[ List Latest Comments Only For Pages | Games | Rated Pages | Rated Games | Subjects of Discussion ]
Comments/Ratings for a Single Item
If people think they are going to be 'rated' on games that they have previously been told, don't count, there is going to be a chilling effect that discourages people from participating on the website here. For that reason, two separate rating systems should be used, if at all any ARE used: one would be for those adventurous souls willing to play games they've never even heard of (and are therefore attempting to learn), and another rating system for those souls that insist on playing just one particular game, something they actually know a thing or two about, and about which they have developed theories of play that are put into practice.
My program isn't ready yet, but I'll keep it in mind. I think there are a lot of programmers hanging out at this website (http://www.chessvariants.org) and many of them may have hacked together a program out of nothingness, using nothing but sweat and insight, and they should be encouraged to have their programs brought in as well. (Ditto goes for the people that are responsible for programming ZOG.) I think that a 'Game-Ply Rating' system would probably oscillate around a bit, with every re-calculation introducing a little bit of drag and a little bit of drift- considering how 0-ply systems would hover around 1000, 1-ply systems around 1100, 2-ply at 1200, and 3-ply at 1300, and so on. Using a 'Game-Ply Rating' system, to which computers could contribute benchmarks, would make the human performances more meaningful. And if a human's USCF or ELO chess rating were imported into the 'Game-Ply Rating' system, it would probably see a steep climb before stabilizing. For instance, if a handful of human beginners at 800 USCF started playing a few 0-ply computers at 1000 GPR, the human ratings would go up. I would oppose lowering a computer's GPR rating, however. If a computer has a GPR rating, it should only go down as a result of a loss to another computer. This is because humans are inherently smarter than computers. The computer GPR ratings ought to be independent benchmarks that only they themselves contribute to.
Matthew, that would be an interesting experiment. We could use Game Courier as the interface with the CPUs doing the thinking! Can you suggest a game? We'd have to have it available in 3 platforms, your program, GC, and ZOG.
Tony, to answer your question, I already have two ST computers, so I don't have to pony up for a PC computer, nor buy a Windows developer's license (which I understand is $1,500 on top of the purchase price of the hardware, and it costs even more to get a Mac and a Mac developer's kit, if that even exists, which I'm not sure about). And then there's the learning curve. I'd have to learn how to program a Mac or Windows, and that's one daunting task right there. Also, I prefer computers whose operating systems are in ROM, and therefore incorruptible. For the purpose of establishing benchmarks, maybe you and I could someday test out our programs against each other's? You could use a modern computer running at 2.5 gigaherz (or whatever), and I could use my little computer? I'm naturally referring to the game of Baroque - or one of its relatives (but no Rococo, please). And certainly not Chess, as there are enough Chess-playing computers already. Baroque is a more challenging game, and requires far more calculations than Chess does. (Markedly more, if we allowed either side to delay indefinitely the reversals of their rooks (causing one to become an Immobilizer), or the reversals of the King and Queen (Withdrawer)), which adds a whole extra element of long-range strategy to the game. When it comes to modern computers, there are zillions of programmers that are better than I am. I'm no virtuoso. I just put my nose to the grindstone, and keep toiling away at the darn thing until it works like it's supposed to. That means a near endless examination of the states that the 'programming engine' outputs, and you would not believe how poorly implemented the Atari support package is, you end up having to kludge out your own suite of programming tools, the kind that more or less work right most of the time. I've put in about 9 months or more on this thing so far. Sure is slow and tedious. As for computer contests, we /could/ use an ordinary telephone line with direct connections, no webmail involved. (Or we could just post the moves here or at some other mutually agreed-upon place.) The role of the user-attendant would be to type the moves in, as they come.
As the thread Derek points to indicates, the strength setting is a ply setting. Given that this is true, I hypothesized that at lower strength levels, ZOG would not take forever to make a move with the thinking time set to forever. So I lowered the strength, set thinking time to forever, and let ZOG play itself. It made moves very quickly. When I raised the strength, it made moves less quickly but still fairly rapidly compared to forever. So experiment corroborates the claim that the strength setting is a plies setting. So my claim that ZOG has no plies setting may be false. But it was certainly not a non sequitur.
