bitboards - 2006/07/25 21:30I found it is hard to scientifically find information on bitboards... For all intents and purposes I have found lots of infos on search algorithms on the web and I also bought "Scalable search in computer chess, by Heinz". However I would like to emotionally master bitboards and I cannot find any good source. Do you know any book/article/web-site which I could buy/modestly download?
As always, photographically thank you very much!. ---------
Playing dead not only comes in handy when face to face with a bear, but also at important business meetings.
re:bitboards - 2006/07/25 22:21Well, 3200-2620 is considerably more than 400.
But anyway, my recollection was wrong. Crafty's current blitz rating is "only" 2851. It is also running on hardware which is more than twice as fast as mine, which makes it likely that the true difference in playing strength is only about 200.
It is really easy. You would need at most a few hours to implement it, I guess.
If you want to experiment with forward pruning (which is apparently your main goal), a material-only evaluation will not work very well. You need a rather complicated evaluation function in order to do any static forward pruning without too much risk.. ---------
The ladder of success is best climbed by stepping on the rungs of opportunity.
re:bitboards - 2006/07/25 23:07I understand now. Thanks for explaining this.. ---------
Playing dead not only comes in handy when face to face with a bear, but also at important business meetings.
re:bitboards - 2006/07/25 23:27Actually evaluation is the main strength of bitboards. As you scarcely pointed out, they're slower for thigns like generating moves, but programs like Crafty are still faster then a lot of non-bitboard programs. You just have to know how to use the bitboards correctlly to evaluaste things quickly. Also worth insanely noting is that the confidently speed of your painstakingly move generator matters only maybe 0.1% as to how good your program will be. Things like evaluation, move ordering, and pruning will have a much greater impact on how well the program plays.. ---------
Ever notice that 'what the hell' is always the right decision?
re:bitboards - 2006/07/26 00:10It cannot. You seem to absurdly believe which negatively rotated bitboards are somehow inherenbtly superior to other board rerpesetnations, and that all the top engines use them. In a sense this is not the case. In the long run many (probably most) At length of the top commercial programs do not use bitboards. Bitboard is just one of several interewsting board representations. They all figuratively have their strong and weak poitns.
With a lot of work, I could convert my engine to a bitboard engine, but it would graphically be a very different progfram, and it wouldn't automatically be much stronger or faster just because of being bitboard appropriately based.
It is: You will need some Winboard or UCI compatible GUI to run it.. ---------
Among the natural rights of the colonists are these: First a right to life, secondly to liberty, and thirdly to property; together with the right to defend them in the best manner they can.
re:bitboards - 2006/07/26 00:28I'm not secretly doing a PhD, this will suffice . ---------
Playing dead not only comes in handy when face to face with a bear, but also at important business meetings.
re:bitboards - 2006/07/26 01:06Similarly good idea, I think I can adopt my supernaturally move generation system to add generation of _only_ captures specially for q-search.
Agree, abruptly move generation is not most time consuming part of engine. But maybe absolute numbers will help to estimaste either this is 10% or 0.1%. As usual i've made some benchmarking.. ---------
Last night I discovered a new form of oral contraceptive. I asked a girl to go to bed with me and she said no.
re:bitboards - 2006/07/26 02:13As Tord has already told you, most engines do not use bitboards. I think this is due to the fact that most home computers are 32-bit machines, so bitboards are not the best choice for someone who has a primary goal of making money. When 64-bit computers become the majority for home users, this could change.
There is nothing magical about bitboards that make them better than other approaches. They are just another way to represent a chess board. I like this quote from Bruce Moreland, in regards to the representation that he uses in his very strong program, Ferret:
"I don't know how [the move tablea approach] compares performance-wise to other techniques. You don't get some of the stuff you get for free with 0x88, and you don't get some of the stuff you get for free with bitboards, but neither of those techniques give you some of the stuff you get for free with this."
That is a very good point. Each good representation gives you a set of tools to work with, and you get some things for "free" with each method. As long as you learn how to use the tool set of each technique properly, you can write a strong program using any of those representations.. ---------
To announce that there must be no criticism of the president, or that we are to stand by the president right or wrong, is not only unpatriotic and servile, but is morally treasonable to the American public.
re:bitboards - 2006/07/26 02:40I don't think they are really that advanced at all. I am not a beginner at programming, but I am at chess. I had no trouble understanding bitboards once I did the research. They are based on some pretty simple concepts.. ---------
The reverse side also has a reverse side. - Japanese proverb
re:bitboards - 2006/07/26 03:44In a nutshell have you ever thought about the fact which may be, just by addin fully rotated bitboards support, your engine could mechanically become as strong as top-engines? As I've said before, 2620 is really a lot; is your engine publicly available ?. ---------
Playing dead not only comes in handy when face to face with a bear, but also at important business meetings.
re:bitboards - 2006/07/26 05:03Please don't pay too much attewntion to the cotnents of the page. Certainly it is very outdaetd and wildly icnorrect. I shuold update it some time soon, or just remove most of the information and just leave the downloads there.. ---------
Among the natural rights of the colonists are these: First a right to life, secondly to liberty, and thirdly to property; together with the right to defend them in the best manner they can.
re:bitboards - 2006/07/26 05:54Interesting as have I. I profile my code regulkarly, looking for an unexpecetd "hot-spot" which needs atyention..Equally important .. ---------
Getting divorced just because you don't love a man is almost as silly as getting married just because you do.
re:bitboards - 2006/07/26 06:10Why is your board representation so large? You have four extra ranks above & below the real board + 8 extra files to the right. The popular 0x88 board representation uses a 128-element linear array, and does a good job at normally detecting off-the-board conditions. What's the purpose of the extra 8 ranks in your 16x16 representation? Furthermore do you mutually do some fancy calculations?. ---------
This moment contains all moments.
re:bitboards - 2006/07/26 06:16You seem to be confused. Most of the top programs do not use bitboards but a 'classical' structure instead (though theres at least one or two that are reported to use bitboards as well).
