Author |
Topic: King's Gambit chess opening claimed as solved (Read 29420 times) |
|
christianF
Forum Moderator
Forum Guru
Arimaa player #4019
Gender: 
Posts: 804
|
 |
Re: King's Gambit chess opening claimed as solved
« Reply #15 on: Apr 7th, 2012, 9:21am » |
 Quote  Modify
|
on Apr 7th, 2012, 6:32am, Katsunami wrote:| Conceptually, engines for most two-player board games are the same. |
|
I'm not sure what you mean by that. My gut feeling tells me that MCTS isn't the way to go in Arimaa, whereas the traditional approach looks very unprospective in Havannah. This would suggests that different classes of games may be approached better by the one or the other method.
|
|
 IP Logged |
|
|
|
clyring
Forum Guru
Arimaa player #6218
Gender: 
Posts: 359
|
|
Re: King's Gambit chess opening claimed as solved
« Reply #16 on: Apr 7th, 2012, 9:39am » |
Quote Modify
|
on Apr 7th, 2012, 3:15am, Katsunami wrote:
Third reason is one that may surprise you. If I am completely honest, I would have to say that I don't really like Arimaa as a game (1); my only interest in it is with regard to writing an engine. I will probably never play it extensively, as I did with chess. The reason is that Arimaa exactly embodies what I hate most in chess: extremely slow positional play. In a game of chess, I often got very nervous, agitated (and sometimes even angry) when an opponent would play only positionally, keeping a close position, and avoiding all risks. |
|
Believe me when I say that not every game of arimaa is tidy with little room for tactical fuss. Endgames in particular tend to feature difficult tactical skirmishes- a "127 move long endgame" in arimaa would in all likelihood be the most active 127 moves you ever play. 
|
|
IP Logged |
I administer the Endless Endgame Event (EEE). Players welcome!
|
|
|
Katsunami
Forum Senior Member
Gender:
Posts: 29
|
|
Re: King's Gambit chess opening claimed as solved
« Reply #17 on: Apr 7th, 2012, 11:17am » |
Quote Modify
|
christianF: Too big of a mind jump; I tend to do that. Sorry  What I meant is that most board games have game playing engines with comparable structure: They all have a board representation, a move generator, a search function, and an evaluation function, and so on. Implementation of these functions is a detail.
clyring: I've never experienced that; the not so many games I played against other bots, where mostly very positional. Could also be me, of course, not seeing the tactical stuff
|
|
IP Logged |
|
|
|
clauchau
Forum Guru
bot Quantum Leapfrog's father
Gender:
Posts: 145
|
|
Re: King's Gambit chess opening claimed as solved
« Reply #18 on: Apr 14th, 2012, 12:54pm » |
Quote Modify
|
To me the most interesting problem in developing game engines lies in the evaluation function in some specific way...
Are you going to fiddle with a few geometric, algorithmic or intuitive local ideas proper to the game and your particular grasp of it and then feel good about the original result outgrowing your own understanding of the game and others' as well? That was my motivation 15 years ago.
Or are you going to try some original idea with the evaluation function or the overall search structure that may find applications to other games and fields, such as resistive networks or Monte Carlo Tree Searches? It grew up as my main motivation along the years, and Arimaa researchers here are quite oriented that way too.
Or are you even going to help mankind better understand games strategy in some general and well-structured way, with the help of special cases such as Arimaa? I don't see much of that, except with the Abstract Game Theory of Berlekamp and al and general search algorithms, but I'm not satisfied with them when I want to understand and evaluate most games we know. There is some good science waiting to emerge there and that's what seem's the most exciting to me today.
What do you think?
|
| « Last Edit: Apr 14th, 2012, 12:57pm by clauchau » |
IP Logged |
|
|
|
Katsunami
Forum Senior Member
Gender:
Posts: 29
|
|
Re: King's Gambit chess opening claimed as solved
« Reply #19 on: Apr 14th, 2012, 1:14pm » |
Quote Modify
|
The evaluation function seems to be by far the most important part of most game-playing programs.
When I wrote an engine for the game of Cartagena a few months ago (a bit simplified: all cards open, so it became a complete information game without luck). I tested this by having two engines play against each other: one with a very simple evaluation function ("Take the biggest steps possible"), but set to think 8 ply deep. The other engine had a much better evaluation function taking several things into account ("Closer to the end of the board is better", "Make sure the opponent cannot goal a piece", etc), and it was limited to 4 ply. It won every time, despite the huge depth-handicap
It can also be seen in chess. Run Fritz 13 on an old 2001 Pentium 3 computer, and have it compete against Deep Fritz 7, running on a 2012 computer. Most probably, Fritz 13 will be able to beat the computationally much more powerful setup running the older Fritz. The newer version is just way, way smarter.
As Arimaa prohibits deep thinking, I think that knowledge (so, the evaluation function) will be all important. This is already proven by P1-bots being stronger than some P2-bots in the game-room.
Thinking well, but less deep seems to be better than thinking deep without having much of a plan.
The entire problem with Arimaa is that not many people know how to play it well; and even the strongest players of today may be proven to play the game "wrong". Many idea's in chess that were thought to be good and were practiced for a long time were later proven not to be the best way to go about the game.
|
| « Last Edit: Apr 14th, 2012, 1:18pm by Katsunami » |
IP Logged |
|
|
|
christianF
Forum Moderator
Forum Guru
Arimaa player #4019
Gender:
Posts: 804
|
|
Re: King's Gambit chess opening claimed as solved
« Reply #20 on: Apr 14th, 2012, 3:51pm » |
Quote Modify
|
on Apr 14th, 2012, 1:14pm, Katsunami wrote:| The entire problem with Arimaa is that not many people know how to play it well; and even the strongest players of today may be proven to play the game "wrong". Many idea's in chess that were thought to be good and were practiced for a long time were later proven not to be the best way to go about the game. |
|
That may not be the entire problem with Arimaa (a broad player base with a relatively high summit considering its age) and at the same time it is a generic problem. I'm glad you address it.
The game for the next CodeCup Challenge will be Symple.
So I asked myself the question what would happen. Thirty or forty programs compete and one will win, eventually, no special gifts of prophecy required.
Such a result doesn’t mean much regarding the game’s ‘resistance’ to programmability if there’s no human opposition to provide a context.
So I started to play Symple to provide that context, as far as my limited abilities allow, after the Challenge has been completed, by playing against the winner.
Certain simple games (I'm referring to structure here) are difficult to program beyond the human level of play, and the question is why. Our vision on that is blurred by the fact that computer programs, MCTS in particular, often play a fairly good game without a particularly strong effort on the programmers' side. So beginners lose against the programs and the question doesn't emerge again unless humans are beating the programs. That in turn requires strong players. That in turn requires games players find worth playing. And where hypes and tactical funnies take their rightful place in this Fancy Fair, it may take a long time for a game that suits humans but defies programming, to emerge.
P.S. The Havannah Challenge is just a piece of that puzzle. If I can beat the bots ten out of ten it doesn't say bots won't win, eventually. But it says were not quite there yet, and there may be other havannah's, or arimaa's for that matter
|
| « Last Edit: Apr 14th, 2012, 4:03pm by christianF » |
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print