A new skill level

[size=16px]Improved Rating – 2nd Attempt[/size] Winning a game would earn you rating in two ways: [i]<Improved Rating Change> = <Rating Change Based on Game Size> + <Current Rating Change> [/i] Where the <Rating Change Based on Game Size> will be given by: [i]<Rating Gain Based on Game Size> = (Value of 2p game x number of people in game)/2[/i] [i]<Rating Loss Based on Game Size> = (Value of 2p game x number of people in game)/(Number of people in game – 1)[/i] This would work similarly as what I described in my first post, but it would also incorporate the current rating system (which was designed to address the differential value of beating people of different rating). This now becomes VERY specific (albeit cheating somewhat by incorporating the current rating, I admit!) Let’s look at three of examples with some maths to make sure this works. Let’s say winning a 2-p game gives you 10 rating (this number would need to be set carefully, but is in the right ballpark): [spoiler]Winning a 2p game would give you the rating change you’d normally get depending on who you beat, plus 10 rating Losing a 2p game would give you the negative rating you’d normally get depending on who beat you, minus 10 rating. Winning a 4p game would give you the rating change you’d normally get depending on who you beat, plus 20 rating Losing a 4p game would give you the negative rating you’d normally get depending on who beat you, minus -6.67 rating. Winning a 9p game would give you the rating change you’d normally get depending on who you beat, plus 45 rating Losing a 9p game would give you the negative rating you’d normally get depending on who beat you, minus 5.63 rating.[/spoiler] To recap, this improved rating does the following: 1) Value winning games based on the a-priori probability of winning assuming everyone was equally skilled. This is missing from the current rating system. 2) Differential value of winning against people who varying skill compared to self (using current rating system) 3) It’s rating-neutral like the current rating system (the sum of all the rating changes in any given game is zero.) [spoiler]Another possibility would be to remove 3) above, so that there is more rating gained than loss. At the moment, high ranked players have a high chance of losing rank. Some of the very high ranked players play very little. One could change this by ensuring that overall, there is a tiny bit more rating gained in any given game than it is lost. Even a very small difference would make people want to always keep on playing. This could be achieved by: <Rating Gain Based on Game Size> = ((Value of 2p game PLUS a small amount)x number of people in game)/2[/spoiler]

I can see how it works, but what exactly does this add? It just gives even you more points if you play large player games - but this already happends, as there are more players you beat, so more players who lose rating and give you some. So what exactly does this improve?

Fair question. I feel it’s a large improvement, though perhaps a subtle one. Let me try to better explain why that is - with my apologies for the length of the post! The rating on this website is peculiar, in that it is shared across all game types (not a bad thing!). A few posts ago I mentioned I thought we would find it hard to agree on what skill was, since there are many different game types. I’m assuming we want the rating system to single out people who are skilled at the games they play. A capitals specialist who wins more than their fair share of games, will have high rating. A specialist in deathmatch in fog with unlimited fortification who wins more than their fair share of games will have high rating. A specialist in deathmatch in the sun with chained fortification who wins more than their fair share of games will have high rating. Etc Now, that’s one side of how it works. The other side is player skill: if you only play against low-ranked newbies, you might win more than your fair share, but you’re not going to get much rating. As you say, ‘winning a game against experienced players is very hard and there should be a big difference there’. The problem as I see it is that the current system limits the rating you’re able to get based on the kind of games which high raters engage in. Now this might start of by chance: more high raters deciding to play X type of game. But once most high raters are playing that type of game, it will be hard to reach very high rating unless you play *that* game type (with those high raters you’re competing against). Suppose none of the people in the top 20 ranks like playing sunny deathmatch capital games. Suppose Bob really likes 6-p sunny deathmatch capitals. He’s very good at it, and wins about a third of the games he plays (so twice as many as the a-priori probability all things being equal). Because of Bob’s skill, he finds his rating goes up and up… But eventually he gets stuck. He realises that when he loses a game he loses twice as much rating as he wins when he wins a game. So even though he wins games twice as often as other people, due to the fact that none of the high raters play the game bob plays, Bob’s rating gets stuck. Now this is an extreme example: most people here play different kind of games. But not ALL kinds og games. Some mostly play long term games. Others mostly play live games. Some play mostly sametime ones. Others play mostly consecutive turn ones. Some favour fog games. Others only play sunny ones etc. Also: I don’t think either the system I’m proposing, nor the current rating give more points when you play larger games (I guess they do in absolute terms, but they only give you proportionally more points, which is fair enough). In the same way, the system I propose doesn’t give you more points overall: it’s still point-neutral, it gives you as many points as the other people in the game lose. You just win a bit more when you win, and you lose a bit more when you lose. What the system I propose does is to give you some points for winning a game (or make you lose points for losing a game). These points are fixed, and completely independent of the people you’re playing with. Winning a game *should* have some value in and of itself. What this would mean for Bob is that he’d be able to get more points, even if few of the high raters are playing with him. It would also mean that high raters wouldn’t need to avoid low rated foes as much as they currently do. The system I’m proposing would reward SKILL more than the current system does. And it does that however you define skill. Incorporating the current rating system, it also does everything the current rating system does.

