Classicists League Table

General stuff.

Moderators: JonS, Buachaille, Fatmo

Re: Classicists League Table

Postby InterMPC » 10 Nov 2010, 02:00

Like most rating systems across the world, I think the rating system should be based on performance, not just skill. Some might say that's semantics however;

Performance means you have to work for your position - you have to go out and perform. You can't rest on your laurels.

Just because you have all the skill in the world doesn't mean you perform the best every game. Consistency is important. Player A might display more skill in 1 game than player B, however Player B is more consistent. Who deserves a higher ranking?

If you play lots of games but don't perform, your results should go backwards. So a system where, you lose you go down, you don't play you stay the same, and if you win you go up makes sense.

I guess my concern is about players maintaining a high rating, even if they haven't played a game in months. In my opinion, that means they haven't performed. I'm not saying we should give incentive to play 5 games at a time, but at least make sure people have at least 1 active game going, or if they just finished have a small window to allow them to join another. What's the best way to encourage this without encouraging people to play 20 games at once increasing the chance for an NMR.

Think of tennis as an example, If I'm ranked number 1 and then don't play for a year should my ranking remain? Or Imagine a heavyweight boxing champion remaining the champion without any fights.
InterMPC
 
Posts: 2225
Joined: 12 Jul 2010, 09:22
Class: Diplomat
All-game rating: (1000)
Timezone: GMT+10

Re: Classicists League Table

Postby QueenOfHearts » 10 Nov 2010, 04:48

What about a streak included? ie. W2 or L1.... That could take care of consistency. Or even a record, with draws being counted as wins.

Also, as to the skill thing, what bout just a running score? I mean,t hats the most logical solution.
`·.„¸¸¸„‹•*˜¨˜°*•›„¸¸¸„.·´QoH`·.„¸¸¸„‹•*˜¨˜°*•›„¸¸¸„.·´
`·.„¸¸¸„‹•*˜¨Always played, never won!˜°*•›„¸¸¸„.·´

Thanks to SidneyKidney for my sexy avatar. It has a copyright on it.
User avatar
QueenOfHearts
Sponsor
Sponsor
 
Posts: 4650
Joined: 18 Jun 2009, 19:44
Location: (F) Mos (sc)
All-game rating: (1000)
Timezone: GMT-5

Re: Classicists League Table

Postby fellowes » 10 Nov 2010, 05:25

InterMPC wrote:Like most rating systems across the world, I think the rating system should be based on performance, not just skill. Some might say that's semantics however;

Performance means you have to work for your position - you have to go out and perform. You can't rest on your laurels.

Just because you have all the skill in the world doesn't mean you perform the best every game. Consistency is important. Player A might display more skill in 1 game than player B, however Player B is more consistent. Who deserves a higher ranking?

If you play lots of games but don't perform, your results should go backwards. So a system where, you lose you go down, you don't play you stay the same, and if you win you go up makes sense.

I guess my concern is about players maintaining a high rating, even if they haven't played a game in months. In my opinion, that means they haven't performed. I'm not saying we should give incentive to play 5 games at a time, but at least make sure people have at least 1 active game going, or if they just finished have a small window to allow them to join another. What's the best way to encourage this without encouraging people to play 20 games at once increasing the chance for an NMR.

Think of tennis as an example, If I'm ranked number 1 and then don't play for a year should my ranking remain? Or Imagine a heavyweight boxing champion remaining the champion without any fights.


Yes, but you can do this with a 0-sum method (i.e. skill focused), with added point-injections for playing. One method of doing this that I'm familiar with: Starcraft II uses this principle (I think) pretty well. Basically, it's normal ELO, but then you get something like 1 point added to your "bonus pool" every two hours (this number probably would need adjustment for Diplomacy). The catch is, you only get the bonus pool when you gain points by winning (you can never lose points to the bonus pool). And you can only take the number of points that you won.

