And, At any rate, how do you "pre.rank" teams?? based on what?
VOTE CHESS RATING INSTEAD OF FIXED SCORING SYSTEM
Sort:
Good points, kokino! I have to agree with your impeccable logic.
BTW, when we give assumed ELO to historic players, it is not using the Glicko system - it is using the conventional system. I am also OK with that - we need not go for Glicko per se.
kokino wrote:
And, At any rate, how do you "pre.rank" teams?? based on what?
Dunno, I guess what I have in mind is some sort of ladder, how does a ladder start, everybody equal?
dsarkar wrote:
Good points, kokino! I have to agree with your impeccable logic.
BTW, when we give assumed ELO to historic players, it is not using the Glicko system - it is using the conventional system. I am also OK with that - we need not go for Glicko per se.
Completely agree, it is useless (and completely inaccurate) to calculate the Glicko RD (Rating Deviation) when the games Databases are incomplete... and this is usual in such historic players. So, better to use simply the Elo Rating.
However, here you can access all the info needed. :)
rooperi wrote:
kokino wrote:
And, At any rate, how do you "pre.rank" teams?? based on what?
Dunno, I guess what I have in mind is some sort of ladder, how does a ladder start, everybody equal?
If we start equal... is the same as start all the teams with 1200 rate... don't you think?
I agree that maybe it might take long to get a consolidated and accurated ranking, but once there... it will be probably the best we could get.
Besides, in a ladder system, (I am not sure of this) there should be some kind of restrictions when challenging teams. I mean, it would be possible for a Team, let's say, ranked 200th, to challenge the 1st? In that case, what would happen if the Top Team loses the game?
It is just a question, I don't know...
kokino wrote:
However, here you can access all the info needed. :)
I doubt, even if we have all the relevant information, that Glicko RD for all teams can be calculated simply based on all the data in the database (e.g.: we find the start dates, but not the end dates of all games - and I dont think there is any field mentioning total number of days taken for a game to end - it has to be manually found out from the data). It will take a modern-day Sherlock Holmes to do that. Can you give an algoritm to do it based on all the info present in the database (I dont even know what are the fields there in the database)?
Good idea, though...
linksspringer wrote:
Interesting discussion. I've already been playing around with rating by downloading all vote chess results (luckily we are not talking about many thousands of games here!) and dividing between fast and slower games, disregarding quick timeouts. Calculating rating by "playing back" all games one by one is doable. One problem though is that some teams have changed their name.
Grand idea! But I think chess.com will need lots of unbiased, reliable, honest volunteers to do that - current staffs will be bursting at the seams to do that manually on all games, or it will take a very long time...
linksspringer wrote:
dsarkar wrote:
linksspringer wrote:
Interesting discussion. I've already been playing around with rating by downloading all vote chess results (luckily we are not talking about many thousands of games here!) and dividing between fast and slower games, disregarding quick timeouts. Calculating rating by "playing back" all games one by one is doable. One problem though is that some teams have changed their name.
Grand idea! But I think chess.com will need lots of unbiased, reliable, honest volunteers to do that - current staffs will be bursting at the seams to do that manually on all games, or it will take a very long time...
No, I already automated that ;-)
How??
Well, all I can say is that our discussion is proceeding in the right direction!
BTW, name change IS a problem - I know of more than one team which has changed name - more than once...
Looking for ideas about the group changing names constrain.... any ideas from other members?? :)
One think is clear, the names showed on each game are the ones when the game starts, not when the game finishes... so if a group changes the name while playing the name will stay...
linksspringer wrote:
dsarkar, I wrote a little Java program that requests the HTML pages and strips out the relevant data. (And I took care to take time between requests so as to not overload the server!) Someone already suggested to me to put the data in a publicly available Google spreadsheet, although I suppose I would need erik's permission first.
A spreadsheet would be great, maybe if a lot of people play around with it somebody can come up with a clever idea.
I hereby request Erik and the rest of chess.com staffs to reveal the numeric, date and name fields associated with the vote chess database(s), and then invite our brilliant programmers like linksspringer to write a program to calculate the Glicko ratings from it.
Jul 30, 2009
0
#35
kokino wrote:
Looking for ideas about the group changing names constrain.... any ideas from other members?? :)
I'm sure there is a unique team ID in the chess.com servers
Yes, I have been looking for info on the HTML and games id, but haven't found any way to independently of group name changes to keep refering to a team... I'll keep looking...
LS, have you checked on the HTML source there the team id? it doesn't change: a group should have it always, independently they change the name... otherwise, info would be lost if someone changes the name, (wouldn't be possible to see games archive on their group page, for instance)
(BTW, it is true, ☺☺BLACK оя WHITE☻☻ have changed the name to ☺BLACK оя WHITE☻ (we have lost 2 faces...lol), but those special characters are not considered in most of sorting processes... (in fact is black--white) so unless you sort looking for exact match, that shouldn't affect to most DB nor spreadsheets... :))
Oppps, forget my last post... :)
66_Mustang wrote:
Besides, many vote games within one team will be made up of different players, so the team rating would not be accurate of playing strength from game to game. One team vote game might be mostly made up of 1500 to 1600 rated players and another within the same group may be made up of 1800 to 2500 rated players.
This is very true - nobody can deny this! We are sometimes surprised at the fluctucation of playing strengths of some groups - in one game they are playing decent, in another game they are blundering like anything.
Also, sometimes a new strong player joins a group, and the strength improves... sometimes key players leave groups for some reasons - in such cases the groups' rating skydives (I have seen a few cases)!
Someone criticized me earlier that attempt at cutting short the time of a move by an admin "is like admins playing each other" - so what will be the significance of the "rating" of a group which changes its name, its players - is it an asset of the admin, or property of the creator superadmin (even admins can give up adminship)?
I still think a simplified scoring system which more accurately depicts the strength rather than the number of games played - is more suitable - a sophisticated "rating" system is overkill...
Look, I'm no statistician, and I don't know how to program. But I've been playing around in a spreadsheet and came up with a few ideas to run a ladder or ranking system:
1) New teams enter at the bottom of the ladder.
2) The naximum movement for one game should be the square root of the total number of teams. This changes as new teams get added to the ladder.
3) To estimate or predict the result of a game, simply use the current rankings of the teams, with a formula something like A/(A+B) where A is the lower ranking, and B is the Higher Ranking. The inverse wiil give the expected score for the other teams.
4) So, say there are 400 teams, If team 1(best) lose to team 400 (newest) team 400 will climb 20 places, and Team 1 will drop 20. If consecutive teams (eg 174 and 175) play, Winner should climb approx 10, and loser fall approx 10.
5) Obviously, no team should be able to go higher than 1, or fall below 400, in this example.
6) Either keep the rankings of teams as they are now, and let it sort itself out over time, or go back in history and enter results one game at a time, but this ssounds like a lot of work to me.
Dunno if this is of any value, I'm strictly an amateur :)
I think I've said this before, somewhere...
What is needed is a RANKING system instead of a RATING system. Winning moves your position up, losing moves it down. The size of the movement depends on the start rankings of the two teams.
That's exactly what Glicko system does....!
(but of course, the much more games you play, the most accurated your position in the ranking will be)