ROLadder announces some of its features.

[Snuffeldjuret]

I don't think they deserve to win that many points. Yet another foundation that we disagree on. The higher ranked team could and should suffer in team rating for using these lower rated individual rating, depending on exactly how you want to use these ratings. All Im saying is that it could be an idea to base it on that, thee question is exactly how you shall base it on that. If a team uses lower rated players, then the team will get a lower rating becuase of that.

Yes that is one way you could abuse that. But if you look at the RO ladder rules atm, it is just as easy to abuse the rules and say that you don't have enough players (a.k.a. no-show). That way you could get away without any punishment if you "do it right" or a smaller punishment. So in your example, it would be SO easily exploited, but it WOULD NOT MAKE A DIFFERENCE.

Basing the rating on old results is also an educated guess, done by the rating system. To add the individual thingy and add another element of the guess could and should (if done good enough) just make the guess better.

... You need to remember that if the selected group you end up playing in a competitive match changes the outcome of a match. That a lot of people will aim of keeping a good new player secret till there is a big match. As then you get that the system will predict the new player is worse allowing the team to gain more points if they win.

Remember if a teams strength isn't consistent that for a rating system that takes into account variance that doesn't have to matter a lot as it will then take into account that a team isn't really consistent. You can make calculations taking into account variance .

That is true Zets, but I have always known that. It's just that we haven't gotten anywhere in this discussion until just now =). Now we are taliking about some real issues here.

Trying to find solutions for real issues like this is interesting. Personally I like the idea of recalculating rating adjusments retroactively, but I am not sure if it could be applied in this kind of atmosphere. The big problem here is that we have so few "outcomes" or "specimen" or what to call it. It might just be too tough to get things to find accurate values enough for the overall system to be better than what you are working on atm.

 

[KingLol]

Ok if you honestly believe that team doesn't deserve to win as many points then that would put you in a very small minority. It appears the fundamental issue here is that you want to reward/punish teams based on what hasn't happened yet which is extremely unfair.

 

[Snuffeldjuret]

Ok if you honestly believe that team doesn't deserve to win as many points then that would put you in a very small minority. ...

... It appears the fundamental issue here is that you want to reward/punish teams based on what hasn't happened yet which is extremely unfair.

These two are not related in the way you present it. The second part is not completely true, especially not the part about what is fair. There is one thing that has actually happened. For instance their best player left. I think that rating shall be a prediction on their current skill based on their previous result. If people would bet money on the match, they would bet differently. Is that unfair? It's just the same way of predicting the result. And that is what Zets is always talking about. The rating shall be able to predict the outcome, and that won't happen if the rating is sometimes inaccurate enough to change the outcome.

The first part I think you agree with me if we look at it in a different way. We probably won't find common ground by discussing it on a forum though as it is one of the worst forms of communication known to man, but I can try.

A totally stupid example, but still...
TeamA has rating 1200
TeamB has rating 1100
TeamC has rating 1000
Imagine if one day all players of TeamA move to TeamB, and TeamC plays against TeamB. In this match TeamB consists of only old TeamA members that should be rated as 1200, and not as 1100 they are now just because they are under the TeamB banner. TeamC lost the match and now loses more rating than they woudl have done if the players of TeamA had stayed there and played against TeamC as TeamA.
You cannot tell me that TeamC was here treated in a fair manner?

A slightly different example but suddenly not even close to stupid.
TeamD has rating 1200
TeamE has rating 1000
TeamD closes down, but all those players among a few others start a new clan called TeamF.
TeamF has 1000 rating
TeamE and TeamF plays, and TeamF only uses players from old TeamD. TeamC loses the match and loses way more points than they would have done if TeamD didn't close down. It was exactly the same players playing, so how is this fair?

Having individual rating on players that is a base for the teams rating would in these two examples IMHO make things more fair and the rating adjustments more deserving. I strongly believe that I am not in a minority for thinking that.

 

[KingLol]

If it was man-for-man the exact same team then you'd have to question why they'd come back as a different team with less points or why they'd move their entire team. In this case I'm guessing you've just made it extreme for the purposes of clarity.

Either way, yes it's still fair. If the lower ranked team is truly at the level they were at before the new superteam shows up then they will make their way back to that spot, and the new superteam will make its way up the ladder - the speed at which these happen is where the ladder convergence comes in.