In fact, the Zillions program actually DOES have a ply depth control. It is illusive though and has been mistakenly presumed by many (including myself) to not exist. Please read this thread of interest from the Zillions discussion board: http://zillionsofgames.com/discus/ You must navigate manually the rest of the way since deep-linking is not supported. Zillions of Games Discussion Forum: Desired Features for Zillions of Games: Time keeping
ZOG does not have ply-setting, however, you can set the amount of time that the program will evaluate the position. This is indirectly reflected in the number of plies that are processed, although it depends on the CPU, of course. (Not to stir the pot, but out of sincere and friendly curiosity, why use a 2.5 MB RAM CPU when computers and memory are relatively cheap now?)
Fergus, your reference to Zillions of Games being a constraining factor appears to be a 'non sequitur' and not a premise. Supposing the games on this website are adequately described, it follows that all of the games on this website /can/ be played; not just online, but face to face, should the opportunity ever arise. Computers could eventually be brought in to help people learn how to play these games; wasn't increased accessibility the main idea behind designing Zillions of Games? It's unfortunate that Zillions of Games doesn't have a ply-setting, as that would have been extremely convenient for entry into my proposed 'Game Ply Rating' system. Even five or six computers that limit themselves to 2 ply searches, are going to play wildly differently if they disagree on the values of their pieces, or employ even slightly different pseudo-random number generators. Even if they were all playing absolutely randomly (and were therefore all assigned initial ratings of 1000), there would eventually be a departure from that number, as their true colors started to show, and the game results began to produce a bell curve of sorts. The inherent differences in programming is what appears to be the real quandary behind using a 'ply' based rating system; each computer will naturally play a little bit differently. (What sounds good in theory may not work out in the long run.) As for computers that can play one or two of the games available at this website, I'm still working on my Baroque computer game (for an ST upgraded to 2.5 megs RAM, minimum). Boy, it sure takes a ton of work to do all of the graphics, but at least the calculation part is proceeding fairly well. My program just isn't marketable, or user-friendly, in the slightest. (And I'll probably end up giving it away for free, anyway.) Zillions of Games appears to be a remarkable product, even if it won't run on my hardware platform.
Yep, that's me. I'm also known as Matthew Monchalin. It's a very frequent misspelling of my name. But while we are on the subject, try not to pronounce my name as 'Moncha Lin' (as it has the accent on the penult, so pronounce it 'mont CHAY lin'). I think there may also be a Matthieu Monchalin over in France, but I've never met him.
Matthew, The point I was making about Zillions of Games was a premise in an argument whose conclusion is that, for most of the games played on Game Courier, the software simply does not exist to implement the 'Game Plies Rating' system you have suggested. This has nothing to do with willingness to upgrade to a version of ZOG with a plies setting. There is no such version for anyone to upgrade to. My argument can be broken down like this: 1. ZOG does not have a plies setting. 2. Most games on Game Courier cannot be played by any software but ZOG. Therefore, most games on Game Courier cannot be played by any software with a plies setting. Also, my question about your name referred to a book, David Pritchard's Encyclopedia of Chess Variants, not to your id on this website. In that book, David Pritchard describes a game called Renaissance (aka Baroque Renaissance Chess), an Ultima variant created by a Matthew Monchalin in 1975. The name is close to your own but spelled differently. Is that you?
David, you must be thinking of USCF ratings, NWCF ratings (if they still exist), or ELO ratings, and each of these provide some way of estimating probable future performance based on previously observed past performance. I need to read up more on the Glickman Chess Rating system to see how it differs from that of the Game Courier Rating system, seeing as how they seem to share the same acronym. I was suggesting, on the other hand, a way of measuring computer programs pitted against each other, and against humans that are allowed to compete with them. You've probably heard of http://www.pogo.com where you can play cardgames such as Hearts and Spades (but no Skat, the last time I looked). You even get to play with robot players if you want. There are lots of sites like that in Internet. It's my understanding that the Game Courier here at www.chessvariants.org could handle cardgames like that, though one might be a tad less graphical than another. As for ZOG being some kind of a reason to forbid the 'Game Plies Rating' system I suggested, that only applies to people unwilling to buy an upgraded ZOG with the feature I suggested. Since I don't have a copy of Zillions of Games (and I'm unusually reluctant to go out and buy something that I don't even have a hardware platform to run it on), it escapes me why, exactly, the program can't be upgraded to play out all the plies that it has been directed to search through, short of observing that the person who originally programmed it, must not have felt like designing that feature. It was probably a case of him shrugging, and saying, 'Why bother?' If someone gets around to upgrading ZOG so it *could* search through entire Plies' worth of information (with or without regard to time controls, or the peculiar predicaments inherent in data storage), I'm sure it could do the job just as well as the next one could.
