Posts:
260
Registered:
Dec 16, 2001
From:
Maryland
Age:
53 Home page
Updated version of my windows Pente program
Posted:
Oct 13, 2002, 4:26 AM
I have put a slighly updated version on my website (http://users.erols.com/msmammel/marksfiv.html). No major improvements or additions yet...but I will have to do more in the future of course... The best improvements are the updated opening book and the ability to select up to level 18 (for those of you with 2.4 GHz computers
Pente 10.4 ( 10/10/02 ) Fixes: -loads in "keryo-pente" games from DSG and "k-pente" multiplayer games correctly -uses and reads more DSG fields in game info. -remembers proper colors of captured stones in multiplayer. -prevents re-swapping by computer after you swap in D-pente -help file slighly improved - hit F1 which brings up a worthless screen but then just click on Contents.
Improvements: -allows searches up to level 18 -screen is maximized when started -S-pente added -larger, updated opening book included
Re: Updated version of my windows Pente program
Posted:
Oct 24, 2002, 11:53 PM
I love this mark! Thank you very much for updating the program. For the moment level 18 is still out of reach of my machine(only 1.7 GHz), but it won't be too many years before it becomes practical.
Dmitri- I can run level 15 overnight and it will actually compute a move. If I run level 18 it is still at 0% the next day. Now that is deep thought!
Re: Updated version of my windows Pente program
Posted:
Oct 25, 2002, 4:02 AM
I actually completed a game against level 12 on my pathwtic 500 MHz machine... it only took the computer 5 hours to make its moves...
THen I tried one on level 18... It made 4 moves right off the bat, then it started thinking for 13 hours for its 5th move... and it was still at 0% of the first move it analyzed, so I gave up.... but I switched the level back to 12 and let it think about the same position, and it has been thinking for about 5 hours evern at level 12...
so I guess it is a touhg position. I played at level 8 and won rather quickly.
I need a faster computer.
If I do not accept a game invite right away, it means I will once I have fewer games in progress.
Posts:
1,032
Registered:
Dec 16, 2001
From:
Powell, OH
Age:
37 Home page
Re: Updated version of my windows Pente program
Posted:
Oct 25, 2002, 3:26 PM
You know what would be really cool? A distributed approach to calculating pente moves. We could use the computer power of many pente enthusiasts to cut down on the calculation time for pente AI's.
Of course this is a very difficult problem, much harder than anything done on this site so far. And I'm not even sure how it would work or if it is possible to split up searching the same game tree? Ideally it would be integrated with the current pente server and on the backend the move calculations could be spread out over many computers on the internet. Anyways, its probably not going to happen anytime soon on the site, but its an interesting idea.
Re: Updated version of my windows Pente program
Posted:
Oct 25, 2002, 7:25 PM
Dweebo- For my Parallel Programming class in College, I grabbed a copy of Mark's old DOS pente game source code and modified it to run on a cluster. My approach was to have each processor in the cluster grab a different move at the top level of evaluation (i.e. if there are 20 processors, you have each one take one of the first 20 moves that the computer thinks are good, and calculate them down the game tree separately)
Since the programming paradigm I was working in was a message passing model this didn't work very well. The reason is that in a traditional single processor execution the processor fills in a hash table of previous moves looked at. This means that if the program finds itself looking at the same exact position that it arrived at from a different path earlier then it simply takes the already computed value for the position from the hash table. This saves a LOT of evaluation time. In a message-passing model like you were talking about(distributed computers), this would mean somehow sharing this hash table... I didn't think of an easy way to do this in the time I had for the class but if as a group we could think of a way to do it, we might be able to set this program up!
Posts:
1,032
Registered:
Dec 16, 2001
From:
Powell, OH
Age:
37 Home page
Re: Updated version of my windows Pente program
Posted:
Oct 25, 2002, 8:04 PM
Yes I remember looking at your paper briefly last time I visited Mark, pretty good idea for a paper
I understand the problems you mentioned about storing common data. Unfortunately I think any sharing of hash keys over the network would be slower than having each processor calculate it for itself. I can also see a problem of multiple computers ending up examining the same positions since pente games often end up at the same position through different moves (especially in the beginning). That problem could be solved by storing which positions are already being looked at in a central server.
I can forsee other problems and potential solutions as well. One problem, more of a limitation actually, of taking the first 20 moves is what if 19 of them end up to be bad at the next move. Then those 19 processors are wasted when they could be looking at other moves deeper in the tree sprouted from the 1 good move. In other words, there needs to be some sort of probability that a given move is good and communication between all processors so that bad moves are quickly ignored and better moves are examined more closely. For example one processor might find a good move, say with probability 0.8 and broadcasts this to all other processors saying "hey check out this move, I think it might be a good one". Then other processors could say "oh, 0.8 is much better than I have seen so far, I'll starting looking at a subtree of that move." Of course I'm sure Mark knows more about that sort of thing, I could be way off. For one, I'm not sure its possible to assign an accurate probability to any given move.
I'm sure there is a way to solve all these issues, but it would take plenty of hard thought and work, but of course it is a much more interesting problem! I am currently not working on pente, taking a break for a few months since it takes so much of my time. But I can see that this sort of thing might bring me back. I am currently taking an operating systems class that has been focusing on distributed computing, maybe in the spring I'll take an actual distributed computing class and learn some things about this.
