Forged Alliance Forever

Re: Global Rank via Score, not Win/Lose

2015-01-06T23:21:02+02:00

Wow! 5 pages of replies.

I think the current system is as fair as it can be and that codepants has also a right to be worried about distortions in rating.

The whole premise of the current system is that after enough games every player will have a statistically accurate and fair rating. Well, this is true or can only be true to an extent and distortions will always exist. Why?
1. Not all players play at the same frequency. There's a comparison of player's skills inadvertently and if all are not compared for the same number and quality of games inaccuracies will ultimately occur. If someone likes to play and does not play only with a strategy to increase rating, it's certain he/she will be matched more often than not with more skilled players. Result?
Lower rating. Is he/she less skilled ? of course not.

2. It is known that I don't agree, global rating to be based on custom games at all, only on ladder. And there is a reason for this. In ladder you will lose for a relatively restricted number of standard reasons:
a. You play against a clearly superior player
b. You have been careless and absent minded
c. You had a bad day and your skills are dull.

In custom games this list is overly expanded. Want an example? You play with allies who are arrogant little pricks that want everything to go their way. Either you won't comply and they will rage quit, or you will and fall with them and their "zuper duper" strategic plan.

Want another? Cheese is easier to do and harder to detect, especially in big maps with a lot of players.

Player selection is agreed between people and not objectively by a machine. This means "not balanced" games will occur.
There's a wider player base. You often don't really know who you are playing with.
There is a far greater possibility to play against "map specialists" who always play(and legitimately) their maps of preference.

And many others.

Solution? Understand how rating works and play mainly for your pleasure.
Avoid games with players who boast too much, or who think better rating automatically makes
them better players.

Statistics: Posted by prodromos — 06 Jan 2015, 23:21

Re: Global Rank via Score, not Win/Lose

2015-01-06T20:58:55+02:00

codepants wrote:
Hawkei wrote:Okay, well lets 'theorycraft' this scenario. ...

(and other people with similar responses)

Let me try again my latest post about the air player who gets overrun. I'll make it even simpler.

- There is a limited amount of mass available.
- The team that uses their mass better than the other team wins.
- Let the amount of mass your team uses effectively be x+y=z where x is your teammates and y is you.
- Let the amount of mass the other team uses be q.
- Therefore, it is likely that if z > q you win, and if q < z you lose.

This isn't always the case (snipes, etc.) but I'm betting most people would agree (and indeed, it's been said in this thread, even by some of the higher-ups) that if you use your mass "better" than your opponent you are likely to win.

If x is sufficiently small then y cannot possibly make up for it. That is the part you all seem to be missing. There is no magic that gets you more mass than the other team if your teammates do something stupid. There is a limited amount of mass and mass extractors and even if you are good at getting it, it is all too easy to do something stupid with it (eco too hard, build units and misuse them, etc).

So your team loses and it's out of your control.

Elaborate.

Alright, I'll try to address your points as directly as I can. The problem with your argument is that over time, those discrepancies will even out. Taking your example, consider Z-Y, which is how well your opponents play less your teammates. If X>Z-Y, you will win, if it's less, you'll lose. Sometimes, Z-Y will be higher than expected, but at other times it'll be lower. Points are rewarded according to the team's performance vs. the expectation. You are complaining that your teammates always let you down, but it's highly unlikely that your teammates underperform their ratings more often than your opponents do. Why would that be the case? Of course, there are examples where your 1300 teammate gets crushed by an 800 and there's nothing that you can do, but there are also times when a 1000 teammate plays great and duels a 1400 to a stalemate, allowing you to win. Not every game you win or lose comes down to your direct control, but these things even out.

Furthermore, as many of the posts have alluded to, the point of the game is to win. If a player is good at the game, his/her team will win more often than it's expected to based on the ratings, and that person's rating will go up. It continues going up until it reaches a point at which the player is losing and winning about as much as he/she is expected to. No alternative metric can measure your ability to win the game better than how much you win the game relative to the skill of your teammates and opponents.

There certainly may be issues with the methodology of combining ratings (summing?) in team games, as IceDreamer alluded to. I'm actually interested in looking into that. But that's a separate issue from using an alternative metric to W/L.

Statistics: Posted by sasin — 06 Jan 2015, 20:58

Re: Global Rank via Score, not Win/Lose

2015-01-06T19:58:06+02:00

Hawkei wrote:
Okay, well lets 'theorycraft' this scenario. ...

(and other people with similar responses)

Let me try again my latest post about the air player who gets overrun. I'll make it even simpler.

- There is a limited amount of mass available.
- The team that uses their mass better than the other team wins.
- Let the amount of mass your team uses effectively be x+y=z where x is your teammates and y is you.
- Let the amount of mass the other team uses be q.
- Therefore, it is likely that if z > q you win, and if q < z you lose.

