PBEM tournament, looking for players!

[zsozso]

OK, to clear out things, let's take the 2 extreme systems:

1. Hierarchical fall-out (aka. pyramid):
Everybody plays only 1 game at each level, only the winner advance to next level. Once somebody lost a game, he/she is out.

2. Full round system:
Everybody plays everybody else, numbers of wins counted and some secondary etc. scoring to decide any tie.

Which one is better?
Number 1 requires far less number of games, but it only decides who is the best, no further ranking is fair. Simple example, if the second best player gets to play the best in the first level: the second is out immediately and has no way to prove how good he/she really is.

Another problem with #1 is, that if one game finished at a level, the winner has to wait for 2 other games to finish, before a new game can be started.

Number 2 is perfect in the sense, that the final ranking properly reflects the skills, but the main problem is that it takes a lot of games. With 25 players, that would mean everybody has to play at least 12 games (with 3-player a game). Playing 1-3 at a time and estimating a game to last ~200 turns with an optimistic 1 full turn per day it takes about 7 months for a game, for those who play 1 at a time, the tourney would take 7 years even if no more people sign up...

So, we would like to get a nice compromise, which does not have the unfairness and waiting mode of the direct fall-out system, but does not take forever like the full round system.

Both the ranking and level-based systems offer some sort of compromise, and as I pointed out in my previous post, they are almost equivalent in their practical effect.

The main problem with the level based system, that it slows down thing by repeating games on the next level among the same 3 players.

The example where player A defeats player B (who is very good and wins many games after) in the first round and does not get enough credit for it, this example was not against the level based system, but against the ranking based point system.

[This message has been edited by zsozso (edited February 03, 2000).]

[Genaciv]

triple post.. ^
[This message has been edited by Genaciv (edited February 03, 2000).]

[Genaciv]

double post ^
[This message has been edited by Genaciv (edited February 03, 2000).]

[Genaciv]

After thinking it over some more.. in NO CASE should third get -ANY- points.. this could be abused to add up bunches of points in level1..

- Genaciv

[Genaciv]

Lord Maxwell, when you're dealing with SMAC, it'll take forever using ANY scoring system.. my method is to sort out the experts from the beginners, and after you use the points gained from wins in the different categories, you can still rank the players using secondary values (ie. TIME it took to win the game). Just like zsozso intended..
As for zsoszo's remark about player A beating player B (who is REALLY, REALLY good) in the first round, and then not getting enough recognition.. well.. i don't understand what he means..
If player B was defeated in the first round, one of his other first round games has to get him to the next stage.. and then for player B to eventually become champion, he will most likely meet player A again.. in level 2, or 3.. and whomever eventually wins (lets not say that.. but rather, whomever is in the lead..) will have deserved their ranking.. period. If you would like to rephrase the arguments against this method, please do, because at this moment i don't see any other better alternatives, and feel that this would be a great method to go by.. it had no ending, so players can push further and further, even if the all-time champion retires from smac, his title can be taken from him..
SUGGESTED REVISION to my METHOD: the differences between the different levels can be a factor of 4 or 5.. if you like.. this way level 1 and 2 aren't THAT DRASTICALLY different, but the difference should still be large enough to separate players..

or.. we could do this:
Level 1: winner=4pts, second=2, third=1
Level 2: winner=8pts, second=4pts, third=2pts
Level 3: winner=16pts, second=8, third=4

I'm not sure what ya'll want, all i know is that whatever we choose with the points, they should DIFFER based on LEVELS.. that's it.. whatever ya'll think is the best to hand out for trying, getting second, or winning, is up to you to decide, i guess.. my only concern is that only winners can advance.. (you could further complicate it by saying someone who gets 2 second places can go up.. or 3 third places go up, or something).. BUT I WAS TRYING TO KEEP IT SIMPLE, AND UNDERSTANDABLE..

LEVELS should be incorporated, points based on them in some way, ...

- Genaciv

[zsozso]

Korn469,

If we award points according to the start-time as you suggest, then your system is practically equivalent to Genaciv's level based system. Just think about it: all games started so far and further ones started until the first completion award the same number of points: 20 to the winner, -5 to the second (-10 + 5) and -15 to the third.

