Author: Alessandro Damiani
Date: 00:48:31 09/07/99
Go up one level in this thread
On September 07, 1999 at 02:58:47, Bruce Moreland wrote: >On September 07, 1999 at 02:41:02, Alessandro Damiani wrote: > >>Hi Bob! >> >>Some time back you wrote here on CCC about the move ordering of Crafty. I don't >>want to download all archive files, so I ask you now. >> >>If I remember it right, you use a SEE to order captures in the search. The >>negative captures (SwapOff(.)<0) are placed after ALL the other moves? >> >>I thought, the "state of the art" is: >> >>1. transposition table move >>2. positive captures (MVV/LVA) >>3. killer 1 >>4. killer 2 >>5. counter move >>6. rest of captures (MVV/LVA) >>7. history moves >>8. rest of moves (static ordering) >> >>I define "positive capture" this way: >>Val(aggressor)<Val(victim) OR NOT Attacked(opponent, move.to). >> >>You said that the SEE gives you a better move ordering when the negative >>captures are put after number 8? >> >>Thank you! >> >>Alessandro (new E-mail address since today: adamiani@econophone.ch) > >I am not Bob, but I have found independently that it does. It is worth trying >both ways. > >This orders "sacrifices" behind non-capturing moves, but most losing captures >are crap. If you no winning captures in a position, one losing capture, and a >bunch of random moves that do nothing, you probably have a better chance at a >cutoff if you do one of the other moves. Also, if you start capturing, you tend >to capture again at the next ply, which means you do more nodes. > >Speculation on my part. > >bruce Hi Bruce! Qualified answers are always welcome! But why "[...] It is worth trying >both ways. [...]"? This move ordering question concerns captures. Is one way not statistically better than the other? Since most losing captures are "crap" the variations they belong to are cut-off by forward pruning. Of course the cut-off is three plies after the losing capture. Are not most quiet moves "crap", too? I have developed a new quiescence search. Move ordering influences the performance of it. Alessandro
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.