I just did some calculation based on "my" formula. Top 20 players would range between 3200 and 8380, so I believe THE ranking could be divided by 4 (as we have GR, CP, H% and TS). If I did no calculation mistakes, the results could be :

Toto wrote:

Tournament points : I see no reason at all not to take them into accounts. Players not playing tournaments would be given a 1000 score, so they would not get penalized.

Good point.  After all, we're looking for a site-wide picture of skills and achievements. 

i seem to be playing better in the tourneys than normal so i'm not adverse to the idea; tho a nominal score for those that don't play would pull most of my hesitation of adding tourney scores.

Toto wrote:

Tournament points : I see no reason at all not to take them into accounts. Players not playing tournaments would be given a 1000 score, so they would not get penalized.

It is clear that this is not true as long as T score is a fourth input.  Toto what is your formula?

I would argue that T score should not be a factor at all (since we are now outside the realm of public games), but if it is, then T score and global rating certainly should not separate influences, since they measure the same thing from different angles.  What should be used is the average of global ranking and T score weighted by number of games played for each, and use this instead of global ranking.

 I agree it would be very cool to have an "overall skill" stat. We all know that such a figure will never be perfect, far from it. But let's try something to improve M57 idea's.

[img]http://www.makemoneymakemoney.net/huang1.jpg[/img]
[img]http://www.makemoneymakemoney.net/huang2.jpg[/img]
[img]http://www.makemoneymakemoney.net/huang3.jpg[/img]

^  ^ Is that spam?

Also - Toto, will you write out the formula you used?

Ozyman wrote:

Also - Toto, will you write out the formula you used?

It's only M57's formula with log (base 10) of CP instead of CP. To make it work with log, you have to give anybody a minimum of 2 points. 

TP should be set to 1000 for any player who never played any tournament (just like you get a GR of 1000 when you join the site and never played).

So it goes :

log10 ( max (CP;2) ) * GR * HR * TP /1000 / 4

For exemple, for BD, log10 (304) * 2244 * 0.77 * 1417/4000  = 1519 (all my calculations were done with this formula)

Thinking about it again, I believe it would be even more fair to give an extra 2 championship points to any player (as having 0 or 2 points is quite different) instead of a minimum of 2.

So the NEW NEW formula would be :

log10 ( CP+2) * GR * HR * TP / 4000

This last change would be insignificant for top players but would allow a player gaining his 2 first CP to double his score. So yes I believe this is better than max(CP;2).

Also, I advocate that, if such a new overall skill ranking was implemented, it would come with a moving average, highest, lowest, complete history, graphs,... A lot of work for Tom, but I am convinced it would be worth it ;)

Is the proposal that this is added as an additional scoring measure (like H Rating) or that this replaces Championship points as the way in which the Top players on WarGear are judged?

I like the idea of a measure... I don't like the idea of the scoring system getting so complicated that players don't know how the system works.

tom wrote:

Is the proposal that this is added as an additional scoring measure (like H Rating) or that this replaces Championship points as the way in which the Top players on WarGear are judged?

I like the idea of a measure... I don't like the idea of the scoring system getting so complicated that players don't know how the system works.

IMO, it should be an additional scoring measure, but it should be highlighted as being the main one (first on the Home page, the first tab in the Rankings page,...).

I guess it's not that complicated if the Help page is giving the formula for those who want to understand how an overall global ranking is calculated.

I like things the way they are.  In fact I would prefer even more top 10 lists (e.g. top team player) to give people multiple goals to strive for.

But, if the single score is the way things are going I think Toto's is dominated by a players GR.  If BD gained 20 CP's his new score would be 1536, while if he gained 20 in GR his new score would be 1533.  I think we all would agree the former is much harder to do then the latter.

Amidon37 wrote:

I like things the way they are.  In fact I would prefer even more top 10 lists (e.g. top team player) to give people multiple goals to strive for.

But, if the single score is the way things are going I think Toto's is dominated by a players GR.  If BD gained 20 CP's his new score would be 1536, while if he gained 20 in GR his new score would be 1533.  I think we all would agree the former is much harder to do then the latter.

I am fine with top team players, it would be nice too (http://www.wargear.net/forum/showthread/1849/Team_Rankings).

I agree with you that introducing logarithm has strong bad effects on the very top championship players, especially BD who is so high. But that was also the idea, as you can add CP's with the growing number of boards, while it's more difficult to improve your GR when you are far above your opponents.

On the other hand, an average player with 0 CP,  a GR of 1500, an HR of 60%, a TP of 1250 would get a 85 score. Gaining 20 points in GR would give him a +1 increase. Gaining 20 CP would give him +293 !!! It seems very volatile but I never saw anyone gaining 20 CP at one time. And the moving average will do it's job.

So the GR is not always dominating... Players who want to upgrade this new ranking will have to choose what is the most rewarding strategy for them. 

Is the proposal that this is added as an additional scoring measure (like H Rating) or that this replaces Championship points as the way in which the Top players on WarGear are judged?