I agree that if a team's "best player" is unable to play, you'd think they'd be more likely to lose (and they probably would be) but why should they then have to be protected from losing lots of point? The team will be punished for not having the strength in depth needed to compensate for the lack of their best player.

The players in a team do not and should not ever matter in rankings, only the results that the team gets.

 

[Snuffeldjuret]

Yes, that is what it is all about. The ladder will converge, if the system allows it to by it's design and activity. This was never a discussion about the final picture in a stable clan scene.
Zets is working on improving the converging abilities on ROladder (at least in my impression lol). In my impression Zets and bswearer were discussing the coverging speed issue (rung vs modifiedELO). I mentioned another way to make things coverge faster, the prediction system. A system which I think is unfair, but it is just as unfair as the whole rating system itself. In the long run in a stable clan scene, both systems end up with same result. So that is imho not an argument why something is more fair than the other how the end result looks in an ideal world.

It is all about reducing everything that is inaccurate, as the clan scene does not seem to be stable enough to become accurate before the situation has changed and thus back to inaccuracy.

 

[KingLol]

The prediction system is unfair because it relies on things which haven't happened, it's based on conjecture. The past results system is fairer because it relies on things which have happened, it's based on facts. Now, if you could take into account every possible factor that will affect the result, then your prediction would be a fact but this is - and will always be - impossible because there are far too many sources of uncertainty. The result of this is that using a prediction system will always be less fair than using past results.

I briefly mentioned this before (can't remember if it was this thread or not) but the IRB rugby world rankings seems to have a good system. It's based on a points exchange system which has the added benefit of teams not being able to lose points if they win.

There's an explanation here:
http://www.irb.com/rankings/explain/index.html

 

[Snuffeldjuret]

Saying it does not make it so. Repeating your own argument has never been the way to win a discussion . I already mentioned that something did indeed happen, but I am not going to repeat myself .

One big advantage of individual rating is that it will be easier to find dates for clans to play on. People won't be as obsessed on getting dates their best players can play on. That is actually what I am most fond of in such a system, as I find it annoying when an opponent says "Hello, we can only play this day".

 

[KingLol]

I'm repeating my point because you don't seem to understand it! A team might have 8 great players who win everything and another 8 who are hopeless. If you think that the skill level of a team is dependent only on the absolute best performance it's capable of and should not include the strength in depth of the squad then why bother having a ladder at all? Why not replace it with a round robin or knockout tournament?

Finding dates for clans to play on is not a problem caused by the rating system, it's caused by there being no way to force teams to play. Also, most of the time then a team can only play on certain days because thats the only day they have enough players. When HoS arrives there will be more teams and more players = this wont be so much of a problem.

 

[Zetsumei]

It's based on a points exchange system which has the added benefit of teams not being able to lose points if they win.

There's an explanation here:
http://www.irb.com/rankings/explain/index.html

The reason why you cannot lose points there if you win is because you can only enter as results 0 (total loss), 0.5 (draw), 1 (complete win). But cannot enter values in between.

If you could have values between 0.5 and 1, then you would be able to loose points after you've essentially won. I showcased that bit in red in the picture.




Essentially that system is very similar to ELO, ELO for them would probably look something like this:



So rather than a straight line elo is pretty much a similar shape but then smoothed out.

 

[KingLol]

In that system 1.0 is a win and the exchange is multiplied for larger wins: moving that into the realm of ROladder, a 6:2 win has a 1.0 multiplier and an 8:0 win would have the 1.5 multiplier.

edit:
thanks for insulting my intelligence by showing me what a smoothed line looks like :troll:

edit 2:
what is the current system used on ROladder? I've just noticed it's a bit wonky, it's too easy for teams to steam up the rankings by playing weaker teams and lots of matches.

 

[Snuffeldjuret]

I'm repeating my point because you don't seem to understand it! A team might have 8 great players who win everything and another 8 who are hopeless. If you think that the skill level of a team is dependent only on the absolute best performance it's capable of and should not include the strength in depth of the squad then why bother having a ladder at all? Why not replace it with a round robin or knockout tournament?

Finding dates for clans to play on is not a problem caused by the rating system, it's caused by there being no way to force teams to play. Also, most of the time then a team can only play on certain days because thats the only day they have enough players. When HoS arrives there will be more teams and more players = this wont be so much of a problem.

