Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> General Discussion >> In tournament WHR Calculator
        
(Message started by: woh on Jan 26th, 2010, 4:11pm)
Title: In tournament WHR Calculator
Post by woh on Jan 26th, 2010, 4:11pm
A tool for calculating in tournament WHR is available.
The tool runs under Windows and IE is required.
The zip contains a readme file with some notes how to use it.
A file with the players and results of the WC 2010 is also included.
Download the tool here (http://home.scarlet.be/~woh/whr/CalcWhr.htm).
All comments are welcome and appreciated.
Title: Re: In tournament WHR Calculator
Post by Adanac on Jan 26th, 2010, 9:36pm

on 01/26/10 at 16:11:19, woh wrote:
A tool for calculating in tournament WHR is available.
The tool runs under Windows and IE is required.
The zip contains a readme file with some notes how to use it.
A file with the players and results of the WC 2010 is also included.
Download the tool here (http://home.scarlet.be/~woh/whr/CalcWhr.htm).
All comments are welcome and appreciated.


Thanks for the new tool.  The program seems to always cluster the WHR around 1500 though, regardless of what the initial ratings of the players are.  I've tried very high and very low ratings and that was the case both times.
Title: Re: In tournament WHR Calculator
Post by Janzert on Jan 26th, 2010, 9:55pm
The default is to use an initial prior of 1 win and 1 loss against a 1500 player. So the initial rating isn't used at all. You can change it to use a prior of one win and one loss against a player with a rating equal to the players initial rating but clicking 'sel' at the top of the WHR column and selecting WHRi. You also need to hit F9 or the Calculate menu item after changing this to recalculate the values.

In case you didn't see it the 'readme.txt' file has some information in it on using the program.

Janzert
Title: Re: In tournament WHR Calculator
Post by woh on Jan 26th, 2010, 10:19pm
Janzert is correct. The standard WHR doesn't use the initial rating. But you can select to use the initial rating in the prior.

I choose not to recalculate the ratings automatically whenever something has changed. Calculating the new ratings also sorts the player accordingly. I think it would be annoying entering multiple result when players are jumping around whenever one is entered.
Title: Re: In tournament WHR Calculator
Post by Fritzlein on Jan 27th, 2010, 6:15am

on 01/26/10 at 21:36:57, Adanac wrote:
Thanks for the new tool.  The program seems to always cluster the WHR around 1500 though, regardless of what the initial ratings of the players are.  I've tried very high and very low ratings and that was the case both times.

If the idea is rank players within a tournament, then pre-tournament ratings should not be used in the prior distribution, right?  Otherwise players do not control their own destiny, e.g. Hippo could go 5-0 and finish with a lower rating than I have after going 0-5.
Title: Re: In tournament WHR Calculator
Post by ChrisB on Jan 28th, 2010, 7:22am
Very nice program, woh!  Thanks for making it available for us to download.

I find the program easy to use and like the flexibility it has to try different priors and "what if" scenarios.  

As far as I can tell, it doesn't have an upper limit for the number of players, but it does get more sluggish as the number of players increases.  At 60 players, it takes about 10 seconds to make a change on my computer.  I still find that workable (5 minutes of total wait to enter a round of 30 games), but I wouldn't want to go much above 60.


on 01/26/10 at 22:19:37, woh wrote:
I choose not to recalculate the ratings automatically whenever something has changed. Calculating the new ratings also sorts the player accordingly. I think it would be annoying entering multiple result when players are jumping around whenever one is entered.

I agree!  :)  I would take me much longer to enter multiple results otherwise and I would likely make tons of mistakes.
Title: Re: In tournament WHR Calculator
Post by woh on Jan 29th, 2010, 7:52am

on 01/28/10 at 07:22:34, ChrisB wrote:
As far as I can tell, it doesn't have an upper limit for the number of players, but it does get more sluggish as the number of players increases.  


There is indeed no upper limit.
I didn't experiment with that many players.
I have put a new version (http://home.scarlet.be/~woh/whr/CalcWhr.htm) on line. It will be as slow as the previous version but I have added an option to use different colors in the result grid. This should make it easier to find to correct spot to add a result. Just select Tools\Colors if you want to use it.
Title: Re: In tournament WHR Calculator
Post by woh on Jan 29th, 2010, 8:09am

on 01/27/10 at 06:15:46, Fritzlein wrote:
If the idea is rank players within a tournament, then pre-tournament ratings should not be used in the prior distribution, right?  Otherwise players do not control their own destiny, e.g. Hippo could go 5-0 and finish with a lower rating than I have after going 0-5.


I think it is worth while to consider the WHR-i as a possible alternative to rank players in a tournament. A player with 5-0 will never be ranked behind a player with 0-5. But yes there is a bigger chance that not all players will be ranked according their number of wins. The more rounds played the less influence the initial rating will have on the ranking and there will be no ties from the first round on.
Title: Re: In tournament WHR Calculator
Post by Tuks on Jan 29th, 2010, 10:52am
i think that its a very accurate way of distinguishing player skills during the tournament and should be given serious thought too, i understand the idea that 3-2 wins qualify over 2-3 wins but im not sure it is really unfair if the 2-3 is actually proven as the better player
Title: Re: In tournament WHR Calculator
Post by Fritzlein on Jan 29th, 2010, 12:07pm

on 01/29/10 at 08:09:02, woh wrote:
A player with 5-0 will never be ranked behind a player with 0-5. But yes there is a bigger chance that not all players will be ranked according their number of wins.

If that is the only obstacle to adopting WHR-i, it can be easily remedied with a weightier prior distribution, i.e. more than one hypothetical win/loss against an average player.  Indeed, I suspect that as the weight of the prior tends to infinity, higher-order effects are damped out and the WHR ranking approaches the old ranking of wins first with sum of opponents' wins as tiebreak.  Does anyone care to validate/disprove this hypothesis on the current tournament standings?

Unfortunately, once one opens the issue of how heavy a prior should be used, everyone will realize that different-strength priors might produce different ranking orders, which gives us something additional to argue about.  Why 0.761034 hypothetical wins/losses instead of 15.2228?  I'm trying (and failing) to think of a good guiding principle that we could all agree on.
Title: Re: In tournament WHR Calculator
Post by aaaa on Jan 29th, 2010, 12:23pm
Minimizing the expected discrepancy of the final standings of players with respect to their "real" strengths shouldn't be the be-all and end-all of a tournament. Foremost, the outcome should normally respect the actual performances displayed in the tournament itself. In light of that, I feel that the number of wins should remain paramount in a Swiss tournament.


Arimaa Forum » Powered by YaBB 1 Gold - SP 1.3.1!
YaBB © 2000-2003. All Rights Reserved.