This isn't always the case (snipes, etc.) but I'm betting most people would agree (and indeed, it's been said in this thread, even by some of the higher-ups) that if you use your mass "better" than your opponent you are likely to win.

If x is sufficiently small then y cannot possibly make up for it. That is the part you all seem to be missing. There is no magic that gets you more mass than the other team if your teammates do something stupid. There is a limited amount of mass and mass extractors and even if you are good at getting it, it is all too easy to do something stupid with it (eco too hard, build units and misuse them, etc).

So your team loses and it's out of your control.

The point is not, "Oh, you didn't help your teammates" or "Oh, you didn't make up for your teammates' weaknesses." The point is that your team, as a whole, did not use their limited resources better than the other team. Regardless of how well you play, regardless of how well you compensate for the shortcomings of your allies, it is always possible for x to be sufficiently low so that y cannot make up for it. You can, essentially, play as well as a 2600 rank player, but if the other team, as a whole, uses their mass better than your team, you will lose points for it -- even if you, as an individual, are better than them.

Reaper Zwei wrote:
Code, your whole argument is that you want your skills to be judged individually, but frankly your playing the wrong game mode for that. Individual skill is important but it is a distant second to the ability of your team to work together.

Actually I don't necessarily want skills to be judged individually. Global ranking is tied to an individual player, not a team. If global ranking was assigned to teams of players, then I'd be fine with the way it is now. Yes, absolutely punish the team as a whole for having lost -- lower the sum of the team's rankings for not working together. But the rank that is assigned to an individual should not go down because their allies did something stupid.

It is not the game you play, it is where the ranking is assigned.
- Code, Reaper, and Blodir, as a team, have x rank. They lose. X goes down. Awesome. Agree completely.
- Code has x rank, Reaper has y rank, and Blodir has z rank. They lose because Code built only scouts on repeat and his mirror used his mass more effectively. Code's rank should go down, Reaper's and Blodir's should be adjust according to how they played.

Blodir wrote:
The goal of the game is to win the game, not gain as many points as possible. Rating is an accurate representation of this. If rating was measured by how many points you got in a game then nobody would try to win, but instead try to gain as many points as possible.

Then, as I have said a number of times now and everybody has ignored, we should either do it another way, or redo the scoring system such that it can be used for ranking purposes. Shouldn't score be synonymous with rank? In every other sport/game I've played, the player/team with the highest score wins...

I am not saying my idea is perfect, I am simply saying I think the ranking system can be done better.

Vee wrote:
That is a problem of education, not a problem of the rating system.

Elaborate.

Statistics: Posted by codepants — 06 Jan 2015, 19:58

Re: Global Rank via Score, not Win/Lose

2015-01-06T06:31:44+02:00

Hawkei wrote:
One essential component of increasing one's rank is understanding how that rank is determined. Equipped with such knowledge players are enabled to make better choices which improve their rank.

don't play to improve rank, play to be a better player. If you try and game the system, you'll always be a bad player, if you try and learn the game, your rank will increase as it should.

Statistics: Posted by Aulex — 06 Jan 2015, 06:31

Re: Global Rank via Score, not Win/Lose

2015-01-06T06:18:44+02:00

Ze_PilOt wrote:
It should not be transparent but completely opaque. People doesn't need to know how the matchmaker works as long as it is.

Yes... Well... I think we have here a difference in communication style. I don't agree.

Firstly, because you cannot prevent a person from knowing something if they are determined to know it. One essential component of increasing one's rank is understanding how that rank is determined. Equipped with such knowledge players are enabled to make better choices which improve their rank.

Secondly, there is one important thing which all players need to understand about the system. That it is fair. This understanding is not fostered by a lack of transparency.

Statistics: Posted by Hawkei — 06 Jan 2015, 06:18

Re: Global Rank via Score, not Win/Lose

2015-01-05T15:29:36+02:00

It should not be transparent but completely opaque. People doesn't need to know how the matchmaker works as long as it is.

Statistics: Posted by Ze_PilOt — 05 Jan 2015, 15:29

Re: Global Rank via Score, not Win/Lose

2015-01-05T01:51:43+02:00

Forgive me for saying. But your query was kind of inviting that response. It was like saying,

"well how come this topic comes up every 6 months? Do you think we're stupid?"

Well the available evidence would seem to imply an answer to that question... Unfortunately I shouldn't have given you that answer because it contravenes FAF forum policy guidelines. I apologise for this.

As to the main point at hand. The rating system is perfectly transparent and easy for most people to understand. In rough terms it is much like a KDR. But the critical difference is that it determines the value of the kill based on the difference in skill between the two players. So in basic terms, this means that beating ZLO is given greater value than beating some random noob... Which is precisely how it ought to be.

