Author: Uri Blass
Date: 06:35:50 01/23/03
Go up one level in this thread
On January 23, 2003 at 08:38:41, Ulrich Tuerke wrote: >On January 23, 2003 at 05:36:12, Uri Blass wrote: > >>On January 23, 2003 at 03:57:45, Peter McKenzie wrote: >> >>>Sounds like you are doing history heuristic. A couple of points (which you >>>might know already): >>> >>>1) Make sure you have a separate history table for each colour. >> >>I have only one history table. >>Is it important to have seperate history table for each color? >> >>Usually when from->to is often legal for white it is not often legal for black. >> >>I may also get rid of the history table in the future and I guess that there are >>better alternatives for order of moves. >> >>I read that some programmers (like Tony Worten) found that getting rid of >>history tables made their program stronger so I guess that they have other rules >>to decide about order of moves that are simply better. > >I still do use a history table. However, I am not really convinced about its >benefits. Use and Non-Use of history in Comet gives results, which don't suggest >that history is a real help. >In former times, before using a transposition table, a history table was much >more useful than today. TT makes it basically obsolete, IMHO. >OTOH, using it doesn't cost much either. So , I have it still in. I remember that for yace history table is clearly productive. I also do not understand why hash tables make history not productive because there are a lot of cases when there is not a move in the hash. >OTOH, using it doesn't cost much either. So , I have it still in. > >I even use some similar tables called "butterfly tables", which store moves >which had turned out to refute a special precdeding move at some time. The table >is indexed by the preceeding move. So - ever when I find that the "preceeding >move" is played I look into the refutation table to see whether there is a >potentail killer for this. However, its effects aren't sensational either. > >BTW, this is not my invention, but had been suggested by 2 Dutch authors (iirc, >Hartman - the author of dappet and some co-author) a lot of years ago. > >Uli I also thought about that idea independetly but did not try it. I believe that a lot of programmers thought about it. This idea is not number 1 in the list of ideas to try and at this time I guess that improving the evaluation can give me a bigger advantage. Latest Movei that beated cyberpango 4-0 has more knowledge than the public Movei in the following areas. 1)King safety(not big scores and part of the knowledge may be wrong but the knowledge seem to help Movei to play better) 2)evaluating passed pawns that are protected by pawns. 3)evaluating isolated pawns. My evaluation is still simple and latest Movei give a constant panelty of 0.1 pawn for every isolated pawn(I do not like it and I prefer to evaluate every isolated pawn based on other factors but I was afraid from new bugs so I still did not do it for WBEC). Movei also still does not evaluate backward pawns or passed pawns when they are not protected by pawns(there are cases when passed pawns are weak,I do not like to give them automatically a bonus). I believe that all of this may give me more than butterfly tables. Uri
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.