Author: Uri Blass
Date: 07:17:46 06/04/02
Go up one level in this thread
On June 04, 2002 at 09:39:18, Vincent Diepeveen wrote: >On June 04, 2002 at 09:25:24, Uri Blass wrote: > >>On June 04, 2002 at 08:59:10, Vincent Diepeveen wrote: >> >>>On June 04, 2002 at 08:29:12, Uri Blass wrote: >>> >>>>On June 04, 2002 at 07:17:03, Vincent Diepeveen wrote: >>>> >>>>>On June 04, 2002 at 02:33:55, Russell Reagan wrote: >>>>> >>>>>I have been wasting a lot of time in search past years. >>>>> >>>>>Like last months i have wasted my time working on forward pruning. >>>>> >>>>>Sure i can search 3 ply deeper with it (and kicking out singular >>>>>extensions and other extensions). >>>>> >>>>>However if i do that, then it scores simply 20-25% worse in tests >>>>>against commercial programs. >>>>> >>>>>The reason was not so hard to find for me. It simply goes for 1 >>>>>move and stays with that move. The rest gets forward pruned. Chances >>>>>to find a superb move is way smaller. >>>> >>>>In this case it is a bad way of using forward pruning. >>>>The idea of forward pruning is to prune tactical blunders and not to prune >>>>positional moves. >>> >>>Fritz7 is doing opposite of pruning tactics. It seems to only prune positional >>>lines. Idem for junior. Junior only sees tactics deeper. It simply >>>gives for a simple rook move like Rab1 (if b-file isn't an >>>open file) 2 ply penalty (or better the move costs around 3 ply >>>to make). >> >>I do not think to try to do the same as Fritz and Junior. >>The first idea to save time is by pruning of tactical blunders and pruning of >>positional moves seems to be less simple but it does not mean that it is >>impossible. >> >>I also do not consider what Junior does as pruning(if your guess about it is >>correct). >>It can be described as not doing extensions in the case of 2 ply panelty and >>doing negative extensions in the case of 3 plies panelty. > >Well fact is that positional lines at depth=19 units sometimes only seem >to get searched 6 ply sometimes by junior. > >That's the biggest positional pruning any program is doing. Most likely >searching positoinal deeper doesn't bring them more because they are >based upon piece square tables. I do not believe that Junior is based on piece square tables. I believe that it is possible to improve the evaluation of programs significantly but there are simple ways to add knowledge(not piece square table). I believe that every top program evaluates mobility in some way. It does not mean that the evaluation is complex. > >>Pruning for me means simply to return evaluation without searching of legal >>moves of the opponent. >>Except null move pruning I do it only when the remaining depth is >>small(otherwise it is very risky). > >That doesn't prune enough to get a bunch of plies deeper though. I agree but it can give some speed improvement. > >What fritz is doing is getting it 3 ply deeper or so and what junior is >doing is getting it 5 ply deeper or so. I think that counting plies is not important and for Junior depth is not plies. Search is important but showing bigger nominal depth is not important. 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.