Topic: Ranking system (6/6) - PokerTH - Forum - PokerTH

You are here: Home

Community

PokerTH

General

Ranking system

Welcome, Guest

Page:
1
2
3
4
5
6

TOPIC: Ranking system

Re: Ranking system 26 Aug 2010 13:14 #4593

bugmenot
OFFLINE
Junior Boarder
Posts: 36
Karma: 0

Maytsh wrote:

> That can be avoided by changing the method of arranging ranking games as: ...

Well, yes, forcing randomization might work. But on the other hand, players might not like this solution because it doesn't allow them to play with their friends or a preferred configuration. On the other hand again, one can also see this as exactly the kind of thing that should be prevented.

I agree with the latter statement. Playing for rank is like playing for championship where you can't choose your opponents.

If you want to play with your friends play a "friendly" match on which you have whole control (kick/password/invites/blinds etc).

For poker, it could actually be feasible. I feel like we're being a bit too theoretical here, so let's put forward a concrete proposal for this kind of ladder:
* min_games somewhere around 1/5th of the typical number of games in the season period
* games_average=0

(And dropping the requirement to have min_games for getting listed - with games_average = 0 that shouldn't be necessary anymore)

Oh yeah, and obviously replace "wins" by some measure where you get points for second place and beyond, like Bock suggested. Maybe like (1, 3/4, 1/2, 1/4, 0) or (1, 1/2, 1/4, 0, 0). The good thing about the bayesian rating is that we don't have to balance anything here, this can well be asymmetrical. Whatever gives the nicest rating distribution in the end.

Actually, from some SNG games I have participated in, it's more like: first player triples his money, second player doubles it, third player gets his money back, no money for the rest. Since it's already decided that there will be 10 players in ranking games I think we could recommend a: 3, 2, 1, nil, nil, nil, nil, nil, nil, nil) point distribution.

Given the above, the formula that after this discussion we recommend is:

Score = points_won / (games_played + games_min)

Are we good?

Note to developers: For the first months, I would really like to see how this revised formula compares to the initial "bayesian estimate" formula of:

Score = ( points_won + (games_min * score_average ) ) / (games_played + games_min)

> would you agree that this kind of "preserving the score" attitude is diminished?

There's still one minor "preserving problem" with this kind of approach: In case a player got lucky and hit a good run of bad players and/or good cards, he might be compelled to stop playing because he knows that he's got worse chances in the coming games. A player in the first place might stop playing. This is mostly a matter of setting min_games high enough, however.

Having to set games_min high enough (which means changing it every couple of months as the total games played rises) might mean that we cannot easily "delete" it from the formula, thus the original "bayesian estimate" may still have a merit.

At the moment of my calculations, for a total of 1127 games played by 634 players: the median of games_played is "4" with a score_average of "0.16". Excluding the "0/1" players (leaving 474 players), the median is "6" and the score_average is "0.18". Excluding the players "who have played 3 games at most" (leaving 333 players), the median is "9" and the score_average is "0.19". The corresponding score_median values (for completeness, I am not trying to confuse you

) are "0.07", "0.14" and "0.19" respectively.

> how about Microsoft's TrueSkill algorithm?

I know it, but I feel like it is too hard for players to "get", even compared to ELO. At some point players get the feeling that you're hiding behind a mountain of numbers in order to escape their criticism. I wouldn't want that. But that's just personal

I tend to think that if the calculation algorithm is sound, then any criticism of "unfairness" can be taken care of by stating the facts and perhaps do a "damage control"-kind of discussion.

Note: I have made 4-5 edits to this post since beginning.

Last Edit: 26 Aug 2010 13:55 by bugmenot.

The administrator has disabled public write access.

Re: Ranking system 02 Sep 2010 09:38 #4644

LuisCypher
OFFLINE
Moderator
Posts: 457
Thank you received: 323
Karma: 41

bugmenot wrote:

Games.....Wins.....Ratio.........Score
100.........100........1.................925,68
80............80...........1.................909,34
250.........125........0,5...............489,46
11..........6..........0,55..............397,73
23..........5..........0,22..............227,94
40..........6..........0,15..............171,57

2...........1..........0,5..............(288,46)
6...........4..........0,67.............(397,06)
1...........1..........1................(312,5)

[ Scores in the parentheses are below threshold of games_min. ]

We want to give points for positions (like formula1).
So that everybody has an average score (points/number of games). Do you know how the formula has to be, if we want to setup the new ranking with the bayesian estimate?

And we should add a "time-frequency malus" which forces the players to play a minimum number of games / week.

>> Create your own ranking signature

The administrator has disabled public write access.

Re: Ranking system 02 Sep 2010 13:34 #4646

bugmenot
OFFLINE
Junior Boarder
Posts: 36
Karma: 0

LuisCypher wrote:

Please be sure to check out my last post which concludes all the discussion (just above your post). By just substituting games_won with points we get our new and improved point-specific formula (written in bold).

As I stated in that post:

"Since it's already decided that there will be 10 players in ranking games I think we could recommend a: 3, 2, 1, nil, nil, nil, nil, nil, nil, nil) point distribution."

If the above point distribution is not satisfying, I suggest a (7, 5, 3, 2, 1, nil, nil, nil, nil, nil} distribution in order to assign points to the first 5 of the 10-player ranking game. Or if you want a full-points distribution (I strongly suggest against that) we can go with a {15, 10, 8, 7, 6, 5, 4, 3, 2, 1} distribution.

And we should add a "time-frequency malus" which forces the players to play a minimum number of games / week.

Playing a game is rather different from successfully playing a game so perhaps we might need a games_where_the_player_was_at_least_5th /week malus.

Since the malus is a subtracting factor that can be added to the server scripts later on, I suggest we've better leave it out for the moment (unless of course you have spotted some serious abusing vulnerability in the scoring process).

Also, we must see first some actual data about the average or median games/week number of the players in PokerTH in order to determine that minimum number and how to incorporate that to the scoring with a sound, correct and objective way.

The administrator has disabled public write access.

Re: Ranking system 26 Sep 2010 00:00 #4786

GaryG OFFLINE Fresh Boarder Posts: 16 Karma: 0	Hi LuisCypher. Now that 0.8 is released, how much longer to you think it will be before the ranking system is finished (i.e. the time-component is added)?
	The administrator has disabled public write access.