Author: leonid
Date: 11:28:58 03/06/00
Go up one level in this thread
On March 06, 2000 at 11:41:57, Tom Kerrigan wrote: >On March 06, 2000 at 07:27:38, leonid wrote: > >>I hope that you will change your idea by the time and will go into complet chess >>game creation. Many things from the creation of mate solver could be very useful >>there. I found, for instance, that nobody put its checking moves at the head of >>the chain of moves. And this idea came simply from the mate solver. I am sure >>that you came to the same alignement of the moves. > >No, this just hilights the difference between a mate finder and a chess program. > >I changed the move ordering in my program so checks are ordered high. The result >is that my search trees increased in size by 34% (sometimes less than 10%, other Here not everything is clear to me. Are you putting all the time checking moves at the head of the moves ligne or not? In my mate solving logic it is all the time with exception of two plies. For now I don't remember exectly advantage that this alignment of moves give but it was done only when statistics said to me to do so. I never did anything in my game only after my personal taste. >times more than 100%). > >This doesn't take into account how much more work the move generator has to do >to see if a move is a check or not (resulting in fewer NPS). This I don't know. At that time I had no counter in this logic. Even now, when I have my nodes counter installed, it is working only in positional logic. >So I'm sure that ordering checks high is very helpful in a mate finder, but it >seems pretty stupid to do in a chess program. Will be happy to know for sure what this order really mean. In positional logic it is not that simple to say primarily because in two plies (that are the most reponsable for the speed) one ply "demand" to put all the checking moves at head of the line and the second one is the next exception. The last one find only moves that contain material advantage (taken of piece, promotion) and use moves not verified in move generator. Recently, when I found that, in reallity, each ply see only around 2.5 moves, tried to use not verified and not check at head aligned moves. Result was mixed. In much more occasion, compared with previous logic, time advantage was there but losses in general were twice as important. Lately expect to come back to this part and try once again. Speed of the positional logic, like I have it now, give the impression that my way of alignment of moves is correct, even if it could be better. NPS rate is really good, if not the best one, and raw speed up to the 9 ply, probably, the best ever. After 9 ply game start loosing because of needed improvement in branching factor. Even here I am not sure where is the problem. It could stay far away from my alignment of moves in positional logic. What should be changed there? For now I still don't know. >(In other words, there is a good reason why nobody does this in their chess >programs.) Maybe everybody is right here as well as everybody is wrong. Even to write mate solver for the chess game sound so new and wrong to everybody. For me it is very far to be so. Leonid. >-Tom
This page took 0.01 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.