Author: A.G.B. Bluemers
Date: 15:06:05 10/14/04
Go up one level in this thread
On October 14, 2004 at 14:03:13, Stuart Cracraft wrote: >On October 14, 2004 at 13:50:41, Pallav Nawani wrote: > >>On October 14, 2004 at 10:28:16, Stuart Cracraft wrote: >> >>>I don't think I'm implementing it properly because my node >>>reductions are not happening. >>> >> >>Count your prunings and see if you are pruning something or not. If the pruning >>threshold is too conservative, you might not be pruning anthing! >> >>Pallav > >It is. Here are the first 10 positions of Win at Chess on a 1 >second search with futility pruning at depth == 1 and depth == 2 >enabled. The final column (otherx) is non-zero showing futility >pruning is happening: > >- 6.00/31.00 0% 0/1 ha=38 1.02 233472 233472/1/229982 9/648/0/3866/78466/986/241 >- 8.50/27.00 0% 0/2 ha=40 2.03 515072 257536/1/254141 >1883/1743/0/10028/96599/998/241 >+ 8.33/24.00 33% 1/3 ha=41 3.06 711680 237227/1/232880 >2004/2388/0/13381/171341/1118/244 >+ 6.50/19.50 50% 2/4 ha=36 3.06 711860 177965/1/232851 >2004/2388/0/13390/171344/1120/244 >+ 5.40/17.00 60% 3/5 ha=33 3.06 712036 142407/1/232829 >2004/2388/0/13398/171357/1123/244 >+ 6.50/17.83 66% 4/6 ha=39 4.07 1011044 168507/1/248489 >2045/2461/170/26468/187772/2849/342 >+ 6.71/18.00 71% 5/7 ha=40 5.07 1271140 181591/1/250605 >2166/2895/340/30082/253136/3501/438 >+ 6.88/18.75 75% 6/8 ha=42 6.08 1526116 190764/1/251184 >2195/3415/510/35943/327868/4640/2789 >+ 6.22/18.56 77% 7/9 ha=43 6.08 1527395 169711/1/251079 >2195/3415/680/36020/327931/4662/2789 >+ 6.40/18.70 80% 8/10 ha=44 7.09 1747555 174756/1/246369 >2497/4198/850/41204/410785/5048/2908 > >Compare the above with futility pruning disabled: > >+ 3.00/63.00 100% 1/1 ha=21 0.32 62942 62942/0/197655 0/107/0/3778/12505/705/0 >- 7.00/43.00 50% 1/2 ha=32 1.33 344542 172271/1/259467 >1874/1202/0/9940/30638/717/0 >+ 7.33/34.67 66% 2/3 ha=35 2.33 538078 179359/1/230951 >1988/1843/0/13194/103614/840/0 >+ 5.75/27.50 75% 3/4 ha=32 2.33 538258 134564/1/230915 >1988/1843/0/13203/103617/842/0 >+ 4.80/23.40 80% 4/5 ha=29 2.33 538434 107687/0/230887 >1988/1843/0/13211/103630/845/0 >+ 6.00/23.17 83% 5/6 ha=36 3.34 837442 139574/1/250548 >2029/1911/189/26331/119352/2667/0 >+ 6.29/22.57 85% 6/7 ha=38 4.34 1096514 156645/1/252382 >2150/2345/378/29945/183543/3321/0 >+ 6.50/24.12 87% 7/8 ha=38 5.36 1349444 168680/1/251569 >2230/2830/567/36594/245744/4669/0 >+ 5.89/23.33 88% 8/9 ha=40 5.37 1350723 150080/1/251448 >2230/2830/756/36671/245807/4691/0 >+ 6.00/22.70 90% 9/10 ha=40 6.38 1511491 151149/1/236991 >2503/3434/945/40470/304424/4981/0 >pawnx/recapx/mthrx/checkx/qfutilx/onereplyx/otherx > >My only concern is that otherx in the first example is such a >small number. Note that the branching factor in the former >was 2.64 vs. without futility a branching factor of 3.82, >further evidence. > >The code itself is here and encompasses at 1 ply before the >horozion and 2 ply before the horizon. > >Still I think something is amiss. This code triggers it > > if (FUTIL_R != 0.0) { > if (depth == iply-1) { > if (parentincheck || ml.cap!=0 || alpha < eval(bd,QUIET)+3000 || >incheck(bd)) > extension = 0; > else > extension = -1; > } else if (depth == iply-2) { > if (parentincheck || ml.cap!=0 || alpha < eval(bd,QUIET)+5000 || >incheck(bd)) > extension = 0; > else > extension = -1; > } > } to see if pruning works you should check for number of nodes per depth per position x with and without pruning ,not the total number of nodes after move time.
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.