even if it is about ability i got the impression that it isnt working well. And might bemore acurate if example the calculation was based on the last 20 games someone played.
the 1 minute game was more i think a reminder of how much frequency has an effect on the formulas used here in a verry strange way (if you play not that often, your rating is likely to jump with large numbers) (how often you win or loos and get 61 pnts?)
rating system.
Sort:
The problem its a bit biased because of frequencie of the games played. Besides the loosing streaks, holiday breaks etc, or the verry freuqent players all these make it i think pretty inacurate.
So i wonder since statistics are so widely used in science.
With a growing collective that plays a random number of games each day.
If it could statisticly create a likely current rank, not ability, or likelyhood of winning
So imagine there 3000 players here and you're going relativly up 4 positions and get ranked 2123 or so, from 2119. (its a bit complex since people could join at each moment).
What you're worried about isn't really a problem. Say you have a bad week for whatever reason. Your rating goes down. You get over whatever was causing your slump and win your next game. Your rating change is based only on the result and the player's ratings at that time. While there is information about your slump games contained in the rating, the formula doesn't go back to look at them again. That makes the current game much more important than the old games, and your rating will quickly adjust back to where it should be.
well ehm i think there it is.
The formula doesnt work like that.
It adjust itself constantly to provide probability of a rank, so the number of games have played (and how) have much impact those will always folow you future ranking. So if in your dataset of played games you put in a "bad" time, this will affect next games in that data set. (call it buterfly effect)
Therefore i wonder if a ranking system could be made from only the last 25 games of all people here, i wont say its easy, different math is required, but since there are computers this might be possible.
Jul 9, 2009
0
#6
after 25 games i'm quite sure that the impact of the previous games has been quite diminished...
Its a math thing.
imagine everyone had 25 games against random opponents.
(but mostly against different people)
what i wonder is ..
well its pure a math thing, if on that a rank system could be made, theoretical.
Maybe you've heard that within 7 friends of friends you reachout almost every person on earth. So 25 games might connect / relate us all well enough to provide a current rank.
hmm i dont think people understand the point, its maybe a wrong place to talk about it too. It should be dropped on a web blog of mathematicans i think. Maybe i will just do that out of curiousity. So i'd like to end this discussion
(it's ok to keep posting but i wont track it anymore).
I get confused about the rating system.
Now i've seen the formula used.
But still i think how much does it realy reflect people, and so if it is right.
I give an example i was rated around 1170 a few weeks ago, then i went on holiday, my chess mind had not been used for a few weeks. So i dropped to around 1000. Next the things i saw got me wondered..
Now if i win from 1300 player i got about 10 pnts, on the otherhand when people loose or win me from (990 or 1200) me they get plus or minus 20 pnts (that's a lot). I imagine for something like that i need to win from a 1500 player or so..(strange)..
Now lets asume its fair, its all based on statistics but also on the number of games. If you had a bad period been ill had an holliday or whatever its stuck in your stats forever, and its this which i wonder if its ok.
I think it might be better to have the statistics based on a selection of games.
Over a fixed period of time, or better over a fixed number of last games, so then the statistics would better aprouch your current level. Well thats what i believe. Because a statistical data set, with corupt data in it, affects a data set. A good statistics system would filter that and would estimate your current level, not the level based on you loosing streak a year ago.
Possibly beeing fixed over a number of games would be the best.
A known downside of the current math is the frequency of games played.
(is this why vacation had so much rating effects?? (long time no play))
I have learned some math about statistics, altough i recognize this as a problem i wonder if a better formula could be written which could deal this problem a rating system based over the n(th) last number of games, between a group of people.
This is just a theoretical question could this be solved in math ?
----
Another strange example today i played my 18th and 19th, 1 minute blitz game, my rating is low in that ~900 and ofcourse i loose these games against a 1300 player as i rarely play these games. i lost 2 games, and it costed me about 61 pnts.. (????) a lower ranking loosing from a high ranked person got realy punished here (usualy its not that way). While this oponent won only 3 pnts together.
i think although its against the rules, if i would start another account, (with not my past statistical data), so to start fresh that acount would for blitz games be soon near 1300 range. As if i currently win from 1200+ players i dont get much point for it. however when you start one doenst get punished that much.