Author: Uri Blass
Date: 04:35:01 03/17/03
Go up one level in this thread
On March 17, 2003 at 06:47:17, Albert Bertilsson wrote: >On March 17, 2003 at 06:43:30, Uri Blass wrote: > >>On March 17, 2003 at 05:47:29, Albert Bertilsson wrote: >> >>>Hi! >>> >>>Since Uri pointed out that my PV was rotten I fixed that to see what happens. >>> >>>I picked out all 105 positions that failed when running for 5 seconds and >>>increased the time to 60 seconds. >>> >>>The results was quite nice, some 40 positions where solved! >> >>still 65 unsolved positions are a lot. >> >>Movei can solve 275 out of 300 in less than a second. >> >>Movei needs something between 0.1 and 0.2 seconds on my A1000 to get 65 unsolved >>positions. >> >>Do you use history tables and killer moves for better order of moves? >>Do you use MVV/LVA or some more logical way to order captures? >> >> >>Uri > >Ok, so I'll go out on the bug hunt! > >Sharper uses MVV/LVA to order captures. > >I've tested killer moves but it only gave 1-2% less nodes, maybe I did it >wrong!? > >I'm currently testing history tables and it seems to give me about 5% less >nodes, maybe I got that one wrong too!? > >/Regards Albert I guess that you did something wrong. I did not have a version without history tables so I cannot say how much history gives me relative to no history. I implemented killers before hash tables(I do not know how much it gave me but it was clearly more than 1-2%). The gain is also not supposed to be linear and is supposed to be biggers at long time control. I suspect that you often have the wrong killer like you had the wrong pv and it may explain your problems with killers move. I guess that there is some problem with using history tables because I expect them to help more than 5%(I do not say it from expereince because I never had a program without history tables). 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.