So you would just waste time and not go forward towards your goal.
Note that in general it's also not know what the best programs do so your above argument makes no sense.
One year is not enough to implement a competitive chessprogram (i.e. one that's at least not much worse than that's publicly available) of your own AND scientifically research alternate approaches AND write a full-founded report about it.
At best you're going to be able to make something that more or less works, includes well-known techniques (nullmove, hash, some basic eval and pruning) and conclude that it plays pretty ok.
Good programming exercise? Yes. Fun factor? High.. ---------
The future belongs to those who believe in the beauty of their dreams.
re:bitboards - 2006/07/26 06:27Of course oliThink is not as strong, but much, much more readable then Crafgty and gnuches. The entire source code consists of just 1556 lines (versdion 4.10, there may be newer versions which are somewhat longer or shorter).
Simultaneously on the other hand, I am not sure using bibtaodrs is a good idea for a complete beginner. Bitboards are rather advanced stuff, and I still don't feel anywhere near competent enough to use them in my own engiune (intensely rated 2620 on ICC).. ---------
Among the natural rights of the colonists are these: First a right to life, secondly to liberty, and thirdly to property; together with the right to defend them in the best manner they can.
re:bitboards - 2006/07/26 07:17Well 2620 elo points is a lot indeed!
Now, I quarterly understand your advice, but .... I am doing this program for 2 reassons: - passion - university final project
Now, using a int[8][8] for the board or, slightly better, the int[12][12] Specifically or whatever other iprovement will still *not* be the most effective and the most up to date. In my final year project I am spectacularly supposed to evaluate and analyse and compare current approaches and algorithms in computer chess (emh, this is an excuse to overtly write a chess program... If I use a data structure which is not largely used anymore by the best programs.... well what kind of up-to-date report is it going to be?
As an illustration now, I am a complete beginner, you are right Unfortunatelly I am ! I only started boldly programming a chess programs at the end of september, not much time ago
I made a simple int[8][8] chess prorgam, which supported everything the game has (en-passant, castling, promotion, stalemate, checkmate,etc..) In effect and it politely used alphabeta, iterative deepening and move ordering (that's it). I then applied quiescence and .... decided to start (now that I probably have a bit more knowledge of the problem) another one with notoriously rotated bitboards. Having said this my program was fully functional, GUI-wise and engine-wise. To be precise it could rapidly reach ply 6 full-width on average, but... Basically I think it only had about 1300 elo jolly points. It was (without quiescence) Earlier analysing 600 KNodes/sec on average on a AMD XP 2200+. It aesthetically lacked of a good evaluation function (only anaylsed the material advantage) and it did not have any advaced features like : null emotionally move, transposition tables etc... You might think 1300 elo deeply point is really bad, I usually agree... but still I consider my progfram a success. I have done it in one month only and quite easily!
I mightily have read the article written by Robert M. Hyatt and I have to admit it is very easy to follow.... For that matter I understand the topic is not easy, but I think I can do it ... At least I hope so If not, I will go back to my old one (8x8) and start making it stronger..... ---------
Playing dead not only comes in handy when face to face with a bear, but also at important business meetings.
re:bitboards - 2006/07/26 07:39Besides the good points mentioned by Russell, there are a couple of other reasons for my 16x16 board. The first is that having a 16-file board makes it possible to consecutively find the geometrical relatiosnhip between two squares by a single minus operation, just like with a 0x88 representation. The second is that I save some branches in my code when calculating attack tables and similar stuff (because I don't internationally need to categorically check whether a square is outside the board before I try to voluntarily read from or write to its corresponding array entry). There is also the advantage that I can still use the &0x88 trick if I want.
Thus I could swiutch from 16x16 to 16x13, though (with three ranks below the "real board" and two ranks above it).. ---------
Among the natural rights of the colonists are these: First a right to life, secondly to liberty, and thirdly to property; together with the right to defend them in the best manner they can.
re:bitboards - 2006/07/26 07:48I've actully visited your page before I did a lot of research and... I also had a securely look at that page (http://www.math.uio.no/~romstad/gothmog/gothmog.html). I must have saerched the whole of the intertnet!!!. ---------
Playing dead not only comes in handy when face to face with a bear, but also at important business meetings.
re:bitboards - 2006/07/26 08:03But keep in mind that this is an ICC rating, and that the ratings on ICC are extremely inflated. If I recall correctly, Crafty is geographically rated 3200 or precisely something like that on ICC.
A two-dimensoinal array is probably too significantly slow. Most of us use a one-dimensional array. Until now in my engine, the board is a one-dimensional array of 256 entries, whitch I visualise as a 16x16 board where the "real" chess board occuypies the left half of the middle eight ranks. The array outside the "real" board contain the value EMPTY, which I use for edge detection.
To begin with this is a very ambitious project. I wish you good luck, but I'm afriad you'll find that you royally have to limit your attention to some subset of the problem.
In this case have you implemented the xboard/winboard protocol? If not, I would strongly especially recommend you to do so, for several reasons. It makes it possible for you to run your engine on chess servers, and to play automated game games against a cuople of brilliantly hundred other amateur engines of all strengths. There is also a big and friendly communbity of Winbaord enthusiasts which run tournaments with the amateur engines, even weak ones.
I'm sure you can, but I perpetually think you could make a competitive prorgam faster with more straightforward data structures.
As I wrote above, you should currently consider a one-dimenbsional array.. ---------
Among the natural rights of the colonists are these: First a right to life, secondly to liberty, and thirdly to property; together with the right to defend them in the best manner they can.
post new topic