So say I have an ELO of 100, and 20 bonus points. I will play as an ELO 100, and gain and lose points as such. Say I now have a victory that nets me 12. Now I add 12+12 to my ELO, which is now 124, but my bonus pool is reduced to 8. Another 12 point win without increasing the pool will give me 20 (ELO 144), as I get 12+the remaining 8.

This addition should fit with your point 5, since you could just go to excell every month and run a simple function to increase the bonus pool of every player by X. It should also satisfy 3, to a large extent, and I completely disagree with point 4 (yes the ELO does not require 0 sum, but then it ceases to be a very accurate measure of skill. If each game on average nets an increase, sevenoaks will be at the top of the list. We can't have that... besides the fact that he'd never get accepted :roll: )

Combining this with option 1 satisfies your points as follows:
1. Solos get more than 3x a 3 way draw, 3 way draws get more than 4/3x 4 way draws, etc., encouraging going for the solo, and reducing players in draws.
2. The win/loss ratings are based on ELO, thus meaning beating up on n00bs/morons is less valuable to your point total. The only thing is, getting (say) a 7-way draw even if you are favored, under this system, will not hurt you (It won't help you either, though).
3. You only get the "free" bonus-pool points if you play, encouraging players to keep playing (and winning).
4. Again, I don't think this is a good point. This system preserves 0-sum behavior, which I and some others consider important.
5. Shouldn't be too bad to code.
6. Not that hard of a system. I could write the official summary if people want.
7. (mine) Preserves "a loss is a loss" behavior.
fellowes
Premium Member
 
Posts: 352
Joined: 23 Jan 2010, 04:43
Class: Ambassador
Standard rating: (1177)
All-game rating: (1185)
Timezone: GMT-8

Re: Classicists League Table

Postby InterMPC » 10 Nov 2010, 07:24

I believe that if sevenoaks was still around he would certainly be near the bottom of the list.

I agree with you that you can't have people who play 30 games at a time getting a good ranking.

I believe that the Elo system currently being proposed would actually limit growth of players who play too many games. If you play heaps of games then you will require a reasonable percentage of wins to get a decent rating. The higher your score, the better percentage of wins you need to maintain it. By spamming games it may keep you off the bottom, but certainly not near the top.

Maybe we need to do some simulations but I honestly believe that the issue you have would be non existent. The reason is this, if you get a 4 way draw against players that do not have as high a ranking as you, your score will go down! That's the beauty of this system. You can't coast along by playing heaps of games trying to get draws (well you could but you wouldn't get into the top 10). You need to get results better than what the Elo system predicts you should get otherwise you go down! Just like in chess if you draw against an opponent who does not have as high a rating as you, your score will go down.

E.g I'm ranked 2nd. (yes I'm so good I will be ranked 2nd *not*) I am 12 points behind the person who is ranked 1. Most likely I won't be able to play with people that have as high a ranking as me, meaning that even a two way draw could see me lose points. I need to concentrate and get a good win percentage, not play a million games and win only 2.

Your suggestion about the bonus pool is really cool. But it's not 0 sum. Maybe when we say 0 sum we are talking about different things. 0 Sum means to me that from every game the results added up will equal zero. For example if a solo is worth 7 then a loss is worth -1. If a two way draw is worth 10 then a loss is worth -4. If everyone starts on 1500 points when they join the league, then the total amount of points of the people in the league will always add up to 1500 multiplied by the number of people in the league.

QoH,

You're suggestions are great as they are easy to implement, and easy to understand. I get the feeling through that the majority of the classicists want a more complex system that caters for the strength of the opponent your playing. Or is this just a wrong assumption on my part?

