| Author | Message | ||
| Kieran Child (Doodinthemood)
New member Username: Doodinthemood Post Number: 1 Registered: 9-2008 |
When setting the computer to analyse a position, and it finds a win in 7, say, it would be good for it to show a solution tree, with the answers to every line outlined, on command. This may already be possible? (surely this tree is hidden in the computing anyway?) but I cannot find how to discover it but by manually playing moves. | ||
| Greg Schmidt (Gschmidt2)
New member Username: Gschmidt2 Post Number: 78 Registered: 1-2007 |
Presumably you want to do this in order to answer "what if" types of questions (eg. "what if I had moved to position y instead of x - how is that still a loss for me?") The only way I know to do this is by saving the game, reloading, then backing up to that point and trying the alternate move(s). Then save the new game and repeat the above for the new position. If you do this enough times, you could manually construct the end game tree by observing the behavior. That would be tedious for many games, however in many situations you needn't go that far to answer the limited set of questions you might have. -- Greg | ||
| Keith Carter (Keithc)
New member Username: Keithc Post Number: 165 Registered: 8-2000 |
Greg, is this a principle continuation issue? Zillions does not have a way to show what moves constitute the win in 7. It would be nice if it did. One way to see what Zillions had in mind is to set it, at that point, to play both sides. It will play out the seven moves. Some of the moves you will see may be poor ones for the side headed for a loss. The Zillions AI will sacrifice pieces to delay the end of a game. | ||
| Greg Schmidt (Gschmidt2)
New member Username: Gschmidt2 Post Number: 79 Registered: 1-2007 |
[Greg, is this a principle continuation issue? ] Kieran mentioned "solution tree". By that I assume he means the remaining game tree rooted at the position that has been shown to be either a win, loss or draw. The game tree would include all possible moves from that point, followed by all possible replies etc. etc.. The "leaves" of that game tree would all be uniformly labeled with the same value win, loss, or draw. A display of this tree would allow one to answer many "what if" questions such as "If I play x and my opponent plays Y"... how does that lead to a guaranteed loss?" OTOH, the "principle continuation" (PC) is not a tree, rather it is a single path through the game tree. The PC from a given position indicates what the computer believes is the best single line of play by all sides starting from the given position. [Some of the moves you will see may be poor ones for the side headed for a loss. The Zillions AI will sacrifice pieces to delay the end of a game.] Once the AI sees that it has no way to win, then all moves look equally bad. In a technical sense, it really doesn't matter what move is played if they all lead to a loss, even the seemingly dumb looking moves. However in a human sense, both Zillions (and Axiom) attempt to delay the loss as long as possible in the hopes that a human player will blunder. More sophisticated programs (eg. Chinook) will use additional heuristics to make moves leading to "complex" positions as a way to put further stumbling blocks in front of the human player. There are some positions where no matter how badly the human player blunders, they wouldn't be able to lose if even if they wanted to. In that case, there's little point in delaying an inevitable loss other than to give the appearance of a good fight. IMO, the computer would be better off resigning in that case. -- Greg | ||
| David J Bush (Twixter)
New member Username: Twixter Post Number: 45 Registered: 7-2002 |
I believe you mean "principal continuation" which means "main or most important continuation" and not "principle continuation" which might have something to do with some rule or principle, I'm not sure. Sorry for being a spelling Nazi, but the difference might make a difference in this instance. Even in what appears to be a relatively simple position, a full solution tree could be very large. When experts provide analysis, they list a few branches and hope the reader is smart enough to fill in the blanks. Zillions does not provide a tree of possibilities for you to examine, but if you follow Greg Schmidt's suggestion and play out whatever position you are interested in against the computer, you may learn what moves work and what moves don't. If you think one side is losing, switch sides and see if you can force a win. | ||
| Greg Schmidt (Gschmidt2)
New member Username: Gschmidt2 Post Number: 80 Registered: 1-2007 |
[I believe you mean "principal continuation" which means "main or most important continuation" and not "principle continuation" ] Ah yes, spelling error on my part. We are indeed referring to the same concept. I suppose I was describing the "principle" behind the "principal" continuation.  -- Greg | ||
| Kieran Child (Doodinthemood)
New member Username: Doodinthemood Post Number: 2 Registered: 9-2008 |
What I was suggesting was something in the way of solutions as seen in answers to chess problems. They don't give all the possibilities for both sides, but all the possibilities for the losing side, and the correct response for the winning side. Suppose that to win a game of chess, you have to get any piece to the fifth rank. Then a solution would take two moves, which would lead to a full possibility tree of over 8,000 variations, but if you only pick one winning line for the winning side, then it would look like this: 1. e4 a6 2. e5# if 1... a5 2. e5# if 1... na6 2. e5# and so on, giving you a solution with only 20 variations. Still a bit excessive, but managable nonetheless. I am currently doing what Greg suggested when building solutions, but it would save so much time for zillions to just show me why it thinks there's a win in 6. Thanks for the very quick responses | ||
| Greg Schmidt (Gschmidt2)
New member Username: Gschmidt2 Post Number: 81 Registered: 1-2007 |
Although there are some similarities between Chess AI and Chess problem solving, they are fundamentally different algorithms. It would be possible to incorporate both algorithms into a Chess program, but obviously Zillions does not do that. Furthermore, using Zillions, I'm unaware of any trick to simulate what you are looking for beyond a very tedious setup and try approach which is untenable for all but the simplest positions. My best suggestion would be to find a specialized program that is specifically designed to solve Chess problems. A google search turns up a freeware program called "MateMaster". You might give that a try. You will have to manually enter the position of interest into the program. If you know a little programming, it might be possible to write a small program which translates a .zsg file for Chess into the required format for the solver. -- Greg | ||
| Kieran Child (Doodinthemood)
New member Username: Doodinthemood Post Number: 3 Registered: 9-2008 |
I'm ok for chess problems (I have fritz). I've just found myself using zillions as an assistant solver for connect four, dotsNboxes, etc. and thought this feature would be helpful and easy to do. I was reluctant to give this reply in fear that you've put so much effort into the previous replies and I wouldn't want you doing that for this one It was just a suggestion to the design team. I'm not a programmer myself, and was rather confused by some of the words being used, but I wasn't really expecting any replies at all, so thank you. | ||
| Greg Schmidt (Gschmidt2)
New member Username: Gschmidt2 Post Number: 82 Registered: 1-2007 |
OK, I promise not to put much effort into my reply. Unfortunately, the Zillions design team appears to have flown the coop so as far as the designers are concerned your suggestions for improving the product will fall on deaf ears.  -- Greg | ||
| Matthew King (Doodinthemood)
New member Username: Doodinthemood Post Number: 4 Registered: 9-2008 |
Aw, was hoping an amateur might be able to rustle up something. Ah well. |