I fully understand what you mean, I just don't agree with it to 100%. Huge difference. The question is if you understand my point? I doubt that as you don't seem to try work with the concept itself, but rather how I present the concept. You are listening too much on what I said, and try to little to really see the advantages of such a system and eliminating the problems you find.

I dont think that "the skill level of a team is dependent only on the absolute best performance it's capable of". If that is how you have interpreted what I have written, then that is unforteunate. Everything can be modified and adjusted, so don't take anything written in a suggestion as written in stone. That goes for every suggestion of there in this world ^^. We are discussing a concept, not the final definition.

It's still a part of the problem with finding dates. Not every problem has just one reason behind it... The rating system IS a part of the problem as a team have it much easier to find members that are not up to snuf for a high level match, if there is no rating adjustment after the match.

 

[KingLol]

Here's how I see it then:

A team is a black box. You don't know what is inside and the only thing that matters is what comes out (the result).

If you knew exactly what was inside it and how it worked, you could predict the output with 100% accuracy.

Because you don't know exactly what is inside and how it works, you cannot predict the result, therefore it's reduced to a guess based on a formula which "some guy" decided and that's the unfair part.

Because you can never FULLY characterise a players contribution to a team performance, that is less fair than having a ranking system based on past results.


Your disregard of strength in depth was not directly stated but it's logically implied. A team with 16 players ranking shoots up when they have their 8 best players and crashes down when they have their 8 worst players and yes, the timing of when they have their best and when they have their worst players can be unlucky but that's tough **** and will be averaged out over time.

edit:
for finding matches: that's NOT a problem with the rating system, it's a problem of clan leaders wanting to preserve their rankings by hook or by crook. Appeasing them by giving them less penalties is not solving the root cause problem.

 

[Zetsumei]

In that system 1.0 is a win and the exchange is multiplied for larger wins: moving that into the realm of ROladder, a 6:2 win has a 1.0 multiplier and an 8:0 win would have the 1.5 multiplier.

edit:
thanks for insulting my intelligence by showing me what a smoothed line looks like :troll:

The one is not a multi player it stands for a statistical chance of victory equal to one in other words it stands for 100% victory, and you cannot win with 150%. 6:2 in terms of RO means that you won 75% of the rounds or 0.75. (although this isn't always accurate as on an unbalanced map you are more likely to draw, for this reason in the current ROLadder system you can cancel matches once the prediction goes over 75%)

I just made that picture to illustrate my point, so no need to harass me.

edit 2:
what is the current system used on ROladder? I've just noticed it's a bit wonky, it's too easy for teams to steam up the rankings by playing weaker teams and lots of matches.

Don't look at the values at that graph as they are not ROLadders values for winning and losing a match and for point differences. Actually currently due to not taking in account the map unbalance means that when over a 60 point difference the weaker team got an advantage rather than disadvantage.

If you look on ROLadder a guide of thumb is that around a 60 point difference the system will predict you will win 6:2 and around a 90 point difference that you win 8:0. So for instance for Team ALC can hardly gain any points for playing Russian legion even for winning 8:0, and against Team Enclave it already predicts that alc will win 6:2. Or if you look at 3.SA it predicts that they will win 6:2 against FMJ and Cinergy (practically giving 3.SA 0 points if they win 6:2). All teams below that have a higher point difference meaning that 3.sa can only gain points by winning 8:0 while losing points for other outcomes.

(although currently punishment points are currently handled directly through the elo system, rather than only visually substract some points, which is generally bad for the scoring system as well).

So when trying to challenge weak teams it is actually harder to obtain points than when challenging stronger teams. The current system on ROL is a bit too simple but it serves its purpose.

 

[KingLol]

1 stands for 100% you cannot win 150% . 6:2 in terms of RO means that you won 75% of the game effectively or 0.75

You defining 1 to be the widest winning margin is where you appear to have become confused. In that system, 1 is the multiplier for the points exchange when a team wins by the narrowest margin possible and if they win by a large amount then the points exchange is multiplied by 1.5 (although in RO the largest winning margin is capped as well at 8:0).

re: the graph values: I wasn't assuming these were the ROladder values, my comment about it being wonky was based on looking the actual ladder points.

 

[Snuffeldjuret]

I already knew how you looked at it, and I thought I told you that I did.

