| Author |
Message |
Fergus Duniho (Fduniho)
| | Posted on Tuesday, August 22, 2000 - 7:22 pm: | |
Yes, Zillions can already play against itself, but only with the same settings, and only the very same ZRF. I would like to run Zillions twice and let them communicate back and forth with each other, so that they could play games with each other. They could send only a minimal amount of information back and forth, namely the move in a game. This would be useful for pitting against each other different ZRFs that play the same game and use exactly the same notation. This is useful for testing how well different ZRFs play the same game, which is important for games like Shogi, which Zillions doesn't play well without some tweaking to the ZRF. I can already do this testing by hand, but it is so time consuming. I would like to just have Zillions play the different ZRFs against each other on its own, freeing me to do other things. |
David GLAUDE (Glu)
| | Posted on Tuesday, September 12, 2000 - 8:28 am: | |
YES,
I want that too.
In fact I would like to have the option to configure Zillions of Games to automaticaly accept challenge from a human opponent across the network and start playing the game.
That way, I could let peaple (registred user of Zillions) play against my PC (if permanently connected) to a game for wich I cannot release the ZRF file (copyright TM issue).
Also, if the network protocol was OPEN (like the DLL interface) it would let me try my version (C, C++, Pascal, ...) of a game against Zillions version using a ZRF.
It will also let me use Zillions engine in an automated way in order to analyse games opening, ...
MOST IMPORTANT: I will be able to make someone believe I play very well a game (where in fact it is Zillions playing ;-).
David GLAUDE |
Ken Franklin (Kenz)
| | Posted on Thursday, April 05, 2001 - 12:13 pm: | |
A silly feature:
A loop option to automatically reset and start another game - until stop is selected. An example is Tic-Tac-Toe from the movie, WarGames; where the Machine continously played the game against itself. |
Clint Gamlin (Cgamlin)
| | Posted on Friday, April 12, 2002 - 11:58 pm: | |
I think we can go one step further here.
One thing that annoys me about Zillions is that the AI engine needs so much tweaking to get it to play at its peak. And then you can go too far and make it SO good that there is no reason to play against it - you KNOW that it is going to beat you 90% of the time, so it seems like there is no point.
I suggest implementing a ranking system similar to that used in chess.
The computer can play against itself, with a number of different AI settings, and play against itself, ranking those settings and testing new settings in a 'natural selection' type way (i.e. a basic genetic algorithm)
Then, when you play against the computer, it works out your ranking (after a few games) and selects an AI setting that is a few rankings higher. As you improve, the computer slowly moves to its highest setting.
It would also help to keep a file on the ranking of each user on each game that they play.
If Zillions gets a central game server, it would also be possible to quickly swap the ranking-to-setting information, so that if one computer has found a better setting than you have, you can get a copy of that setting.
I see a lot of potential for Zillions, and a game server would help it immensely. |
Keith Carter (Keithc)
New member
Username: Keithc
Post Number: 144
Registered: 8-2000
| | Posted on Sunday, December 30, 2007 - 7:36 pm: | |
So it has been about 67 months since the last post on this topic. I am going to reply anyway.
FWIW the Axiom engine for Zillions has an autoplay file that does allow AI vs AI games with different game files and/or different play settings that will loop to play as many games as desired. |
Greg Schmidt (Gschmidt2)
New member
Username: Gschmidt2
Post Number: 46
Registered: 1-2007
| | Posted on Sunday, December 30, 2007 - 9:34 pm: | |
For Axiom, any "tweaks" between the pair of games must not alter the format of the move string that is generated as one instance is required to "understand" the other's moves. Both programs must speak the same language when it comes to interpreting a legal move. That's not strictly a problem when it comes to "tweaking" the AI. It might be a problem if the tweaks are done via altering the moves.
The original post mentions all sorts of specialized ideas such as using a genetic algorithm to try new settings and have the computer learn to match the opponent's skill. I think this underscores a recent post of mine where I suggest that there are an infinite number of these kinds of specialized features that various people desire that require an infinite amount of development time to support, in a specialized way, for a product such as Zillions. I'll reiterate that IMHO the best way to address these issues is by having a very open and progammable interface that exposes the fundamental "building blocks" of the system and includes a general purpose programming language. That way, a user could layer these "pet features" ontop of the existing base without having to hope for and wait for a future release from the developers.
I'm not suggesting that Axiom is the end all be all for that, but I think it is a definitive step in that direction. Much of what was suggested is already possible in Axiom if someone wanted to go through the trouble of writing the scripts to do this sort of thing.
One of my longer term goals is to use the autoplay feature of Axiom to train a neural network (and/or possibly a genetic representation) to learn an evaluation function for a specific game. This is possible today. |
|