Author: Robert Hyatt
Date: 18:13:18 08/21/01
Go up one level in this thread
On August 20, 2001 at 16:55:57, José Carlos wrote: >On August 20, 2001 at 14:30:15, Robert Hyatt wrote: > >>On August 20, 2001 at 11:56:35, José Carlos wrote: >> >>>On August 20, 2001 at 09:43:22, Robert Hyatt wrote: >>> >>>>On August 20, 2001 at 05:22:36, Gian-Carlo Pascutto wrote: >>>> >>>>>On August 20, 2001 at 04:41:41, José Carlos wrote: >>>>> >>>>>> Then I tried not doing it in the PV, and the strange things went out. The tree >>>>>>got smaller and the search deeper. >>>>>> I don't know if this is the correct way of doing it (never do a null-move in >>>>>>the PV), cause I like to try myself before looking other's sources to see how >>>>>>they do it. But looks reasonable. In the PV, we don't expect to fail high, so >>>>>>why trying a null move? >>>>> >>>>>I don't do this, and crafty doesn't seem to be doing this either. >>>>> >>>>>However, your explanation makes perfect sense and I will try it. >>>>> >>>>>-- >>>>>GCP >>>> >>>> >>>>I used to do it but I found that trying nulls _everywhere_ actually made my >>>>tree smaller. I do it this way now because it was more efficient when I >>>>tested it.. Not by a lot, but better. >>> >>> Mmm, some time ago I asked about my not enough deep search, and some of you >>>pointed me to the move ordering. I found that null moves in the PV caused the PV >>>to go crazy sometimes. So I stoped doint null moves in the PV (according to my >>>previous post reasoning) and I did the trick. My fail high on the first >>>move/fail highs rate moved from 86-92% to 90-96%. And my search went to 'normal >>>depths'. Now I doubt again... Is there something I have to take into account >>>when doing null moves in the PV to avoid getting strange PV's? I know, the >>>question is too vague but, maybe someone has an idea. >>> So, everyone does null moves in the PV? >>> >>> José C. >> >> >> >>I don't see how nulls on the PV can cause "strange PVs". If you have a position >>that drives your search nuts, post it and I will run it with Crafty with null >>on and off to see how it behaves. >> >>Null-move definitely causes quirks here and there, particularly false fail highs >>if you use PVS and you have to handle them correctly or blow up. > > I'll recompile the program as it was before and try to find an example, but it >was very difficult to catch it, because the strange things happened in games, >not in test positions. It must have something to do with hasing too. Anyway, >I'll try to find and example and post it here. > > José C. Yes... beware of hashing as well. you have the same position before and after a null-move. If you use one hash table, and you don't consider side to move correctly, then null-move will show you the error of that thinking. Quickly. :)
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.