| Author |
Message |
Andrew Juell (Ajuell)
| | Posted on Monday, April 16, 2001 - 9:36 pm: | |
I have a few asymmetrical games I'd like to playtest CPUvCPU for balance, but I usually have a lot of other stuff running with erratic heavy CPU usage...would it be simple to have an option to search to a fixed depth (and then apply quiescence or whatever search extensions ZoG usually uses) or to search a fixed number of nodes before terminating, rather than relying on a timer which could give one side substantially more 'real' time to think? |
Patrick S. Duff (Pduff)
| | Posted on Tuesday, April 17, 2001 - 12:18 am: | |
While we're on the subject of Zillion's move search, I have several improvements to request. First of all, three minutes is not enough. I want to be able to choose up to thirty minutes search time per move. Suppose the moving player has only one legal move, or Zillions has determined that all other moves lead to a loss. I want Zillions to make the forced move and start thinking about the opponent's move. Instead it just sits there thinking deeper and deeper into the move tree. Here's another problem I've seen frequently: Suppose there are ten legal moves, but only two of them prevent the other player from winning on their next move. Further suppose that Zillions has a best move which leads to a loss-in-five, but it hasn't exhausted the move tree yet. Why do I see it generate millions of nodes considering each of the ten legal moves in depth, instead of quickly eliminating the eight loss-in-one moves and only considering the two remaining moves in depth? If the opponent has a winning response, it shouldn't matter that they also have other possible responses -- Zillions should assume the opponent would make their winning move, prune that branch of the move tree, and immediately go on to search a different move. While Zillions is searching, I'd like a toggle option to see all of the moves in the current "Best: " move sequence, or the current "Searching... " move sequence, or both, not just the first move in each one. Finally, I see that Zillions only uses a small amount of memory, even when searching deep in the move tree (millions of positions). I want to be able to save the current search state and exit Zillions. Then at some later time, I want to start Zillions, load the search and resume it. |
Andrew Juell (Ajuell)
| | Posted on Tuesday, April 17, 2001 - 1:41 am: | |
Not to mention...there MUST be some way of getting Zillions to save the partial move-tree branching from the move actually made for use in later analysis, right? Even in something like double-move chess, it just starts its analaysis from scratch every new sub-move...didn't it already explore this line of play to N-ply just a few seconds ago? Why throw all that information away? |
David GLAUDE (Glu)
| | Posted on Tuesday, April 17, 2001 - 12:58 pm: | |
Well, Some of us want to use Zillions for scientific, game solving, game testing, opening books testing and creating, ... But obviously there are some annoying limitation that make our live more difficult that it could. I think we need to remember that Zillions of Games market if Mr. everybody that like to play games on computer from time to time and that is sick of searching for free/share/pay-ware for every game he would like to play. I believe ZRF writer and game addict/researcher are a very small percentage of the user community. However, if some easy to implement/nice to have feature could be added to help us, it would be nice. If some inside information on the way Zillions is thinking: alpha-beta/what-optimisation, fixed-depth? is a partial move is a ply or the full move is a ply? What does the :-o :-) :-( "lost in 4", "win in 3" mean. Why is memory usage not increasing over time?... It would really help understand our tool. As for request of feature, I would love Zillions to have a "memory"/"opening book" capability. I am thinking about a repository of played games with result and possible anotation of mistake/good move. This could be use next game to avoid making the same mistake again. Of course this might require human intervention, but I don't like winning always the same way(s) what ever the level. I would like my opponent to improve (because I am improving) over time. David GLAUDE glu@who.net |
Jeff Mallett (Jeffm)
| | Posted on Tuesday, April 24, 2001 - 6:21 pm: | |
>...would it be simple to have an option to search >to a fixed depth (and then apply quiescence or >whatever search extensions ZoG usually uses) You can do this now. Go to the Computer Opponent dialog and... * Set the search time to infinite * Set the variety to none * Set the strength according to the fixed depth you want (the minimum setting is a depth of 1. This will allow you to compare as you want. |
Jeff Mallett (Jeffm)
| | Posted on Tuesday, April 24, 2001 - 6:54 pm: | |
>Suppose the moving player has only one legal >move, or Zillions has determined that all other >moves lead to a loss. I want Zillions to make >the forced move and start thinking about the >opponent's move. Instead it just sits there >thinking deeper and deeper into the move tree. You must be mistaken about what happens when the moving player has only one legal move. Zillions plays the move instantly regardless of the settings. In the case where it has only one move that doesn't lead to a loss, it will consider its move for as the long the time settings dictate that it should search. As far as Zillions knows at that point, all moves may lead to a loss, but it simply hasn't found the loss in that one line yet. If you want Zillions to make the forced move and start thinking about the opponent's move, you should hit the "Move Now" button. >Why do I see it generate millions of nodes >considering each of the ten legal moves in >depth, instead of quickly eliminating the >eight loss-in-one moves and only considering >the two remaining moves in depth? If the >opponent has a winning response, it >shouldn't matter that they also have other >possible responses -- Zillions should assume >the opponent would make their winning move, >prune that branch of the move tree, and >immediately go on to search a different move. Zillions does assume the opponent will make the best moves previously found. However, due to a concept called "iterative deepening", it often goes back to moves it previously found to be bad and re-examines them a little deeper to see if they really are as bad as it previously thought. Because of pruning techniques these searches take little time compared to searching deeper in the main line. To put it another way, the move that looks like it loses a piece outright at a depth of 1, might look like a winning piece sacrifice at a depth of 8. |
Patrick S. Duff (Pduff)
| | Posted on Wednesday, April 25, 2001 - 10:42 pm: | |
If Zillions has determined that all moves except one loses, but it isn't sure yet about that last move, I'd like it to go ahead and make that move without going any deeper in the game tree. Others might prefer the current behavior. Let me suggest that this be made an option the user can toggle on/off. |
Patrick S. Duff (Pduff)
| | Posted on Wednesday, April 25, 2001 - 10:51 pm: | |
I'd like to be able to turn off the thinking brain animation, but still display the search progress information. The animation was cute for about the first five minutes I saw it, but I soon got bored with it. Now I actually resent it a bit, because I just think about it wasting a tiny bit of CPU time every second it runs. For this reason I usually turn off the search progress bar for hours at a time, turning it back on briefly only long enough to check on progress. I know it's a very small annoyance, but there it is. |
|