Author: Marc van Hal
Date: 16:05:06 05/13/02
Go up one level in this thread
On May 13, 2002 at 10:20:22, Vincent Diepeveen wrote: >On May 12, 2002 at 22:50:03, Christophe Theron wrote: > >>On May 12, 2002 at 01:37:40, Sean Mintz wrote: >> >>> >>>>We have reached GM level and are slowly approaching the super GM level and you >>>>want us to throw everything to the trash bin and start over with radically new >>>>ideas? >>> >>>No no no! I do not want us to throw everything away! It's just my opinion that >>>there is too much effort in to what we all already know and we aren't really >>>trying anything new. And I don't mean this like one of those ''anything is >>>possible'' speeches but main focus of the computer chess world has been very >>>small. Someone must agree with me. >> >> >> >>Actually I agree with you. Some time ago I was complaining against the high >>number of Winboard engines that are all essentially the same rehashing of known >>techniques. >> >>It is rewarding to use these techniques because with them many programmers can >>write chess engines that are quickly playing stronger than their authors. >>Creating a chess program has become an interesting programming exercise, nothing >>more. >> >>But I think if someone wants to print his name in the field, he should try >>something different. Maybe not completely different, but at least bring >>something new. I think that the way human player search the tree is extremely >>interesting and his not even remotely approached by current computer chess >>programs. > >the main problem is that i as a human search completely based upon >knowledge and experience. The only time i am 100% sure i didn't miss >anything i did a kind of 2 ply alfabeta search. > >The mainline checked out real deep then with very shallow searched >trees. > >Also i catch myself and many others upon the fact that even the move >played in the game, the mainline has not been clearly examined. Just >a 'global' look whether there is no general trick. > >Such things are all 'knowledge' based and demand a rather intelligent >approach. Nothing brute force in short! > >So searches that do not focus upon knowledge (CNS for example) or >trust upon bruteforce (the well known search ways) they are completely >chanceless IMHO. > >>Learning is also almost totally absent from computer chess. That's another area >>to explore. > >Did you have in mind position learning or booklearning? > >Best regards, >Vincent > >I am not a programmer but you mentioned Alpha Beta and learning Which brings me to an idea which might be intresting to help you on your way. What if a program has played a game and the log file is stored and it did see that there was a diference in the valeu of the position from what it expected before. If it was higher it should be stored as A if it was lower it should be stored as B. But in the maintime it could compare the moves which where played by the oponent and if this move was better then expected it should be stored as a if it was worse as B But there also should be C if it was a move which valeu did not realy difers from the moves expected. (If this is treu the moves should be reanalysed deeper if there realy was no diference in the valeu of that move. You also could make it analyse a database of high qualety games For more position learning in the same way to take out the good and the bad positions. The only bad part about it should be that after the program has played or analysed a game it should need some time to work this all out. Deep positional understanding maybe can be increased a litle in this way. I am intrested in what you think about this idea or am I just talking air, or about a technic which is already been used? Regards Marc van Hal >> Christophe
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.