Author: Uri Blass
Date: 23:55:17 03/09/03
Go up one level in this thread
On March 09, 2003 at 21:36:00, Ulrich Tuerke wrote: >On March 09, 2003 at 18:13:31, Uri Blass wrote: > >>On March 09, 2003 at 17:56:55, Ulrich Tuerke wrote: >> >>>On March 09, 2003 at 15:43:37, Russell Reagan wrote: >>> >>>>I think that most people subscribe to the school of thought that says that >>>>quiescent search is not perfect, so do it fast and "good enough" for most >>>>situations. If you could have a perfect quiescent search, what price would you >>>>be willing to pay? One ply of full width search? Two ply? Time to depth takes >>>>twice as long? I am interested what programmers with more experience than myself >>>>think about this. >>> >>>I think, that's an interesting field to experiment with. I have tested a lot of >>>ideas in quiescence search: pushes of passed pawns (if some pre-conditions >>>fulfilled), checking moves (if king to be checked is condidered to stand risky), >>>hanging pieces attempting to escape (only very close to the boundary of full >>>search). >>>OTOH, you're right; i think that this kind of q-search is expensive. Only >>>extensive testing can tell whether it's worthwhile. I do not know yet. >>> >>>Uli >> >>I am surprised that for you it is expensive, >> >>Comet is not one of the fast searchers so I thought that for you it should be >>not expensive. > >I don't see your point. Why should a certain percentage of extra-nodes be >"cheaper" for a slower searcher than for a faster searcher ? The reason is simple. I assume that fast searcher may get significantly less nodes per seconds when it does checks in the qsearch when it is not the case for slow searcher. > >> >>I found it productive for test suites and I did not find a proof for big >>difference in games. >> >>I decided to keep it because I know that my function to generate checks in the >>first plies of the qsearch is not close to be optimal and I probably can get a >>significant improvement in speed by doing it faster in the future. > >I do it different. So, we can't compare easily. In case the evaluation has >signaled that either king is "in danger", then I generate checking moves (to >that king) also deep in quies-search. OTOH, in a quiet position, I don't do them >at all. > >These checks - together with escape of hanging pieces - slows down considerably. >I haven't really measured for a long time, but I guess this could account for >almost one iteration less in given time. same for me but one of the reasons is the fact that it caused movei to be signficnatly slower in nodes per seconds(It is more than 30% slower in nodes per second because of checks in the qsearch thanks to my way that is not efficient) I do not do escape of hanging pieces today but only escapes of the king or hanged pinned pieces(not pawns). I do these things also more deep in the qsearch but generating move that threats check is done only in the first 2 plies of the qsearch. It is possible that I may do better if I have rules when not to generate checks in the first plies of the qsearch but I have not these rules. 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.