Arimaa Forum - Print Page


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

        Arimaa >> General Discussion >> syed
        
(Message started by: mistre on Sep 7^th, 2007, 12:19pm)

Title: syed
Post by mistre on Sep 7^th, 2007, 12:19pm

It looks like we have a challenger for the number 1 player of Arimaa - syed! At a rating of 2430, he is only 10 away from passing fritzlein.

Let's look at how syed has gained such a lofty rating:

2834 games played - this is just absurd.
2585 wins - wow impressive.
91% win percentage - again impressive.

But there must be a catch! And there is...

Games vs. Humans - 0

To me this is just incredible. You would think at some point he would break down and play a human - if not to just to see how he stacks up.

Wouldn't it be great to see syed take on fritzlein or chessandgo?

If you look at the Bots syed plays against - nearly 92% of his games have been against a P1 bot or worse. His record vs Aamira2006p2 (a 1600 rated bot) is only 15-17 and against Bomb2005p2 his record is a mediocre 11-18. He has played only 5 fast games and 0 blitz games. On top of that, on the rare occasion that he plays a P2 bot - he makes the game unrated.

So a more realistic rating for syed would be in the 1800 to 1900 range.

Still, it is impressive that he has played Arimaa bots 2834 times and shows no sign of slowing down...

Title: Re: syed
Post by Fritzlein on Sep 7^th, 2007, 12:59pm

Not only is 2834 games played a record for a human, it is more games than all but four bots! I'll check my database some time, but I think the only bots with more games played than syed are

ShallowBlue
Arimaazilla
Bomb2005P1
Bomb2005Blitz

and syed is just about to pass Bomb2005Blitz. That's what I call fanaticism.

As for the ratings, well, we know the system is broken. I'm actually curious to see exactly how broken the system is. Syed already has briefly been the #1-rated player, so that milestone has been reached. But can he make it all the way to 2500? How about 2600? I'm on the edge of my seat...

The other day Omar mentioned to me a possible way to get "real" ratings parallel to the current ones, without having to exclude bot games. The basic idea would be that you can't choose your opponent. Let's call it a "title rating". If you wanted to get a title rating, you would sign up for a game in the gameroom, and you would be paired with someone else (possibly a bot) who was also trying to get a title rating. The pairings algorithm would try to create relatively even games, plus a variety of games for everyone. Eventually, if your title rating got too high, the only people you could get paired against would be other humans. This is nowhere near implementation, but worth thinking about as a possible project some day.

Title: Re: syed
Post by mistre on Sep 7^th, 2007, 7:17pm

How about a separate number assigned to everyone like a strength of schedule rating? Take the average rating of all the opponents you have played against. That would give a general idea of who is overrated or underrated.

Title: Re: syed
Post by IdahoEv on Sep 7^th, 2007, 10:08pm

on 09/07/07 at 12:59:28, Fritzlein wrote:

How about 2600? I'm on the edge of my seat...

It's too slow and methodical a process to generate much edge of the seat excitement, methinks.

Quote:

The basic idea... Let's call it a "title rating" ... you would sign up for a game in the gameroom, and you would be paired with someone else (possibly a bot) who was also trying to get a title rating.

That sounds like a lot of effort for the developer for something that very few people would use. People allow their games to be rated because it's the default. If they had to schedule special games just to have a true rating, we would have very few such rated games.

I would imagine it would be much easier to tweak the system to divert abuse than it would be to generate a new one like that.

For example, add a threshhold function to the points swapped. If the ratings aren't close enough that the higher-rated player would gain at least, say, three points from the win, then the higher-rated player cannot gain any points at all by winning.

Or, limit the number of times one can gain points from any particular opponent rated below yourself.

There are also, I think, some good long-term solutions to the general ratings inflation problem, as well, which is best thought of as an economics problem and I think is also part of what's going on here. (Because constant-performance bots are seeing increasing ratings with time). But I'll save those concepts for a separate post, because they're a bit off-topic here.

Title: Re: syed
Post by The_Jeh on Sep 8^th, 2007, 1:27am

Go to http://www.econ.iastate.edu/prosci/moore/iahs/ then click on the Comments and Descriptions link. This site rates high school football and basketball teams in Iowa, and it does a very good job.

Is it possible we could implement something similar? Then a player's rating would not change based off each new game, but would rather be the result of a cumulative analysis of all the match-ups that have occurred. The only problem is that in Arimaa, there is not a margin of victory. There is the game length, but a longer game length does not necessarily mean a more evenly matched game.

Or I've got another idea - not for ratings but for an independent ranking system. Just apply the Schulze voting method, treating the players like voting choices, and win percentage as the preference for one player over another.

Title: Re: syed
Post by The_Jeh on Sep 8^th, 2007, 2:20pm

Well, he's finally got an edge on you, Fritlzlein. I guess slow and steady wins the race. You'd better go beat a few bots, and quick.

However, you still dominate p8-wise.

Title: Re: syed
Post by Fritzlein on Sep 8^th, 2007, 3:19pm

on 09/08/07 at 14:20:09, The_Jeh wrote:

You'd better go beat a few bots, and quick.