Then all winners will want to play other winners which is the same as level 2 against level 2. Then those games award the same number of points, so after those the winners will want to play each other - equivalent to level 3 games, etc.

The only difference is, that playing a winner or a loser in the second round does not have so dramatic point difference. On the other hand the same problem occurs as I described above:
Assume player X will be the eventual champion. If player Y beats him in the first game, then he gets only 10 points for that masterful act, while player Z beating the champion in a later game may get a lot more for the same achievement.

A very simple equal scoring (each victory worth 1 point) would avoid that. The only problem with this, that it does not differentiate between beating weak and strong players. So we should either stcik to the simplest scoring or choose one that is sophisticated enough to avoid the timing problems.

[korn469]

zsozso,

the value of the win should be determined when the game starts not when a the games are over if a player starts five games at a time then the records at the time of the games's start should be the record that determines the final outcome

and we should weigh the victory

if player A comes in first
player B comes in second
and player C comes in third

player A should get points from both B and C and B should get points (but not as many) from player C

lets start everybody out at 100 points

if you win a game you get 10% of the other players 100 points if you come in second you get 5% of #3 points

and if you surrender to a player they only get 5% of of points and you automatically come in second but only get 3% of the third players points

so you will want to play somebody with alot of points...and all point values are set at the begining of the game not the end

so if player A has 100 points and player B has 90 and player C has 80 here's how it would work

A wins
B comes in second
C comes in third

A get 9 points from player B and 8 from player C

B loses 9 points to player A and gets 4 from player C

C loses 8 to A and 4 to B

so
player A gets +17
player B gets -5
player C gets -12

the players total point value are determined when the game starts not when it finishes

we round all fractions and so if you have 94 points the winner would only get 9 points and you'd only lose 9

what do you think of that?

korn469

[zsozso]

Good point Lord Maxwell!
We could have followed a very simple direct fall-out hierarchical system: each player plays a single game, winner advance only to second level games, and so on until a single players comes out on top of the pyramid. The reason I dropped this idea right away, because that would take forever. The level based scoring is basicaly a variation on that.

On the other hand, there is also a big problem with the opponent strentgh based scoring systems (both ideas from Korn469 and Pagan[CyC]):
Those systems work fine for sequential games, i.e. when a single player never plays games parallel. Let's take an example why it fails.
Assume, player X starts 5 games at the same time, one of them with player Y. Player Y beats player X, but X wins all other 4 games. Now, if you do the scoring with all this knowledge at hand, it is obvious, that win of Y is a very valuable one. On the other hand if Y is so good, that he beat X faster than X won his games, then Y will get a low score for his victory. So, Y may get 4 different score values depending how fast/slow he wins, and he gets the highest score for the slowest victory against the same player. This is exactly the opposite of the realistic value of the victory!

I do see the importance of ranking victories against better or worse players, but we need a system that works for parallel games. Maybe we need to re-adjust the scores of earlier games in light of later ones ?

Zsozso

[Lord Maxwell]

Genaciv: Your scoring system has one major drawback, it would take to long. We need to use a point based system for the simple reason that most games will take 6-9 months to even near completion. (By then even a fast game won't have reached 2300. So most wins will be by economy or war. Maybe peacekeepers on empath guild and early size 16 cities with pop-booming can squeeze in a diplomatic victory... Not in smac-x though...)
I don't think that there will ever be a third round in your sýstem. (Knowing the fickelness of attention online.

[Pagan[CyC]]

An iterative method of scoring that takes into account the skill level of opponents may be what you are looking for. I wrote a program to do this for one vs one Quake results a few years ago, a basic description is below. NOTE: Round in this sense is an iteration of the scoring calculations, not rounds of competition.

1st round is purely win/loss

2nd round modifies the value of each win or loss based on the 1st round score of each opponent.

3rd round modifies the value of each win or loss based on the 2nd round score of each opponent.

Nth round modifies the value of each win or loss base on the (N-1)th round score of each opponent.

Rounds continue until the greatest change in score for a player is below a certain threshold.

With as few players as we have it wouldn't be that hard to do by hand or with a spreadsheet, it would almost certainly be less work than modifyig the Quake program. I'd be happy to do it if you'd like to do the scoring this way.

