Arimaa Forum - Print Page


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

        Arimaa >> General Discussion >> P8 ratings
        
(Message started by: mistre on Nov 15^th, 2007, 8:54am)

Title: P8 ratings
Post by mistre on Nov 15^th, 2007, 8:54am

For the most part, I think that the P8 ratings more closely match a player's actual skill level more than the their actual rating.

However, why does P8 drastically overrate players who have just started and have no losses or very few? Look at Mhi, Shrek, Lamcf_Lamcf and mnn for example.

Also, is there any thought as to more prominently featuring the P8 ratings throughout the site? Currently, they are tucked away only on the Top Rated and Established Player links.

Title: Re: P8 ratings
Post by Arimanator on Nov 15^th, 2007, 2:33pm

Speaking of which, I think it'd be useful to add the ordering of the players by their P8 ratings in the page, at least as an option.

Title: Re: P8 ratings
Post by mistre on Dec 9^th, 2007, 8:27pm

Another thing that would be neat is to see the P8 rating change after every game instead of only once per day. That way you can see how wins and losses vs. certain opponents affect the rating. Is it possible to lose a game and have your P8 rating increase?

Title: Re: P8 ratings
Post by Fritzlein on Dec 10^th, 2007, 8:07am

We talked about whether it would be possible for a loss to a high-rated player to increase your p8 rating, or a win against a low-rated player to decrease your p8 rating. I think both are impossible, but the mathematical proof eludes me, so I could be wrong.

It would be possible to update the p8 ratings after every game, although the calculation is rather more CPU intensive than the standard rating calculation. You would probably have to wait a few seconds to get the updated rating.

My current bias is that p8 ratings, although better than the current ratings, are far from a panacea. Rather than go through the hassle of making the p8 ratings the main server ratings, I think we would get more benefit from simply having a "vs. humans" rating and a "vs. bot ratings", i.e. the same formula as at present but with two categories of games. That would be less work for more gain.

Title: Re: P8 ratings
Post by mistre on Dec 12^th, 2007, 1:06pm

I am interested in knowing why you have a lack of confidence with the P8 ratings. For the most part they are very accurate IMO, except when a player has just started and has zero or only a few losses. Maybe a fix for this is possible. P9?

Look at Syed and ArifSyed's ratings, who we know is the same person. Syed's main rating is 2424 and Arifsyed's is 2042. However, his P8 ratings are identical - 1896!

Trying to inflate a P8 rating is very difficult and nearly impossible. Look at Ariminator who has got his main rating up to 2678. However, his P8 rating lags way behind at 2161.

Your suggestion of splitting the ratings into 2 (vs computer and vs human) is not without its own perils. The computer only rating would quickly become meaningless and would be even more of a target for bot bashing. The human only rating would be interesting, but would also be a target of abuse as it would be possible just to pick on new players to inflate it. Also, players may start avoiding certain matchups because they know they will lose and there is not the comfort of beating a few bots to gain back the lost points.

I think we can all agree that the current rating system is OK, but it can be worked over by bot bashing. Not that there is anything wrong with bot bashing - but to use it just to see a number go higher is really counterproductive to the whole point of ratings in the first place.

If we can all agree that the purpose of ratings is to be able to compare relative strengths of players for the purpose of
finding opponents and for seeding tournaments, then the current system doesn't work very well.

While the P8 system isn't perfect, I think it just needs a few small tweaks for it to be very good. It already is much better than the main rankings anyways. Since these ratings are already kept, would it really be more work to make this rating more prominent on the site?

While I don't propose getting rid of the current main ranking system, maybe the two should be flip-flopped. The P8 ratings could be displayed on all of the pages and the other rating could be relegated to the background.

Just my 2 cents.