Been there, done that. Once I bot-bashed until I was 200 points above the second-highest rated player. I don't mind if syed has his turn now. My only regret is that chessandgo may miss his chance to have the number one rating, but since he is simultaneously World Champion and Postal Champion, he doesn't really have anything to prove.

Title: Re: syed
Post by chessandgo on Sep 8^th, 2007, 5:31pm

on 09/08/07 at 15:19:49, Fritzlein wrote:

My only regret is that chessandgo may miss his chance to have the number one rating

No worries, Karl. I am still attracted just as much as before by the #1 rating (among everyone but syed, that is), and I've been trying my best for a long time already, but you're holding it tight. I'll go on doing what I can, but I'm pretty certain you're going to stay #1 for a long time ... enjoy ;)

Title: Re: syed
Post by aaaa on Sep 8^th, 2007, 7:25pm

The big question of course is whether Arifuddin is actually related to Omar, but Syed seems to be a pretty common surname.

I'm actually the anti-syed in that, having finished the ladder, I now shun rated games against bots to keep my rating pure, although I hope the low rating hasn't led to people avoiding challenging me for a game (it would probably be more due to the throwaway handle).

Title: Re: syed
Post by Fritzlein on Sep 8^th, 2007, 8:46pm

on 09/08/07 at 19:25:14, aaaa wrote:

The big question of course is whether Arifuddin is actually related to Omar

Yes, he is Omar's uncle.

Title: Re: syed
Post by The_Jeh on Sep 8^th, 2007, 11:22pm

on 09/08/07 at 20:46:18, Fritzlein wrote:

Yes, he is Omar's uncle.

What an incomparable expression of family devotion.

Title: Re: syed
Post by RonWeasley on Sep 9^th, 2007, 6:18pm

I'm not sure about this coincidence. Lockhart once introduced me to his "neice", but she didn't look related.

Title: Re: syed
Post by 722caasi on Sep 10^th, 2007, 9:58pm

I think there should be a separate page for top p8-wise rated players.

Title: Re: syed
Post by mistre on Sep 11^th, 2007, 8:47am

There is. Just go to the Top Rated Players page and click on the link for p8 Ratings.

Title: Re: syed
Post by 722caasi on Sep 11^th, 2007, 8:53am

What i mean is, it should be ranked by p8 score or difference, not by the old rankings.

Title: Re: syed
Post by camelback on Sep 15^th, 2007, 9:27pm

I just want to add my 2 cents on this topic.

Arimaa rating is frustrating and deceiving. When I joined arimaa, I used to download games of top players and I also downloaded games of syed. Now it seems meaningless. When you are new to the game, not only you have to know about the game but also know about the "rating" and find the real top players. In fact, there are not 1 but 2 players in the top 5 who don't deserve to be there. It's high time this needs to be changed.

One simple solution may be to have a cut off rating, say 1900 or 2000, beyond which you cannot get points by playing a bot. Or to seperate bot rating with human rating.

Title: Re: syed
Post by mdk on Sep 16^th, 2007, 1:11am

on 09/15/07 at 21:27:26, camelback wrote:

i think it all depends on what you determine makes a player deserve to be ranked in the top 5. yes 2 of those players gained their ranking from playing bots but one beat the toughest bot an impressive number of times while the other has played far more games than any other human. who is to say that such feats don't deserve a higher ranking?

regarding looking at the games of top rated players, if you are trying to learn how to beat a bot then there is nothing wrong with looking at the games of a bot basher. if you are looking for deeper strategy then you will go to the top human v human games, namely those containing fritz and chessandgo.

i admit that at times i do have a problem with the rating system but i really feel that there are better ways that time could be spent than trying to figure out how to fix the rating system. i agree that the ratings are often inaccurate if you are trying to predict the outcome of a game or determine who the best player against humans is. the only simple solution i can think of is to have one rating for games against bots and one for games against humans. overall i think this would encourage more human human games.

Title: Re: syed
Post by DorianGaray on Sep 16^th, 2007, 2:21am

on 09/16/07 at 01:11:54, mdk wrote:

If I may be authorized to give my opinion, since I have been implicated into this "indictment". I think that P8 ratings are not worth much more than the regular ones. If you look at Syed's P8 rating it is still INCREDIBLY high given that as someone pointed out he's never played against a human opponent and given his "style" he'd probably be beaten by a newbie with only two weeks game-awareness under his/her belt. And even in the domain of bot beating he's only played against bots above P1 a minuscule fraction of the time and not recently. So technically he gets his ratings from P1s. P1s are not worth anything, for all we know Syed could as well be a bot himself. In fact that could work. Imagine that someone who created a decent P2 bot decides to make it play continuously against P1s. Don't you think that that bot would get well above 2500 after a time, and thanks to the flawed alternative an above 2300 P8 rating as well? ::)

Something doesn’t smell right here. :-X

A good alternate rating would be one that would give Syed his due, that is not more than 1700, and that’s generous! As for me, well, I should be excluded from the list altogether since I haven’t played a rated game for 14 MONTHS. ;)

Title: Re: syed
Post by The_Jeh on Sep 17^th, 2007, 11:47am

I see that Syed has broken down (on Omar's request???) and consented to a postal game with arimaa_master. Will he be dominated, or is this a repeat of Boise State vs. Oklahoma?

