denbeste@bbn.COM (Steven Den Beste) (01/24/88)
There does not exist a good computer GO player that I have heard of, on any machine up to and including a Cray. The reason is that all the approaches which have been developed over the years for Chess (mostly at CDC to try to win the Levy bet - which they failed) don't work well for GO. Briefly, the algorithm consists of two parts: A section which determines all possible moves for a player in a given board position and a section which gives a given board position a metric of desirability. With these two, you trace a tree of the N best moves for me, followed by the M best moves for my opponent at each of the results of those N moves, followed by... These are known as "plies" I think, and once the approach was worked out, making the computers better consisted of two things: Making the evaluation code better, and making the computer evaluate more positions per move. This won't work for GO for two reasons: instead of perhaps 40 possible moves in an average board position, in GO it is often more than 250 - and any of them might be the one you want to make. Worse is this: There is no easy metric on board positions for desirability. In chess you use things like "-3 for ever major piece in jeopardy of being taken without an equal loss by the other side", "+5 if the opposing king is in check", "-5 if my king is in check" etc. (examples pulled from my ear - don't quote me.) It ain't that easy with GO (as any experienced player can tell you). Someone here asked about a GO program - I know nothing about specifics, but I'd be willing to bet much bucks that one of three things are true about it: 1. It plays TERRIBLY. 2. It doesn't play at all but allows two humans to play on the Amiga. (useless - a board is better) 3. It plays GO-MOKU (also known as "five-in-a-row" a game as different from GO as checkers is from chess) -- Steven C. Den Beste, Bolt Beranek & Newman, Cambridge MA denbeste@bbn.com(ARPA/CSNET/UUCP) harvard!bbn.com!denbeste(UUCP)
spencer@eris (Randal m. Spencer [RmS]) (01/24/88)
Recently on *comp.sys.amiga* denbeste@bbn.COM (Steven Den Beste) wrote: ...There does not exist a good computer GO player that I have heard of. Well, I suppose that means I am supposed to comment on this... The program is called "GO for the MAC" and soon "GO for the Amiga", and yes, programming GO is not easy. The point is to learn GO from the computer and from replays of previous matches. Also good when there is no human (or at least on interested human) around to play. People are still better machines. ... ...Someone here asked about a GO program - I know nothing about specifics, but I'd ...be willing to bet much bucks that one of three things are true about it: ... ...1. It plays TERRIBLY. Plays lots better than me! ... ...2. It doesn't play at all but allows two humans to play on the Amiga. (useless ...- a board is better) It plays you, it just plays as an 18 kyu (spelling?) level player. ... ...3. It plays GO-MOKU (also known as "five-in-a-row" a game as different from GO ...as checkers is from chess) It plays WHAT? ... ...Steven C. Den Beste, Bolt Beranek & Newman, Cambridge MA Yes, this is a port of the Logan code, quite a nice little package, take a look! Now any more of this conversation can be taken to BIX where I will dane to talk commercial (since I am paying to be there, here I come for fun). -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Randy Spencer 3461 N. Edison St. Arlington VA 22207 (703)241-2140 spencer@mica.berkeley.edu I N F I N I T Y BBS: (415)222-9416 ..ucbvax!mica!spencer s o f t w a r e AAA-WH1M -=-=-=-=-=-=-=-=-=-=-=-=-=See-=-=-you=-=-=-=in-=-=-=-D.C.-=-=-=-=-=-=-=-=-=
john13@garfield.UUCP (John Russell) (01/25/88)
In article <6250@cc5.bbn.COM> denbeste@bbn.COM (Steven Den Beste) writes: >Someone here asked about a GO program If there is such a beast in the public domain I'd be very interested in getting a copy via e-mail. GO, not Go-moku that is. John (john13@garfield.uucp , ...utai!garfield!john13, or john13@garfield.mun.cdn) -- "Am I dreaming, or was there a show on this weekend called 'Jimmy the Greek: Live at the Apollo'?" -- David Letterman
wolf@ssyx.ucsc.edu (Mike Wolf,4264777) (01/25/88)
In article <6250@cc5.bbn.COM> denbeste@bbn.COM (Steven Den Beste) writes: >There does not exist a good computer GO player that I have heard of, on any >machine up to and including a Cray. The reason is that all the approaches which >have been developed over the years for Chess (mostly at CDC to try to win the >Levy bet - which they failed) don't work well for GO. > [text deleted] > >Steven C. Den Beste, Bolt Beranek & Newman, Cambridge MA >denbeste@bbn.com(ARPA/CSNET/UUCP) harvard!bbn.com!denbeste(UUCP) Actually, there are quite a few GO programs which play competitivly at the amature level. I don't remember the terminology very well, but I seem to recall that the recent winner of the world computer go tourny was at the 18 kyu level. I think a high number is better, but I could be wrong. +------------------------------------+---------------------------------------+ | Michael Wolf | An old Scandinavian quote: | | BITNET: wolf@ucscj.BITNET | "You can lead a herring to water, | | ARPA: wolf@ssyx.ucsc.edu | but you have to walk real fast, | | UUCP: ...ucbvax!ucscc!ssyx!wolf | or else he'll die." | +------------------------------------+---------------------------------------+
dvl@hpisoa1.HP.COM (Doug Larson) (01/29/88)
> Actually, there are quite a few GO programs which play competitivly > at the amature level. I don't remember the terminology very well, > but I seem to recall that the recent winner of the world computer > go tourny was at the 18 kyu level. I think a high number is better, > but I could be wrong. > | Michael Wolf | An old Scandinavian quote: | Actually, there aren't any GO programs that are any better than 15 kyu. And 15 kyu is very weak for a human player. In amateur ranks, there are two types of ranks: dan and kyu. Dan ranks get stronger with higher numbers and kyu ranks get weaker with higher numbers. The weakest dan player (1 dan) is one rank stronger than the strongest kyu player (1 kyu). Amateur ranks: Very strong 8 dan < 2-7 dan in here > Moderate 1 dan Moderate 1 kyu < 2-9 kyu in here > Weak 10 kyu <11-19 kyu in here > Very Weak 20 kyu A human player that works at it a little can normally reach the 20 kyu level in less than 2 months. With work, a human player can reach 10 kyu in a year or so. All this depends on the amount of work, the quality of one's opponents, the quality of teaching, and (of course) ability. A human player who has never played and has just had the rules explained to him could be considered to be in the 30-40 kyu range - but ranks weaker than 20 kyu are *very* shaky because of their great variability from game to game. Ranks weaker than 10 kyu are fairly shaky - these players can perform quite differently from game to game. Professional grades in japan run from 1 dan to 9 dan (Higher numbers are stronger). I will just say that professional players are *very* strong and leave it at that. If you are a chess player, here is my (very rough) psychological comparison of chess ranks with GO ranks. Of course, take this with a large lump of salt ... Amateur Go ranks vs. chess ranks (my estimates) 6 dan 2300 (master) 1 dan 2000 (Bottom of expert range) 5 kyu 1600 (Bottom of B class) 10 kyu 1300 (Middle of D class) 15 kyu 1000 (Bottom of E class (I think ...)) Doug Larson ( 1 dan ) hplabs!hpda!dvl