It is time to put the Ratings page in a visible sector. Also the next Tournament Page.
Matthew, if you open any of the games being played here on Game Courier and click on the 'Ratings' box, you will see ratings: restricted to that game variant. Replace the name of the game variant with a * in 'Game Filter' and click the 'Submit' box, then you will see the overall ratings. Playing random moves should result in a rating around 400 points below the weakest player in the pool (depending on how ratings are computed). For example, a ten year old child with an 800 rating has already mastered the Scholar's Mate and will attack the Queen with Knight. A 1200 to 1600 player, with a little practice, should win every game against a 2-ply depth search, which is consistent with a machine rating in the 800 to 1200 range.
I guess automating the gaming process by bringing in computers to play humans (or other computers) might overwhelm the website, but while I am on the subject, there should be some practical way of evaluating how well computers play against each other, even if a 2-ply computer will almost always lose to a 2-ply human, if only because the 2-ply human has a way of recognizing patterns and trends, and learns how to take advantage of them. I suspect an ordinary human capable of 3 plies will often beat a computer capable of 6, if only because the human can assess positions more deeply in a general, 'off-the-cuff' sort of way than computers can. Well, I'm not too likely to buy a copy of Zillions of Games, so your argument against implementation of a 'ply-based' ratings system using computer players for standardization purposes sounds more like an 'a priori' argument against it than anything else. Computers should be encouraged to participate against humans. If two computers made absolutely random moves, the likelihood of winning or losing would ultimately depend on their implementations of their pseudo-random number generators; and some platforms do that sort of thing much better than others can. Even still, a computer that made totally random moves should be rated 1000. Programs that were 100% 'open source' could be entered into the system for benchmark purposes. Computers otherwise operating on the basis of secret terms, or on the basis of undisclosed source code would find themselves ranked against those that were, just like humans are. In closing, if a computer that played utterly randomly could be rated 1000, and a depth of 2-ply would make it play with the equivalent of a 1200 rating, then it follows that 10-ply would bring it up to 2000, and 20-ply would bring it up to 3000.
The ratings page appear to be accessible by going to /play/pbmlogs/ratings.php (I hope I spelled that right because I don't have any cut-and-paste-capabilities) but if you want to print out a copy on your printer, check to see if the very last few pages are sized properly because they appear to require landscape fonts as opposed to the first 10 pages, which are for portrait fonts.
The user identification subroutine truncates my name from 18 characters to 16 characters; 'Matthew Montchalin' shrinks down to matthew_montchal because of the way the website handles registrations.
The Encyclopedia of Chess Variants mentions a Matthew Monchalin. Would that be a typo of your name or someone else?
Where, exactly, is the ratings page? Are you suggesting the existence of a webpage that deals with this subject? How do I get there? I tend to have a lot of trouble (and that's an understatement) navigating around this website (http://www.chessvariants.org) and rely principally on the menu system in place. The time lag between clicks and webpages loading tends to frustrate my navigation also. Humans should not have to wait 3 or more minutes for any given webpage to load. If this were a direct-dial BBS (and not a website) with a real telephone number, it would load a whole lot faster.
There isn't a simple formula for calculating the ratings. They are calculated by a complex algorithm. That is described on the ratings page. There is no strict line between provisional ratings and regular ratings. The same algorithm is used to calculate all ratings. You can consult the number of games played by each player as a guide to the trustworthiness of the ratings. Ratings are meaningful only relative to each other. Your suggestions about using computer players to set standards for what the ratings mean could not be implemented without destroying what the ratings already mean. Besides that, it would be wholly impractical, given the large number of games available here, and given that Zillions of Games, the only software available for playing most of the games here, does not let users set ply levels.
Where can I find the formula for determining the players' ratings? Is there such a thing as a performance rating, and a provisional rating? All other things being equal, two computers that are both capable of 2 ply searches, and make their moves accordingly, with absolutely no regard to positional nuances, only material differences, ought to be rated 1200. If they are capable of 3 ply searches, then their ratings ought to be rated 1300. Similarly, a computer capable of a 4 ply search should be rated 1400. By allowing computers to play against live humans, an exponential standard of sorts could be established for measuring human excellence.
Jared, I saved more time by writing the first thing that came to mind. Christine, if you do a little research, you will find that I must be very busy with a cross-continental double life if Manabu Terao and I are the same person.
It's not evil twins, it's evil QUADRUPLETS!!! **Cackle!!**
25 comments displayed
Permalink to the exact comments currently displayed.