and bad matchups between any two fresh teams, based on race-pairing and of course coach-skill difference that you have no idea about when looking at a win-loss record of 0 0 on both sides.

@ugh said in Mvp + rank information:

Your ignorance, however, obviously knows no bounds.

Uh oh, I get the feeling you're about to go off half-cocked, as usual.

@ugh said in Mvp + rank information:

I'll give you a simple example for what I mean, since you obviously can't understand it. TVPlus alone cannot predict the outcome of a game with two fresh teams of equal TV, but anybody who thinks about this for half a minute would concede that there are good and bad matchups between any two fresh teams, based on race-pairing and of course coach-skill difference that you have no idea about when looking at a win-loss record of 0 0 on both sides.

As predicted, there you go. The smart thing to do would be to ask "have you considered the situation...." rather than assuming I hadn't. As it happens, there are methods laid out to handle new teams and new coaches under full TVPlus AND we have looked at both the prevalence of such matches AND analyzed the utility of using a roster v. roster modifier in match prediction.

If we use FUMBBL's CRP data (which covers the entire time it ran under CRP) we find that matches in which both teams are playing their first game, and both teams have the same TV account for roughly 1% of all games, in spite of the fact that most teams play fewer than 10 matches, and the mean TV is low. It's almost 2% in COL. Thus, what we're talking about is the fact that under the current system, up to 2% of the time it has the same lack of information that it would have had under TV matching in the same situation 😉

Next, I have, in the past, aggregated win percentages for each race pairing, and created a "win percentage difference" variable which covers the aggregate bias in matches involving a given racial pairing. I included that variable during my logistic regressions back when Arne and I were talking about enhancing the predictive ability of matchmaking... and guess what? The analysis considered it a variable that lacked overall significance and opted to exclude it from the final formula.

Finally, in full TVPlus we discussed on a system whereby zSum does not start at 0, but rather begins at a value that is the mean z-score for zSums of all the teams the coach has played, adjusted to the zSum distribution of the new team. That means that new coaches begin with a zSum that is the mean value for the roster, while experienced coaches begin at a zSum that conforms to the zSum values they've managed to achieve with past teams, and thus reflects their overall performance as a coach. In fact, I explained that very system to you on the Steam forums during your endless arguments that TVPlus cannot be used for ranking.

So, no, this is not a case of MY ignorance, it is the usual case of you being ignorant of the topic and shooting off your mouth fast, furious, but ultimately falsely.

Hey, guiz.

Just a couple of thoughts (which I don't claim 100% are mine) to how we probably could "hotfix" some of the issues without changing MM and inducements too much.

  1. Though I'm not a big fan of full TV++ system (as it reduces value of good coaching throughout a season severely), I agree that new teams who enter a ladder later in a season may have significantly harder times trying to qualify for play-offs. So may be it would be a good idea to apply TV++ in its full potential to at least those new teams (when they play against already experienced teams), during their, say, 5-7 first matches?

  2. Will it be a good idea to apply full TV++ in play-offs and finals, as well? The reason behind it is anybody who got to this stage is proved to be a good coach and good strategist already. So may be at this point we should just reduce all the remaining differences between their teams and put them into conditions where their on-pitch skills alone will decide the best one? (though it seems to me that this way we just leave it to dices to decide the winner, but may be I'm wrong). Another benefit of this is that we'll have more interesting and unpredictable finals - which usually attract a lot of spectators. Shouldn't we at least put forward a thrilling, flashy show at the end of a long, daunting season?

last edited by Mori-Mori
  1. I'm hoping Cyanide will implement rookie team protection, preventing new teams from being matched outside narrower TV ranges, e.g. 300TV for fresh teams.

  2. TV+ is only used in ladders, not playoffs.

@dode74 , just to make context of that last post more clear. I tried to find how exactly TV++ in its fullest should operate, but couldn't. From different references and general idea coverage here and there (including mentions of it by @VoodooMike), though, it seems to be not only about MM (the extent of its usage in BB2 atm), but as well about redrafted algorithms of inducement calculation for underdog. Specifically, if CRP's inducements are balanced around idea of giving an underdog roughly ~30% chance of winning, TV++ is capable of bumping it to 50/50 situation.

Next,

@dode74 said in Mvp + rank information:

I'm hoping Cyanide will implement rookie team protection, preventing new teams from being matched outside narrower TV ranges, e.g. 300TV for fresh teams.