And "some guy" is always deciding it. No matter which system is used. Either that "some guy" decides that people get to voice their opinion and he might redirect the responsability of choosing system. But it will always be one "some guy" and no matter how, you will have to think the system set in place would then be unfair.

I don't think the team with those 16 players shall have to go up and down chaotically in rating, if they don't have to. If applied good enough, individual rating would have the team's rating more static unless ofcourse their own skill changes over time. We have different ideas of what a good system is, so we should not discuss a solution as it is impossible to even come closer to one.

No it is a problem with the rating system, as clans leaders as you described would not cause that problem in a good system with individual rating.

This is pointless. I won't discuss this with you anymore. You may have the last word if you'd like, I won't respond. I might continue discussing this with others if the opportunity comes up though. Now good night and sleep tight!

 

[KingLol]

Well I'll accept your invitation:

Conclusion:
Neither system is perfect, but past results is less objectionable because it's more heavily based on facts, reducing the "some guy" influence. It essentially represents how good the team has been up until their most recent match. There's a reason no team ranking system bases the rank on individual members.

 

[Zetsumei]

You defining 1 to be the widest winning margin is where you appear to have become confused. In that system, 1 is the multiplier for the points exchange when a team wins by the narrowest margin possible and if they win by a large amount then the points exchange is multiplied by 1.5 (although in RO the largest winning margin is capped as well at 8:0).

re: the graph values: I wasn't assuming these were the ROladder values, my comment about it being wonky was based on looking the actual ladder points.

Ah my apologies then for not reading the full paper on that thing (I should be going to bed ).

That means that the system does have a big disadvantage though, as it would mean that you would need to force the limit of gaining any points close to the current 60 point mark on ROladder. Meaning that unlike now where you can still gain something by winning 8-0 against a team at over 60 points, you would then gain nothing as 0*1.5 is still 0.

About scores on rol, I get what you mean, it looks like not all scores are ordered correctly, might be due to the server transfer. Thank you for telling that.

 

[KingLol]

It wouldn't be too tricky to take the system used on the IRB rankings, there are 3 parameters that can be adjusted:

i.) gradient of the points exchange line - i.e. steeper gradient --> larger points exchanges

ii.) point exchange caps - move the flat lines capping the max number of points you can win/lose from a match

iii.) changing the 1.5 multiplier (or even the 1.0 multiplier I suppose although that would open up the possibility of winning teams losing points)

Because you can't gain points for losing with that system, it prevents teams challenging high ranked teams on very unbalanced maps in order to get a 2:6. This also means there's no particular advantage to playing lots of matches.

 

[Zetsumei]

It wouldn't be too tricky to take the system used on the IRB rankings, there are 3 parameters that can be adjusted:

i.) gradient of the points exchange line - i.e. steeper gradient --> larger points exchanges

ii.) point exchange caps - move the flat lines capping the max number of points you can win/lose from a match

iii.) changing the 1.5 multiplier (or even the 1.0 multiplier I suppose although that would open up the possibility of winning teams losing points)

Because you can't gain points for losing with that system, it prevents teams challenging high ranked teams on very unbalanced maps in order to get a 2:6. This also means there's no particular advantage to playing lots of matches.

It wouldn't be hard to modify, although the same parameters could easily be modified with ELO as well to obtain the same results. And ELO will likely be more accurate.

Similarly you can choose in both systems to only include absolute wins or allow for partial wins as well. And you can choose to give an additional reward for exceptional performance.

As said before, currently teams can freely decline matches with a point difference of over 60 points in ro:ladder. But if 2 teams like it, they can still play beyond there, and can still gain and loose points based on how they perform. As when arranging the match you can simply both pick a balanced map or decline and otherwise decline the match.

How active you are with an ELO or IRB system won't really change your score that much, as you're playing against a prediction. When playing against weaker clans generally you can hardly gain any points.

Although an issue with both systems is that by default it doesn't take account of the effect of time on the accuracy of the scoring. So say if you stop playing for a year you effectively would have the same score. Something that needs to be accounted for as well.

