I think this may be overkill because a large percentage of q-search nodes will cut off due to static eval, or due to one capture that is good enough, or they will be really static so no candidate moves are available. So I expect the number that get to the 3rd, 4th or 5th move w/o anything happening is relatively small.
That said like anything else testing will tell if this is a good idea or not.
--Jon
Pruning in QS
Moderators: hgm, Rebel, chrisw
-
- Posts: 4366
- Joined: Fri Mar 10, 2006 5:23 am
- Location: http://www.arasanchess.org
-
- Posts: 27798
- Joined: Fri Mar 10, 2006 10:06 am
- Location: Amsterdam
- Full name: H G Muller
Re: Pruning in QS
Yet it has been shown that move sorting in QS has huge impact, and even subtle things like using SEE for ordering has measurable impact. With random ordering you almost always reach a point in the game where the search depth drops from 11-12 ply to 1-2 ply, due to QS explosion.
Plunder raids usually kill you, and some variants are far more susceptible to plunder raids than Chess. And as Daniel remarks, piece drops are a big problem as well. Considering too many checks also would make the QS prone to explosion.
Plunder raids usually kill you, and some variants are far more susceptible to plunder raids than Chess. And as Daniel remarks, piece drops are a big problem as well. Considering too many checks also would make the QS prone to explosion.
-
- Posts: 20943
- Joined: Mon Feb 27, 2006 7:30 pm
- Location: Birmingham, AL
Re: Pruning in QS
I actually tested this, and while I don't remember the specific numbers, weeding out equal captures is an Elo loser. You don't want to evaluate something like an outpost knight if it can be removed safely by the opponent (NxN PxN or whatever). It also helps recognize overloaded pieces and such.Rebel wrote:My experience is that every equal capture should be investigated and that the costs for doing so are to neglect. It's a long time ago and although rare on the bigger scale I have seen quite some horrible tactical blunders because the lack off it in the past.Ron Murawski wrote: * Don't try equal captures if the victim is protected
If there are more attackers then defenders, then this 'equal' capture is probably winning. My opinion is that SEE is needed here, or some sort of attack board.