I like the idea of a measure... I don't like the idea of the scoring system getting so complicated that players don't know how the system works.

I think this is an interesting discussion, but I don't think the idea of a combined rating makes sense since there is really no way to agree on the relative worth of CP, GR and HR.

i moved up with the new proposed math formula... i approve!

but for sure i would love to see team players rewarded... even fun stats that could help someone pick out a stranger for a partner... in team transfer/placement games, how many does s/he give away compared to the average? (ie selfish or team player?) win % of course. currently my team trophys dont show up as a tourney victory, so other than the trophy case itself, it seems less recognized.

Things could get complicated, but if anyone cares, there can be a sticky topic, or links to the explanation if people care. if not, no harm done adding something.

BlackDog wrote:

I think this is an interesting discussion, but I don't think the idea of a combined rating makes sense since there is really no way to agree on the relative worth of CP, GR and HR.

I tend to agree with BlackDog. I was kind of viewing the debate as a diversion.  There's certainly a lot of subjectivity involved when you try to weight different measures in an effort to come up with a overall value, and the way CP is calculated throws a giant monkey wrench in the equation.

That said, on with the idea. I'm beginning to warm to the log idea, but I have a tweak to propose.

log₁₀(CP+2) is not enough.  The disparity between 0 and 300 is still huge when you consider that it's a multiplier. 

log₁₀(CP+10) is still not enough but at least it starts everyone with a multiplier of 1, not unlike the way a player who doesn't participate in tournaments will have a tournament rating of 1000.

log₁₀(CP+100) is too much.  Given the current range (~0-300), it yields a 2.6 : 2 ratio between top and bottom.

Consider also that we don't know where or even if there is equilibrium with CP. With more and more boards will it become possible to hit 500?  ..1000?

I propose backing out the zeros and using either the mean or the median CP score.  Right now, the median is approximately 16, giving us log₁₀(CP+16).  Better.  I'm guessing the mean is closer to 25, which would be my choice at log₁₀(CP+25)

One argument for this method is that the number will scale with the range of the CP score over time.

my head hurts like i had too much beer... then too much mead... then too much ale

M57 wrote:

BlackDog wrote:

I think this is an interesting discussion, but I don't think the idea of a combined rating makes sense since there is really no way to agree on the relative worth of CP, GR and HR.

I tend to agree with BlackDog. I was kind of viewing the debate as a diversion.  There's certainly a lot of subjectivity involved when you try to weight different measures in an effort to come up with a overall value, and the way CP is calculated throws a giant monkey wrench in the equation.

That said, on with the idea. I'm beginning to warm to the log idea, but I have a tweak to propose.

log₁₀(CP+2) is not enough.  The disparity between 0 and 300 is still huge when you consider that it's a multiplier. 

log₁₀(CP+10) is still not enough but at least it starts everyone with a multiplier of 1, not unlike the way a player who doesn't participate in tournaments will have a tournament rating of 1000.

log₁₀(CP+100) is too much.  Given the current range (~0-300), it yields a 2.6 : 2 ratio between top and bottom.

Consider also that we don't know where or even if there is equilibrium with CP. With more and more boards will it become possible to hit 500?  ..1000?

I propose backing out the zeros and using either the mean or the median CP score.  Right now, the median is approximately 16, giving us log₁₀(CP+16).  Better.  I'm guessing the mean is closer to 25, which would be my choice at log₁₀(CP+25)

One argument for this method is that the number will scale with the range of the CP score over time.

Fully agreed that +2 is not enough. In fact it's the minimum for the calculation to work. Trying to find a more suitable number is a brilliant idea.

I also thought about a +10 for the same reason you explained.

I don't know about using the median or the mean CP. Both would obviously work perfectly but could be seen as a bit complicated, according to Tom's objection. 

So, all together, I would go for a +10. 

BlackDog wrote:

Is the proposal that this is added as an additional scoring measure (like H Rating) or that this replaces Championship points as the way in which the Top players on WarGear are judged?

I like the idea of a measure... I don't like the idea of the scoring system getting so complicated that players don't know how the system works.

I think this is an interesting discussion, but I don't think the idea of a combined rating makes sense since there is really no way to agree on the relative worth of CP, GR and HR.

IMO the idea of a combined rating is one of the best ideas we had to make this site better.

I believe we are very close to have reached a good formula ?

BlackDog wrote:

I think this is an interesting discussion, but I don't think the idea of a combined rating makes sense since there is really no way to agree on the relative worth of CP, GR and HR.

+1

But, continuing with the formula discussion:

Am I the only one here who has a big problem with tournament games being given weight here?   I feel like there is a reason that we have a separate ranking system and statistics for ranked games and tourney games (and I recall this being discussed at great length).

I play lots of games in tourneys that I wouldn't normally play, partly because I know they have no affect on public game rankings.  And although I generally try not to play to advance rather than win, it is a legitimate strategy in tournaments that has no corollary in ranked games.

BD