 

[-[SiN]-bswearer]

With football you use different tactics and a different roster as well, or is football or as you Americans call it soccer not a team sport? Heck even American Football in the NFL seems to utilize a modified Elo system.

well, i was referring to american football (NFL or NCAA which is college level), although football/futbol/soccer could still apply to the point i was making regarding hypothetical team "strength" based off of previous results. you simply cannot make accurate predictions based off of previous results due to inconsistent variables, at least not accurate enough when those guestimations affect rankings so significantly. like i said, that's why NO major team sports competition uses any form of ELO to base standings. the BCS system that determines national rankings for collegiate american football sorta implements a modified ELO system in regards to the computer predictions....but that "computer math" represents only 1/3 of what determines a team's rank. the other 2/3 are the results of human polls where coaches/former players/writers/panelists physically vote based off of their personal observations and analyses of team's performances. and to note, the BCS system is regarded as a majorly flawed system and is despised by over 90% of those who follow the sport

and no the NFL does not use any form of ELO. it's a simple regular season league, single elimination playoffs format.

You do not need to keep all variables constant, as you can take into account inaccuracies in a teams ranking.

but how do you account for those inaccuracies? if you're making predictions, you have to have constant variables, let alone the most important variables at the very least.....participants of a match are by far the most important variable in this case. heck, you could even make valid arguments that ping differences, server differences, and individual's computers and equipment differences can also significantly affect these predictions.

let's look at it like this: everyone knows what happens when you put mentos into a bottle of diet Coke because we've all seen what happened when it was done in the past. now say you ask a scientist to make a prediction of what will happen if he puts mentos into a bottle of diet Coke. of course he would tell you it's ~100% probable that there will be an "explosion". well what happens if you don't keep a variable consistent? what happens if you supply him with mentos and some coffee instead of diet Coke? it's the same thing that happens when you ask a computer to make a prediction that teamX will win/lose the next match based on the results of their previous match when they used players #1,2,4,5,8,10,12,13 but then for this match, they end up using players #2,3,4,6,7,9,11,14 instead. the variables are not consistent. the previously created prediction is invalid. to base rankings off of potentially invalid predictions is a flawed system. the only reason why it works with individual sports (chess, tennis, backgammon, racquetball etc...) is because the main variable is ALWAYS consistent. the player is always the same player (barring physical injury). in team sports, keeping this variable 100% constant isn't possible.

The players in a team do not and should not ever matter in rankings, only the results that the team gets.

well the problem is that for any type of prediction system (or a past results system), those players do matter. they impact the predictions regardless because they are variables in the equation. the players involved in a previous match are part of a process that generates a prediction with the assumption that those same variables will remain unchanged. but what happens when they do change, i.e. different players participating who where not accounted for when the prediction was generated? you've got bad predictions, thus the results of the match and the following changes in standings are also invalid. rankings and standings are two different things. in the case of ELO, they're combined, but the problem is that a teams hypothetical strength (aka rank) directly impacts their position in standings. that would be fine IF the prediction system was legitimate. again in the case of individual based competition, that is possible because you can keep the most important variables constant. however in team based competitions, you can't keep those variables consistent, thus producing inaccurate predictions, invalid rankings and ultimately a flawed standings system.

Here's how I see it then:

A team is a black box. You don't know what is inside and the only thing that matters is what comes out (the result).

If you knew exactly what was inside it and how it worked, you could predict the output with 100% accuracy.

Because you don't know exactly what is inside and how it works, you cannot predict the result, therefore it's reduced to a guess based on a formula which "some guy" decided and that's the unfair part.

that right there is a great way of looking at it and i totally agree with you. with individual sports, you can make legitimate assessments of what's inside the "black box" based off of previous performances because what's in that black box will always be the same (it's the same one person every time). however with teams, you can't always know with 100% certainty what is in the box. one match it can be players #1,2,3,5,6,7,9,10 then the next match it's #4,8,11,12,14,16,18,20 and the next #1,3,6,8,10,14,18,19. because of that you can't make accurate predictions.

the only thing that should affect rankings/standings is the result of the match. you win, you move higher in standings. you lose, you either drop or don't advance (in the case of standard rung systems). having bogus gimmicks like predictions are taking away the simple competitive nature of sport. hence look at the "sports" that use ELO type systems......chess, backgammon, scrabble, video games (mostly 1v1 max 5v5 matches with consistent lineups)......no offense, but those just come off to me as nerdy intellectuals putting way too much thought into something that isn't that difficult to understand. :troll:

bracket tournaments, leagues, knockout, rung ladders. they are simple, "duh" ways to have competitions.