Maximum K values

[Flaming_Spinach]

There has been no final decision on the K values. I would like to do another round of analysis now that we have the complete 2006 season as a data set.

I will say that there will be no crazy 100K events. Who does that?

Cool.

Can you at least confirm wether or not the attendance of a tournament will play a role, since that seems to be a rumor around here?

 

[Rainbowgym]

First of all, I am absolute not aware how the ELO system works, and not able to calculate or whatever.
In the POP documents it says K value for rating was 32 last season (that's what on the OP website).

Now something to shoot me for if you wish, or meaby it makes sense:

Would it be good to only let 4 CC, 4 Battle road, 1 state, 1 Regional and 1 National count for the rating?
So you can attend 10 CC's, but only your 4 best rating giving ones count?
Or does this give to many opportunities to people who can afford to travel, to end high?

I try to understand this Kvalue thing, but I do fail each time.

 

[Flaming_Spinach]

First of all, I am absolute not aware how the ELO system works, and not able to calculate or whatever.
In the POP documents it says K value for rating was 32 last season (that's what on the OP website).

Now something to shoot me for if you wish, or meaby it makes sense:

Would it be good to only let 4 CC, 4 Battle road, 1 state, 1 Regional and 1 National count for the rating?
So you can attend 10 CC's, but only your 4 best rating giving ones count?
Or does this give to many opportunities to people who can afford to travel, to end high?

I try to understand this Kvalue thing, but I do fail each time.

Everything you want to know about the ELO system can be found here: http://en.wikipedia.org/wiki/Elo_rating_system

Put simply, with 2 peoples rating, you can (in theory) caculate the % chance each person has to win the match. A difference of 200 points means the higher-ranked player has a 75% chance of winning the match.

The K-Value is the total number of points that the players ranks will change after the match is over. With a K=32, equally ranked players each have the chance to win or lose 16; if you have 2 players of different ranks, the higher one has the chance to win less or lose more than his opponent, the lower-ranked one has more to win and less to lose.

Simply, the higher the K, the more variable the scores are.



As for your second point, it has 2 problems:

1. As you state, people who go to more events will naturally have a larger pool to select their "good" results from.

2. ELO rankings are supposed to be a measure of your TRUE score, not just your score of all your good days.

 

[Rainbowgym]

However I don't understand all on that Wiki document (heck it's in English, which is not my language)
I do find it interesting that it seems K-value is/was something related to a players rating and not related to a tournament.

If I understand correct POP uses a system were in a tournament called X the K value= Y( so for each player the same)
While in that document it looks like each players K=Value varies no matter what kind of tournament, but depends on his personal rating.

 

[NoPoke]

The wiki article is very good. Here is THE MOST IMPORTANT line from the whole article.

A statistical system, not a reward system

The DCI uses 'ELO' as the basis for a reward system. And I'm expecting Pokemon to do much the same with a REWARD system.

This isn't wrong, there is nothing saying that you can't use the logistic equation as a basis for a reward system. Just be aware that many of the decisions on size of K value that are described in the wiki article then no longer apply. Most sports are based upon a reward system. Very few use a statistical system.

Most of SLOW DECK's concerns in his other thread arise from the fact that we will have a reward system and not a statistical system. I'm nowhere near good enough at stats to know if it is possible for a statistical basis to be found for pokemon.

 

[Rainbowgym]

LOL, I did not mention that part because of what it says and how it's used.

I do feel(without proper education) if we got a reward systeem were K-value is calculated upon player ranking it would be much different, but meaby more fair to find the best players of a region.

(again this is what I "feel" I am not able to calculate whatever, I am only using "intuition" and minimum knowledge, which helped me for whole my life to locate "problems" very acurate)

 

[NoPoke]

intuition and statistics dont mix. *grin*

The Statistical system 'ELO' is used in Chess, and Go. Both of which are pure skill. It is also used in Scrable and Tantrix, both of which have a very small amount of luck but not sufficient to undermine Arpad Elo's work.

Microsoft, who employ some very clever mathematicians, have extended the Elo system in their True Skill rating system that they use in their online gaming systems. But agian these games are essentially devoid of luck.

A statistical system to determine Pokemon player skill is a chalenge not least because matchups are so important. Even the best player alive wont win against a moderate player if the matchup is very bad.

 

[Flaming_Spinach]

While in that document it looks like each players K=Value varies no matter what kind of tournament, but depends on his personal rating.

There is a difference between Chess and Pokemon.

In chess, the K is determined by the players rating. The higher the rating, the smaller the K is. This is because chess does not want people ranked extremely highly to suddenly drop several ranks if they lose 3 or 4 games in a row.

Pokemon uses a fixed K dependant on the type of tournament. ALL cities have K=8, States have K=16, etc. Higher ranked players have just as much at stake as lower-ranked players.



Does that help?

 

[Rainbowgym]

I know Pokemon uses a fixed K value, last document I saw it was 32 for each tournament.

What is wrong with being high ranked and not drop as hard?
First you have to get that high rank by winning.
and with "personal" K value a lower placed player can go up much more by defeating a high ranked player.
Somehow that sounds more balanced to me, but I cannot explain why. (intuition again?)