Do you think this alone would be enough for somebody joining week or two after the initial start? Specifically, will there be enough teams with TV under 1300 (accounting for effect of TV++ MM as well) at this point? What I thought is - as explained at the start of the post - if there is some system allowing to "equalize" new and experienced teams through a more "generous" inducement give-aways in automated manner, may be it should be applied as well during a few their initial matches (on top of any new teams protection by the MM itself), to facilitate their admission into ongoing season? This way, perhaps, they could be even paired with a more developed/successful teams as well.

@dode74 said in Mvp + rank information:

I'm hoping Cyanide will implement rookie team protection, preventing new teams from being matched outside narrower TV ranges, e.g. 300TV for fresh teams.
TV+ is only used in ladders, not playoffs.

Again, I was referring to the part of TV++ system which is designed to equalize 2 teams more evenly with a greater amounts of inducements. The question was: assuming such equalization is possible, would it be a good idea to apply it to playoffs and finals to rule out any remaining differences between teams which have made it there, to make it more about coaches' actual on-pitch skills, and less about whatever lucky or unlucky events affected the composition of their teams on the way to the finals?

last edited by Mori-Mori

Specifically, will there be enough teams with TV under 1300 (accounting for effect of TV++ MM as well) at this point?

Yes. The majority of teams don't get past 5 games played.

The question was: assuming such equalization is possible, would it be a good idea to apply it to playoffs and finals to rule out any remaining differences between teams which have made it there, to make it more about coaches' actual on-pitch skills, and less about whatever lucky or unlucky events affected the composition of their teams on the way to the finals?

No, imo. A playoff is a different beast to a matchmaking league. A matchmaking league is designed from the outset where anyone can join in and (population issues allowing) can get a reasonably even match; we've tacked on a competitive aspect to it in the form of qualification and playoffs, but those aren't the primary aim of the league.
A playoff is an elimination competition between teams (as opposed to just coaches) where getting to the next round is based on using the materiel you have at your disposal to beat your opponent regardless of who he is. If you want it to be about coaches there is NAF rez format available.

@mori-mori said in Mvp + rank information:

Though I'm not a big fan of full TV++ system (as it reduces value of good coaching throughout a season severely), I agree that new teams who enter a ladder later in a season may have significantly harder times trying to qualify for play-offs. So may be it would be a good idea to apply TV++ in its full potential to at least those new teams (when they play against already experienced teams), during their, say, 5-7 first matches?

It only reduces the value of good coaching throughout a season if you rank that season based on w/d/l record, and that's not how you do rankings under full TVPlus. Under full TVPlus you rank based on normalized zSum - meaning you're taking the distribution of zSum values for all teams of that roster, finding where a given team falls on it, and converting that into a Z-score, then ranking all teams based on their relative z-score. It actually increases the value of "good coaching" on ranking.

We're not using full TVPlus at the moment, only the TVPlus rating for MM. That doesn't diminish the effect of good coaching, it just (to the extent it can with small pooling sizes) mitigates the effects of lucky pairings on ranking. It means that it tries to pair the best coaches with one another such that their records become more distinct in the rankings.

@mori-mori said in Mvp + rank information:

Will it be a good idea to apply full TV++ in play-offs and finals, as well? The reason behind it is anybody who got to this stage is proved to be a good coach and good strategist already. So may be at this point we should just reduce all the remaining differences between their teams and put them into conditions where their on-pitch skills alone will decide the best one? (though it seems to me that this way we just leave it to dices to decide the winner, but may be I'm wrong). Another benefit of this is that we'll have more interesting and unpredictable finals - which usually attract a lot of spectators. Shouldn't we at least put forward a thrilling, flashy show at the end of a long, daunting season?

You can still use full TVPlus in play-offs and finals, but you replace individual coach zSum with a normalized roster zSum based on the zSum distributions created during the normal season from the top ~16% of teams. Doing so will roughly balance teams mechanically while no longer balancing them on coaching skill.

Honestly, I think in structured tournaments there's a case to be made for having NO inducements at all - inducements are, inherently, a handicapping system and applying such a thing to a tournament makes very little sense. If you want to use them then there's a good case for having them properly balance the teams at a mechanical level, leaving only coaching (and luck, of course) as the major determiner of outcome. TV difference won't do that... tournament TVPlus could (that's using normalized roster zSum rather than individual coach zSum).

@dode74 said in Mvp + rank information:

  1. I'm hoping Cyanide will implement rookie team protection, preventing new teams from being matched outside narrower TV ranges, e.g. 300TV for fresh teams.

  2. TV+ is only used in ladders, not playoffs.

He was referring to full TVPlus, not what's in place at the moment, I believe.