I might add another column to the testing spreadsheet to see how fellowes proposal compares to the other two that are in there! I'll post again when its done.
InterMPC
 
Posts: 2225
Joined: 12 Jul 2010, 09:22
Class: Diplomat
All-game rating: (1000)
Timezone: GMT+10

Re: Classicists League Table

Postby InterMPC » 10 Nov 2010, 11:17

Here is the new link with the updated table including fellowes first proposal. This will allow you to have a play and see how his suggestion would effect results compared to the previous version.

https://spreadsheets.google.com/ccc?key ... y=CNy8hNwG

fellowes, i think I have replicated what you were explaining. Let me know if I got anything wrong.
InterMPC
 
Posts: 2225
Joined: 12 Jul 2010, 09:22
Class: Diplomat
All-game rating: (1000)
Timezone: GMT+10

Re: Classicists League Table

Postby raphtown » 10 Nov 2010, 12:03

InterMPC wrote:Here is the new link with the updated table including fellowes first proposal. This will allow you to have a play and see how his suggestion would effect results compared to the previous version.

https://spreadsheets.google.com/ccc?key ... y=CNy8hNwG

fellowes, i think I have replicated what you were explaining. Let me know if I got anything wrong.


Something is wrong... If you set it to a 2-way draw, with 1st winner at 500 and 2nd winner at 2000, all the losers at 1500, it gives NaN errors in the 'fellowes' system.
The Classicists are a group dedicated to reducing player NMRs.
User avatar
raphtown
 
Posts: 2257
Joined: 17 Apr 2009, 19:07
Class: Diplomat
All-game rating: (1000)
Timezone: GMT-8

Re: Classicists League Table

Postby InterMPC » 10 Nov 2010, 12:22

Sorry, I only made the table check for differences in ratings of 1000 or less. My bad. I'll fix.
InterMPC
 
Posts: 2225
Joined: 12 Jul 2010, 09:22
Class: Diplomat
All-game rating: (1000)
Timezone: GMT+10

Re: Classicists League Table

Postby InterMPC » 10 Nov 2010, 12:32

I've fixed the issue. You can now do up to a difference in ranking of 3000. Any more than that and there will be errors.

There are some other anomalies, but they seem to be unintended consequences of the method used. Unless I have misinterpreted something fellowes said.
InterMPC
 
Posts: 2225
Joined: 12 Jul 2010, 09:22
Class: Diplomat
All-game rating: (1000)
Timezone: GMT+10

Re: Classicists League Table

Postby fellowes » 10 Nov 2010, 16:50

InterMPC wrote:I've fixed the issue. You can now do up to a difference in ranking of 3000. Any more than that and there will be errors.

There are some other anomalies, but they seem to be unintended consequences of the method used. Unless I have misinterpreted something fellowes said.


Hmmm. After tinkering with it, it seems to be mostly working normally. There is still a massive issue with ELOS even 1000 apart not giving anyone any points, but this is probably not going to be a major concern in the short-run. Should be a matter of tweaking ELO tables/values/scaling/etc.

Also, there seems to be an issue when a high-ranked player (2500) scores a 2 way draw with low-ranked players (1500). He LOSES points. This should not happen in my option 1 (I think this is the one you said you were implementing). The winners are NOT credited fractions of wins against each other, just against the losers.

An explanation of what exactly is going on in rows 38:44 would help a bit (all columns), if you want my input fixing this issue.

EDIT:
I think I understand it a bit better. If 1 and 2 draw, 1 with 2500 and all others with 1500, we currently have
G38 = .5, G39 = 1, when both of these values should be 0 (this can be done with an "if" - since winners don't lose points to eachother, just set these values = 0 if both are winners, so E39 and F39 should be 0 too).
fellowes
Premium Member
 
Posts: 352
Joined: 23 Jan 2010, 04:43
Class: Ambassador
Standard rating: (1177)
All-game rating: (1185)
Timezone: GMT-8

Re: Classicists League Table

Postby fellowes » 10 Nov 2010, 18:39

Also, you haven't implemented the "bonus pool" functionality. Though that will make it harder to tinker with values to test that everything else is working right (each run will change bonus pool values), so let's not add that till the end.
fellowes
Premium Member
 
Posts: 352
Joined: 23 Jan 2010, 04:43
Class: Ambassador
Standard rating: (1177)
All-game rating: (1185)
Timezone: GMT-8

PreviousNext

Return to Miscellaneous

Who is online

Users browsing this forum: No registered users and 1 guest