Author: Bas Hamstra
Date: 14:41:54 01/21/02
Go up one level in this thread
On January 21, 2002 at 17:28:11, Antonio Dieguez wrote: >On January 21, 2002 at 17:18:03, Bas Hamstra wrote: > >>On January 21, 2002 at 16:47:09, Antonio Dieguez wrote: >> >>>On January 21, 2002 at 16:30:22, Bas Hamstra wrote: >>> >>>>On January 21, 2002 at 15:56:59, Antonio Dieguez wrote: >>>> >>>>>>It should matter in quiet positions, right? >>>>> >>>>>Well, It should matter depending on how often the pv changes, so may be more in >>>>>unquiet positions. >>>>> >>>>>>Well, here is data, opening >>>>>>position: >>>>>> >>>>>>Rootwindow <-inf, inf> takes 2404K nodes >>>>>>Rootwindow <-50, 50> takes 2420K nodes >>>>>> >>>>>>...to complete 10 ply. >>>>> >>>>>strange, I don't know the logic there, i would have guessed a bigger >>>>>difference. please don't be angry but if that is what I would get in amyan >>>>>regularly I would be sure it has a bug or I made a mistake in the experiment. >>>> >>>>It is possible of course, but I don't think so. It just don't seem to buy me a >>>>lot. As far as I know it has never been proven anyway. Plus (based on only a >>>>couple of tests) I simply don't see it here. >>> >>>but search with alpha+1,beta is better if beta is not infinite. I don't think >>>that using pvs changes that much, the results of your tests, if have done well, >>>are counterintuitive. >> >>Yes. I am not claiming it is better of course. It just happened in this >>position. But it is interesting to figure out how this can happen in *any* >>position. One thing I can think of is that you lose bound info. > >ok but it shouldn't matter much when using a conservative window, except in big >fail low or fail high. > >>>>>>Convince me with data :-) Should be no big deal to do this little test for >>>>>your engine? Difference? Other position is also fine. >>>>> >>>>>I'm not using much pvs inside the three as it does not help me, may be other >>>>>person may do that quicker. >>>>>Be well... >>>> >>>>Do you use straight alphabeta? PVS might be a tiny bit more efficient. At least >>>>that is what everybody says. But the difference is small IMO. >>>hyatt says that for him, the difference between alphabeta with aspiration and >>>pvs is only 10%, for me is near zero. >> >>Well that is a number that does not seem unrealistic to me. Only slightly >>better. On the other hand not having a rootwindow spairs you the trouble of >>having fail-low's. After which you have no hashmove, *bang*. What's better >>overall remains a question for me. > >"After wich you have no hashmove" >you mean the hash moves of positions inside that tree? they don't have to be >deleted by the fail low... altough yes possibly if the thing is failing low may >be some of them were not so good but they are there. > >best wishes... >me. Suppose you search with <-50, 50> and get a fail low on the root. Now you have to re-search. And what is the hashtable giving you? A move with score < -50. It might as well be < -1000 and not a good move to play first. If you do nothing about it the search might totally crash. That's why Crafty uses "internal iterative deepening". Best regards, Bas.
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.