There was also something about (don't know if this was in chess) your rating becomes lower if you are a certain period inactive. But than you will have to decide which period of inactivity will make you loose point. So that would be a bad option?


Does anybody think POP will introduce something like you can only play 1 Regionals and 2 CC's???
or something like 1 done and gone even if you don't get first place??

 

[dld4a]

Does anybody think POP will introduce something like you can only play 1 Regionals and 2 CC's???

I have been wondering kind of the same thing.

or something like 1 done and gone even if you don't get first place??

:nonono: Win 1 and done maybe, but even that's bad for CC's...

 

[Flaming_Spinach]

There was also something about (don't know if this was in chess) your rating becomes lower if you are a certain period inactive. But than you will have to decide which period of inactivity will make you loose point. So that would be a bad option?

I've heard about that, although I don't remember where. I think it is an excelent idea.

Does anybody think POP will introduce something like you can only play 1 Regionals and 2 CC's???

I hope not for Ciities and Battle Road. I think it would be terrible to have 5+ cities in your area, but you can only go to 2. All it would do for POP is convince people to not buy cards after they go to their first 2 events.

or something like 1 done and gone even if you don't get first place??

Same thing. Why would I want to play in a Cities if I knew simply by playing I would be excluded from all other Cities (which last about 3 months). Who would want to play this game if they could only play in 5 tournaments a year?


*****


@PUI (assuming you're listening) - If you are going to use a K-value that uses tournament attendance in its calculation, may I propose this one:

K = k(a)

Where:
K = K-value used in calculations
k = k-value based on tournament type
a = a-value dependant on attendance at the tournament in question

k (little k) is dependant on the type of tournament in question. Cities and Battle Road have a k=4 and States and up have k=8. This is due mostly to the number of events. Cities and Battle Road (apparently) will have much more events to go to, but States and higher will be much more limited, so these events have a NATURALLY higher k.

a is dependant on the attendance at the tournament in question. Log-base-2 makes the most sense since our tournaments already use it.

K is the product of k and a. It is the actual value used in calculations.


I think this makes sense because using this formula, the attendance of an event is taken into concideration, but it does not weigh to heavily on the calculations.

 

[toby]

There is a difference between Chess and Pokemon.

In chess, the K is determined by the players rating. The higher the rating, the smaller the K is. This is because chess does not want people ranked extremely highly to suddenly drop several ranks if they lose 3 or 4 games in a row.

The ELO system is a filter. It combines old information (old ELO score) with new information (current game results). The ONLY thing that should matter is the statistical confidence in old information versus new information. Chess does not only use ratings, but also uses number of games played. The only reason K should depend on a players rating is if there is a correlation between the rating and the confidence that the rating is accurate. Perhaps a correlation like that exists--I don't know. However, there is a correaltion between the number of games played and the statistical confidence in the rating.

What makes REAL sense? Well, and Alpha-Beta filter, where the Alpha and Beta are more easily recognized and controlable. Both Alpha and Beta would be straight forward and justifiably based on the Variances of 1) the old rating, and 2) the current game result. Actually, the whole thing could be made completely justifiable. Things like rating staleness could be easily taken into account, by calculating the Variance of the old rating by only using games played within the last x months, or something.

Needless to say, that isn't going to happen.

 

[SLOW DECK]

More I study K, and ELO system. K should be 48 to 64, and should be constant through out the year.

If K start's low, you might as well not play. K is the update for each match played. Small tournaments means Low number of matches, thus a small tournament will mean little chance of increasing your ELO because small number of matches.

K constant or slightly increaseing. Starting low won't make sense. If someone win's some local cities, and then get's there butt handed to them in the larger states and regionals, then the points system adjusts that.

Keeping the tournaments ELO rating limited to premier events, it removes the weekly tournament big shots.

 

[Petros]

???

PLZ!!!!!!

Can someone tell me where to find the meanings of the abbreviations you're all using? I am new here, so I don't know any of them. For example: What is POP, K, ELO and that kind of stuff? Where to find the meanings? Are there any topics for them (I can't find them), of do I have to make a new topic myself?

Thanks for the help anyway

 

[toby]

PLZ!!!!!!

Can someone tell me where to find the meanings of the abbreviations you're all using? I am new here, so I don't know any of them. For example: What is POP, K, ELO and that kind of stuff? Where to find the meanings? Are there any topics for them (I can't find them), of do I have to make a new topic myself?

Thanks for the help anyway

Allow me to do your leg work. Post #6 gives: http://en.wikipedia.org/wiki/Elo_rat...atical_details

A Google Search turns up: "POP stands for Pokemon Organized Play"

 

[dld4a]

PLZ!!!!!!

Can someone tell me where to find the meanings of the abbreviations you're all using? I am new here, so I don't know any of them. For example: What is POP, K, ELO and that kind of stuff? Where to find the meanings? Are there any topics for them (I can't find them), of do I have to make a new topic myself?

Thanks for the help anyway

This might help to.
Encyclopedia of Pokemon Terminology