Am I correct in thinking that you mean: my rating system is better for beating newbies?

No. That's not what I'm saying.

What I'm trying to say (and obviously failing at) is that I think skill, however you define it, is made up of two parts: 1) Beating skilled players. This part is covered by the current rating system. 2) Winning more than one's fair share of games. This part is not covered by the current rating system. I'm sorry I'm not doing a better job of explaining myself!

You are right in saying that the current skill level has the property that just winning games is not enough to get a really high rank, for that you need to play against better players. I think however that that's a good thing. How can you be a general if you never play high ranked players? Even though this is a good thing IMO, I do agree that it has a negative side effect of high ranked players not wanting to play newbies anymore. Rank is one reason, their tendency to (unknowingly) suicide is another. I'm not sure though if we should fix that by having a different rating system, or maybe just allow games without ratings (which is on our todo list).

I definitely agree with you that one should continue to have to play against good players to get good rating. I don't think the modification I was suggesting changed that - it just valued winning games a bit more (but still less than it values playing high ranked players). I think we are also in agreement that one shouldn't be able to get goot rating by beating newbies. I'm not sure if high ranked people avoid newbies because of rank: as you say, potential suicide is probably more of a factor. Personally I'm wary of playing with newbies because they are so unpredictable: they might well give the game to someone else. While I'd be happy playing with someone new at classic risk (I'm assuming they know how if they are here!) I would feel compelled to give them advice before playing with them at capitals. I don't think newbies not mixing with other players is a huge issue. I think this mostly sorts itself out given newbies become non-newbies with a bit of time :) I think high ranks not mixing with experienced players with significant lower rating is more of an issue. I like the idea of having pointless games: I think the main advantage of that one would be to allow you to try games you're less familiar with. It would also allow you to try games where skill plays a less significan role in winning (for instance I sometimes enjoy playing fog deathmatches, but I find these ones are very dependent on luck). I'm not certain this would solve the problem: I expect high rankers would continue to play their high rank games the same way (as well as maybe ocasionally try out different gametypes in pointless games). A couple of different ideas: 1) Would it be possible to have a system, similar to rank limiting games, but instead of not letting in people with rank lower than X, those with a rank lower than X don't contribute to ratings? So say you have a 3-p game with a rank limit of 3k. Two of the people in the game have 4k rating. The third only has 2k. Whatever happens in the game, the third person wouldn't contribute to rating (they neither win or lose any rating in that game). This might have to be a premium only setting, as it's very similar to rank-limiting games. 2) Maybe people new to the site shouldn't contribute any rating for the first X games? To be honest, I didn't find rating was a thing until about a month after I started playing - I so I doubt new people would mind/notice. This would avoid new people being farmed for rating too (though I don't see that happening as much anymore).

Well u can become a General without ever playing a player(s) with a higher rank then what your at...would take many many games..but is possible.. Imo to have a accurate rating of skill will require a system that takes into account of those who are above or below yr current rank.this I'm sure wouldn't be a easy to do..1 to figure out a mathematical equation an 2 write the code to implement it,the code prolly being the hard part. Although this would be cool to have.. But is it worth the time an effort to do? Their are a many things proposed or in the works I think(from wat I C reading the forums) are far more important to make happen an to make this site a cut above the rest.. From wat I read..the programmers to do list is huge already I think replacing rating with % of wins overall is (1) accurate (2) easy to implement for code (3) the players can research the other players game history an C if he is good or just a noob basher. Another thought..wat about only getting a rating from those above yr rank..no rating from those below...so if yr rating is 0 an you play lesser ranks.you remain at 0..only way to up yr rating is to play higher ranks..so on yr homepage it could show as... % of wins an your rating against higher ranks...so if player is nothing more then a noob killer...this would show on his stats

There is one thing that I will never do: having some players get or lose rating or any sort of skill measure, and other players in the same game have nothing to lose and therefore can suicide as they want. If you don't lose rating for losing a game, you are much more likely to "just try" as "you don't lose anything anyway".