Title: Re: syed
Post by arimaa_master on Sep 17^th, 2007, 1:57pm

on 09/17/07 at 11:47:02, The_Jeh wrote:

I see that Syed has broken down (on Omar's request???) and consented to a postal game with arimaa_master. Will he be dominated, or is this a repeat of Boise State vs. Oklahoma?

Yeah, I was totally shocked seeing syed to be my opponent - anyway I think that was just an accidental mouseslip or so and he will possible forfeit on time,(no offend intended), although I would be happy if he will play.

Title: Re: syed
Post by Fritzlein on Sep 17^th, 2007, 2:47pm

Recently, I have been a strong advocate of not caring about ratings. It certainly applies to me; the less I care about my rating the more I enjoy playing, and the more I care about my rating the more Arimaa is stressful. To some extent I still let ratings ruin my fun, and that is too bad.

That said, I find camelback's argument very persuasive. When I first learned Arimaa, I dutifully replayed and studied all eight of Omar's wins in the 2004 Challenge. That turned out to be a waste of time compared to replaying even one or two of Omar's games versus 99of9.

on 09/16/07 at 01:11:54, mdk wrote:

regarding looking at the games of top rated players, if you are trying to learn how to beat a bot then there is nothing wrong with looking at the games of a bot basher. if you are looking for deeper strategy then you will go to the top human v human games, namely those containing fritz and chessandgo.

The problem is that a newcomer has no way of making the distinction of whose rating is based on what. To anyone who has just joined, syed is the best player, because "top-rated player" and "best player" are essentially synonymous. Especially chess players, who are used to chess ratings being quite accurate, would have no reason to suppose that syed (rating 2495) would be an underdog to jdb (rating 2010).

There is nothing wrong with bot-bashing. Indeed, the way the Arimaa Challenge is promoted makes it seem like bot-bashing is the whole point. I learned a lot from playing bots, because I learned to avoid the type of mistakes they know how to punish. But I learned very little about Arimaa from replaying games between a bot and someone who knew how to beat it. The human (consciously or unconsciously) repeats weak moves that provoke weaker responses. Studying bot-bashing games won't even necessarily teach you how to bash a different bot, since they don't all share the same blind spots.

A newcomer will have to find out by trial and error that the games of our top-rated player are very similar to each other. I'll wager that syed's last 100 games against Bomb2005P1 show less variety of strategies (and even tactics!) than my last ten against PMertens.

My solution of shrugging about inaccurate ratings and trying to care less is also denying that ratings have legitimate uses. But don't ratings have legitimate uses? Apart from wanting to know who would be favored in any given match (which is of some value, albeit debatable) it would indeed be nice if newcomers had some guidance as to which games were worth replaying. For the latter purpose, a misleading rating may be worse than no rating at all, because if we had no ratings folks would judge by tournament wins instead, and would therefore study chessandgo's games!

Title: Re: syed
Post by IdahoEv on Sep 18^th, 2007, 12:27am

To my mind, ratings should ideally represent playing strength only: it should be that variable which best predicts (in a statistical sense) the outcomes of future contests based on the results of previous contests.

Playing strength of course can't be completely accurately represented in a single variable and there will always be variations in performance. But rating could of course be a much better predictor than it is.

Title: Re: syed
Post by Janzert on Sep 18^th, 2007, 1:22am

on 09/18/07 at 00:27:19, IdahoEv wrote:

And actually I'm fairly certain that the current system does a good job of achieving this, if you add one little restriction to that statement. ;) It would be; as long as the future opponents are the same (statistical) population as the previous opponents.

But of course what everyone really wants is for the rating system to generalize from the subpopulation of previous opponents to the entire arimaa playing population. At least of course that's what I basically want, so will assume it must be what everyone else desires as well. :P

Janzert

Title: Re: syed
Post by The_Jeh on Sep 18^th, 2007, 4:00pm

on 09/17/07 at 13:57:53, arimaa_master wrote:

Forget about it. It's showtime.

Title: Re: syed
Post by Fritzlein on Sep 18^th, 2007, 4:29pm

It's great that syed is taking on this challenge. Of course, it is unrated, so he's not putting his rating on the line, but at least he's willing to risk his reputation, and maybe learn something at the same time.

I checked my database, which has games through September 15, and the all-time leaders in games played are:

7361 bot_ShallowBlue
5408 bot_Arimaazilla
3192 bot_Bomb2005P1
2945 bot_Bomb2005Blitz
2908 syed

When we look at only humans the list is

2908 syed
1776 Arimanator
1440 naveed
1128 PMertens
1109 Fritzlein
971 blue22
938 Belbo
900 omar

When you throw in the other lives of Arimanator, he's around 2500 games played, but still behind syed. Finally, just because I want to be at the top of some other list now that syed tops the ratings, lets look at the number of rated games versus humans:

425 Fritzlein
389 PMertens
317 chessandgo
297 robinson
246 naveed
236 99of9
234 arimaa_master
221 seanick
212 jdb