You have to understand that with a pure KDR most games would go for 5 seconds, and the Pro's would become very frustrated. Because, everyone would see who they were up against, determine that they had minimal chances and that the game was not worth playing. They would then promptly quit and look for an easier opponent. Having a shifting value system, such as true skill, means that however mismatched the game is, it is always worthwhile to play the game. The more mismatched the game becomes, the greater the opportunity to increase score and upset the higher player's ranking.

Of course a higher ranked player will play the mismatch. Because, quitting the game would be accepting defeat. They are forced to defend their title. So they WILL fight.

Statistics: Posted by Hawkei — 05 Jan 2015, 01:51

Re: Global Rank via Score, not Win/Lose

2015-01-05T01:28:09+02:00

It's that kind of arrogant response that keeps this topic alive. Remember as I said before, unless the ladder is transparent there will be those that question it and those that feel it favors the select few.

Resin

Statistics: Posted by Resin_Smoker — 05 Jan 2015, 01:28

Re: Global Rank via Score, not Win/Lose

2015-01-05T00:39:30+02:00

Resin_Smoker wrote:
If the above was the case then why do we keep revisiting this topic every six months?

Resin

I can't give you an answer. I wasn't the one who brought this up. Maybe because there are some people who are behind the bell curve?

Statistics: Posted by Hawkei — 05 Jan 2015, 00:39

Re: Global Rank via Score, not Win/Lose

2015-01-04T21:29:51+02:00

That is a problem of education, not a problem of the rating system.

Statistics: Posted by Vee — 04 Jan 2015, 21:29

Re: Global Rank via Score, not Win/Lose

2015-01-04T19:42:21+02:00

Blodir wrote:
The goal of the game is to win the game, not gain as many points as possible. Rating is an accurate representation of this.

Ask a few of the newer players what they think of it, I doubt most of them would agree.

Statistics: Posted by Resin_Smoker — 04 Jan 2015, 19:42

Re: Global Rank via Score, not Win/Lose

2015-01-04T19:42:40+02:00

The goal of the game is to win the game, not gain as many points as possible. Rating is an accurate representation of this. If rating was measured by how many points you got in a game then nobody would try to win, but instead try to gain as many points as possible.

Statistics: Posted by Blodir — 04 Jan 2015, 19:39

Re: Global Rank via Score, not Win/Lose

2015-01-04T19:37:32+02:00

Hawkei wrote:
Resin,

If we adopted WLR as the definition of score we would be introducing luck as a very large factor in determining rank. It would also effectively discourage lower skilled players from participating in the ladder. I can think of no better way to force people away from ladder than this.

The change will result in shit ladder dynamics and cause people to stop playing it, en mass. Because it will remove the inherent dampening factors and cause exponential increases in the score of higher players. While at the same time deflating the scores of lower ones. Meaning that the definition of what constitutes a particular scored player will change over time. The score system would become totally meaningless.

Instead of those high score players being the ones who are actually good at the game. They will consist of players who work the system by "Search Stalking" other opponents. By trying to determine who is and who is not searching at a particular time. Hence, by avoiding higher ranked players someone can effectively inflate their own score. I don't think this is what we want.

If the above was the case then why do we keep revisiting this topic every six months?

Resin

Statistics: Posted by Resin_Smoker — 04 Jan 2015, 19:37

Re: Global Rank via Score, not Win/Lose

2015-01-04T18:08:52+02:00

Resin,

If we adopted WLR as the definition of score we would be introducing luck as a very large factor in determining rank. It would also effectively discourage lower skilled players from participating in the ladder. I can think of no better way to force people away from ladder than this.

The change will result in shit ladder dynamics and cause people to stop playing it, en mass. Because it will remove the inherent dampening factors and cause exponential increases in the score of higher players. While at the same time deflating the scores of lower ones. Meaning that the definition of what constitutes a particular scored player will change over time. The score system would become totally meaningless.

Instead of those high score players being the ones who are actually good at the game. They will consist of players who work the system by "Search Stalking" other opponents. By trying to determine who is and who is not searching at a particular time. Hence, by avoiding higher ranked players someone can effectively inflate their own score. I don't think this is what we want.

Statistics: Posted by Hawkei — 04 Jan 2015, 18:08

Re: Global Rank via Score, not Win/Lose

2015-01-04T16:13:35+02:00

Hawkei wrote:
Resin_Smoker wrote:Rank should be solely based off of a players win / loss ratio. (WLR)...

No. It should be based off a WLR which is normalised for the differential in player skill. Such that wins against superior opponents carry greater weight than wins against inferior ones. Which is precisely the system which we have now

Without normalisation for player skill differential a WLR is meaningless.

It is also the reason why a seasoned player will not play with a new player.

Statistics: Posted by Resin_Smoker — 04 Jan 2015, 16:13