I hadn't considered that! It's great to talk about these things with you - you've obviously thought about this a lot :)

Yeah, we tried to create a new skill system before once, but that was only in the admin forums I believe. Didn't really found a new skil llevel, unless you want to use something like Trueskill or Elo, but that's probably a lot of work and hard to explain to people. That's the main good thing about the current rating system we have - it's easy to understand how it works, and quite accurate as well (though defenitely not perfect). And thank you for suggesting these things. I (we) do not accept everything, but it's a good thing to think about it.

I did not mean yr rating never goes down..if should be affected by both higher an lower players..but only goes up when u defeat higher ranks..sorry for not being more clear on wat I was trying to say..I'm sure Matty ,you an those who create an run the site have beat this subject to death..I was just throwing out some ideas :)

Why the need to try to quantify skill? Is it just to weed out rank hiders? The rating system is probably the most accurate guide we have for someones skill. There is only one way to test skill and that is to play others over and over again. Then you get to see over many games not just when they win from someone elses fail but when they plan, execute and succeed in ripping you to shreds, not only taking into account the possibility that someone else will fail but that they won't need a fail before them to unleash their plans. If you can come up with a number to express this then you will have your skill rating

Create a weighted rating based of an unweighted rating. Use the standings list to generate each. [b][u]Unweighted: [/u][/b] [b]A [/b]= #of "is beating" + 2X "is conquering" +3X "is dominating" etc. (where the player is the one doing the beating) is tied # ignored [b]B [/b]= #of "is beating" + 2X "is conquering" +3X "is dominating" etc. (where the opponent is the one doing the beating) [spoiler=Practical Note]So go to any player's "standings" page, A is a weighted sum of all the players someone is beating and B is a weighted sum of all the players which are beating that player. And all the ties are ignored. So the more at the top of the list, the better the player.[/spoiler] Unweighted we will call U; where [b]U[/b] = A/(A+B) Now A is strictly self correcting for game selection but not opponent strength. If you play 6 player games, you get to "beat" 5 players, and are rewarded in this ration proportionally to the number of opponents you beat. B is proportional to win / loss ratio, since a loss is a loss, and will always just be 1 player each loss no matter game type or number of players. (team games will distort A high, since you "beat 2, 3, 4 players" while essentially playing heads up - so that's one flaw, but a minor one. The big issue is this does not account for how good the players are. Hence I called this the unweighted ratio. [b][u] Weighted Ratio:[/u][/b] Same exact calculation as unweighted but each opponent is multiplied by their unweighted Ratio [b]C[/b]=#of "is beating"(each multiplied by their own U) + 2X "is conquering"(each multiplied by their own U) +3X "is dominating"(each multiplied by their own U) etc. (where the player is the one doing the beating) [b]D [/b]= #of "is beating"(each multiplied by their own [1-U]) + 2X "is conquering"(each multiplied by their own 1-U]) +3X "is dominating"(each multiplied by their own [1-U]) etc. (where the opponent is the one doing the beating) [spoiler=Practical Note]So now that we have established an unweighted ratio for each player between 0 and 1 we will use that to give credit for beating good players proportional to their skill being good. We will also credit the player for losing to good players with 1-U making the losses take up less of the denominator inversely proportional to the unweighted ratio of opponents who beat you[/spoiler] Weighted Ratio is W where [b]W[/b]=C/(A+D) Finally you get two stats to compare: Skill Rate = W x 100 [independent of games played] Skill Points = W x #of games played [skill and games played consdidered] [spoiler=Side Note]This is the simplest ELOCHESS / Sagarin ranking system I could think of to not be impossible to figure and calculate regularly, unlike the old skill points which would be done now and then because it was a pain to calculate[/spoiler]