| Author |
Message |
Mats W (Kålroten)
| | Posted on Saturday, March 04, 2006 - 8:10 am: | |
The person who was fast enough to download 4-handed Chaturanga before I sent Ed a bugfixed version should download it again.
The bug consisted in that one of the pair of dice can never show a 3.
Mats |
Greg Schmidt (Gschmidt)
| | Posted on Saturday, March 04, 2006 - 2:50 pm: | |
I believe that would be me. Thanks -- Greg |
Mats W (Kålroten)
| | Posted on Monday, March 13, 2006 - 1:10 pm: | |
This game seems to appeal to people. It is rather fun. But I don't understand why the engine behaves as it does. Why does it sometimes refrain from capturing an enemy king? So I don't understand how I should improve it. I simply can't understand it. If some of you can find a way of improving it it would be good. And sometimes it happens that it captures its partner's pieces, but this treachery is like real life, so it's ok. This game is fun for kids, I suppose, and ideal when playing for money. Much better than poker.
Mats |
Greg Schmidt (Gschmidt)
| | Posted on Monday, March 13, 2006 - 2:51 pm: | |
[And sometimes it happens that it captures its partner's pieces,...]
Zillions doesn't have any notion of players that are partners. I believe it uniformly considers all other players in the turn order to be unfriendly. Would you agree that is what is going on and have you been able to find any tricky techniques to work around that?
Nice game BTW.
-- Greg |
Mats W (Kålroten)
| | Posted on Monday, March 13, 2006 - 3:24 pm: | |
Yes, I punish him if he captures his partner's pieces, and that works reasonably well. The big problem is that he refrains from capturing enemy kings sometimes, although I encourage it. It doesn't matter much, perhaps, since it's only a dice game, but it's irritating. |
Mats W (Kålroten)
| | Posted on Thursday, March 23, 2006 - 7:56 am: | |
4-handed Chaturanga. I fixed a serious bug concerning promotion. Playing strength is now improved, somewhat less unwilling to capture enemy kings, and positions the king better. By the way, there is a curious relative of 4-handed Chaturanga, called Enochian chess, which was played by members of the Golden Dawn movement, and was also used for divination purposes. |
Mats W (Kålroten)
| | Posted on Thursday, March 23, 2006 - 7:58 am: | |
Sorry, I game the link to the not yet updated 4-handed Chaturanga. The bugfixed version can be downloaded here. |
Greg Schmidt (Gschmidt)
| | Posted on Thursday, March 23, 2006 - 10:55 am: | |
On two occasions Mats wrote:
[But I don't understand why the engine behaves as it does. Why does it sometimes refrain from capturing an enemy king?]
[Playing strength is now improved, somewhat less unwilling to capture enemy kings, and positions the king better.]
Did you ever discover what the original problem was or did your recent change just happen to improve the situation?
-- Greg |
Mats W (Kålroten)
| | Posted on Thursday, March 23, 2006 - 11:47 pm: | |
I don't understand what the problem is, just increased the reward for capturing a king.
Mats |
Greg Schmidt (Gschmidt)
| | Posted on Friday, April 07, 2006 - 9:07 pm: | |
[And sometimes it happens that it captures its partner's pieces,...]
More on this...
1 - Random players:
Not knowing how Zillions utilizes minimax search when multiple players are involved, I can only speculate. However, it does seem reasonable to assume that when random players are involved, that should pretty much destroy any real effective lookahead. I wonder if the search engine actually uses the same random sequence as in actual play (in other words, after completing the search, does it reseed the random number generator back to what it was prior to the search?) - does it have perfect information about what random numbers will actually be drawn when the move is played out? If so, that seems like cheating. So when random players are involved, either the search is useless, or it is unfair.
2 - Multiple players:
Also, I speculate that the engine is lumping all unfriendly players together as a single "super opponent" which can make the union of all possible unfriendly moves. I can't really see how else Zillions would effectively deal with multiple players - minimax is really designed for two players. If you think about it, the engine would lose information regarding the sequence of moves when it flattens all opponents into the super opponent thus further degrading performance.
Although I can't provide a concrete scenario, perhaps it is not capturing the king because it sees a larger, non-existent threat looming from the super player.
Are there any other theories on how the search engine deals with random and multiple players?
-- Greg |
Greg Schmidt (Gschmidt)
| | Posted on Friday, April 07, 2006 - 9:38 pm: | |
A couple more thoughts...
1 The search engine could compute all possible outcomes for random players. In other words, rather than just consider one roll of the dice, it could consider all outcomes. In that case, you would expect the engine to minimize the damage done by the the opponent's "luckiest" move. That would seem to cause the engine to play very passively since it's always trying to protect itself against the worst case random scenario, which generally is unlikely to occur. So now the options for search with random players is a) Useless, b) Unfair, c) Passive.
2 Multiple player games - especially those involving random players are rare:
I doubt there are many multiple player (more than two non-random players) games in the "free games" section other than yours. Does anyone know if there are any? Furthermore, if there are others, it's even more doubtful that they involve random players (someone please correct me if I am wrong about this). My point in mentioning this is that 4-handed Chaturanga could easily be pushing the limits of what Zillions can effectively handle - even for games with low branching factors.
-- Greg |
Mats W (Kålroten)
| | Posted on Sunday, April 09, 2006 - 4:57 am: | |
Hmm...I could always solve this problem by creating a variant where king capture is forced. Normally one would always want to capture an enemy king. This game would be entertaining if Zillions played it better. |
Dan Troyka (Dtroyka)
| | Posted on Friday, April 14, 2006 - 11:16 pm: | |
Grand Camelot is also a four-handed game in which the players form teams. I dealt with the problem of partner cooperation by defining only two players each of which controls two armies. A special dummy piece toggles from turn to turn to determine which of the two friendly armies actually gets to move on a turn. This solves the general AI problem of partners treating each other as enemies -- they'll co-operate because they're owned by the same player. I'm not sure how much this method would help 4-handed Chaturanga. The dice will likely severely limit the computer player no matter what you do. |
Mats W (Kålroten)
| | Posted on Friday, April 14, 2006 - 11:54 pm: | |
This could be a good idea, but then four players can't play the game over the net(?), and one cannot capture one's partner's pieces(?). Especially in Camelot it would be convenient to capture one's partner's pieces as part of a knight's charge, to get at the enemy pieces behind. (I haven't tested Grand Camelot.)
Mats |
James lee mckigney (Gryro)
New member
Username: Gryro
Post Number: 1
Registered: 1-2009
| | Posted on Saturday, January 17, 2009 - 5:44 pm: | |
is there a possibility of 4 handed chess and it's variations? i really love that variation. |
|