Pagan[CyC] <-- looking for an easy way to come out 1st! ;-)

[korn469]

instead of genaciv's level ranking system i think that we should have a point ranking system that works like this (al la starcraft's system)

Battle.net maintains both a win/loss record and a rating for each player who plays in ladder games. Players gain rating points for each win, and lose rating points for each loss. Because ratings take into account not only the number of wins and losses but also the skill of each opponent, they are a useful measure of the true skill level of a player.

The rating system used by Battle.net is the same Elo rating system that used in competitive chess, golf, and tennis. The Elo rating is a scientific model which provides a meaningful way to compare the skill level of different players, based on the assumption that skill levels follow a normal bell curve

Battle.net computes the probability of each player winning a ladder game based on that player's rating and the ratings of each of the players he is playing against. This probability is used to determine how many rating points the player wins or loses for winning or losing the game. A player who wins against a much higher rated player will gain many points for the victory, while a player who wins against a much lower rated player will gain very few points.

For two player games, the following formula is used to compute the probability of winning:

Probability = 1 / (1 + 10^(-difference_in_ratings / 400))

When more than two players participate in a ladder game, probabilities are adjusted downward accordingly. The probability of a player winning the game is the same as his probability of winning a tournament against all of the players in the game. The following formula is used:

AB = probability of player A beating player B
AC = probability of player A beating player C
etc.
probability = 1/3 * AB * (CD*AC + DC*AD) +
1/3 * AC * (BD*AB + DB*AD) +
1/3 * AD * (BC*AB + CB*AC)

Each player who completes a ladder game wins or loses points according to the following formulas:

Winner's rating increases by K * (100% - probability_of_winning)
Loser's rating decreases by K * probability_of_winning

the website i found this at is here http://www.battle.net/ladder/ladderfaq.shtml

korn469

[korn469]

Zsozso,

i think that you should leave on co-op victory, and make it where there can only be one winner but a player could "surrender" by annocuing it here in the forums and therefore be able continue to play, the player they surrendered to would have to accept their surrender...however once you surrender you would be ineligable to win the game

by surrendering you'd get a draw instead of a loss though, if the player you surrendered to won, if not you'd get a loss...but make the scoring system where you get more points for a win where your opponents don't surrender

it is my opinion that having a consolation prize for the leading player to give a weaker player would speed up the game and would still allow for diplomatic interaction later in the game, but that's just my opinion

korn469

[zsozso]

Paul: When I wrote all victory conditions enabled, I meant the types of victory, i.e. conquest, economic, diplomatic, trascendence. Didn't even think about cooperative.

As MtG points out, it does not make sense to allow cooperative victory. Unfortunately, I did leave it on in the scenario files, so it is ON in theory...

Now I'd like to take this opportunity to ask all players to avoid cooperative victory, please! i.e. if two players pact to eliminate the third, they should cancel the pact before taking the last base.

I will turn off the cooperative victory flag for the games started from now on, but it is too late for the game ACT001-015 and AXT001-006.

As for the scoring system, I like the level based thing that was suggested by Decx first then Genaciv. But I am not sure about the 10 times scoring system. After all, you might be playing a very tough player in level 1, so the victory could be just as valuable as a level 2-3 or whatever. If the players are placed into separate 'leagues' according to levels, then it is better to say, that 3 level 1 victories simply do not raise the player to level 4, only to level 2. Then they are ranked in the level 2 table according to their victories from level 1, but they will only play each other in level 2 games.

So, all games started so far are considered level 1. Winner will be anounced at whatever level the game stops, i.e. no 3 players can make to that level to play another level game. E.g. if there are only 4 players in level N, then all 4 combinations are played between them, but only 2 players made victories, then those two are in level N+1 and the one with better rank is the Champion.

In fact we can talk about Champion starting from level 2, but the title may move o another player during the course of the tourney.

Thoughts, comments ?

[NeverNeverNavarro]

sorry, didn't mean to post that 4! times... I sounded really pissed off! I'm not btw!

Count me in if there are still spaces, etc.

If I had had more time to do it I would have!!



Navarro

[Genaciv]

Just a reminder that i'm interested in starting a second game as the University.. and put me on as able to do 4 or 5...

- Genaciv