Author: leonid
Date: 16:06:25 09/12/99
Go up one level in this thread
On September 12, 1999 at 14:58:48, Bert van den Akker wrote: >On September 12, 1999 at 06:43:49, leonid wrote: > >>On September 11, 1999 at 22:09:38, Bas Hamstra wrote: >> >>>HashTables are very important for movesorting. Best moves from previous searches >>>are in the hashtable. That way you can very much improve the probability that >>>you play the best move first, which is immensely important for the alpha-beta >>>scheme. When you search to ply n with hashtables you have to examine less nodes >>>than without, because of that. >>> >>>Estimates of how much less vary from 20% to 50%. In case of 50%, assuming using >>>nullmove, yes that is almost a ply. >>> >>> >>>Regards, >>>Bas Hamstra. >> >>Your message, and the one above, is more that useful. This is just moves >>ordering that still must be improved in my game and, after what I can see, >>hash table help even in this aspect. When I tryed my engin against the best >>one my "branching factor" was miserable. Now I work in improving it. >> >>With my respect, >>Leonid. > > >With my program I have done the following test on position: >2r1r1k1/3p1p1p/p2Qp1p1/q1N2P2/Pn1B1P2/6P1/2P3bP/2KRR3 b - - > >to see what is the influence of >Hash tables, Null move and History tables on the search time. >All times are in seconds. > > > > >D=Queen >T=Rook >L=Bishop >P=Night > >With >Hash table: 524288 entries >Recursive Null move: R=2 >History table for move ordering: On > > > > >Depth Score Time_for_ply Time_PV PV >1 -223 0 0 Tc8-b8 >2 -196 0 0 Tc8-b8 Kc1-d2 e6xf5 Te1xe8+ Tb8xe8 Dd6xd7 >3 -192 2 2 g6xf5 Dd6xd7 >4 -154 4 1 g6xf5 Dd6-e5 f7-f6 De5xf6 >5 -195 13 8 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ c2-c4 Db4xc4+ Ka2-a3 >6 -289 9 1 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 Te1xe8+ >Tc8xe8 c2-c3 >7 -282 25 5 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 >8 -277 115 18 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 > >total time 168 > > >With >Hash table: Off >Recursive Null move: R=2 >History table for move ordering: On > > >Depth Score Time_for_ply Time_PV PV >1 -223 0 0 Tc8-b8 >2 -196 0 0 Tc8-b8 Kc1-d2 e6xf5 Te1xe8+ Tb8xe8 Dd6xd7 >3 -192 3 2 g6xf5 Dd6xd7 >4 -154 3 0 g6xf5 Dd6-e5 f7-f6 De5xf6 >5 -195 18 11 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ c2-c4 Db4xc4+ Ka2-a3 >6 -289 34 6 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 Te1xe8+ >Tc8xe8 c2-c3 >7 -282 79 19 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 >8 -277 324 76 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 > >total time 461 > > > >With >Hash table: 524288 entries >Recursive Null move: Off >History table for move ordering: On > > >Depth Score Time_for_ply Time_PV PV >1 -223 0 0 Tc8-b8 >2 -196 0 0 Tc8-b8 Kc1-d2 e6xf5 Te1xe8+ Tb8xe8 Dd6xd7 >3 -192 2 2 g6xf5 Dd6xd7 >4 -154 4 1 g6xf5 Dd6-e5 f7-f6 De5xf6 >5 -195 16 11 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ c2-c4 Db4xc4+ Ka2-a3 >6 -289 35 4 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 Te1xe8+ >Tc8xe8 c2-c3 >7 -282 145 26 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 >8 -277 808 110 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 > >total time 1010 > > > > >With >Hash table: 524288 entries >Recursive Null move: R=2 >History table for move ordering: Off > > >Depth Score Time_for_ply Time_PV PV >1 -223 0 0 Tc8-b8 >2 -196 0 0 Tc8-b8 Kc1-d2 e6xf5 Te1xe8+ Tb8xe8 Dd6xd7 >3 -192 2 2 g6xf5 Dd6xd7 >4 -154 4 1 g6xf5 Dd6-e5 f7-f6 De5xf6 >5 -195 14 8 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ c2-c4 Db4xc4+ Ka2-a3 >6 -289 10 1 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 Te1xe8+ >Tc8xe8 c2-c3 >7 -282 29 5 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 >8 -277 113 19 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 > >total time 172 > > > > >With >Hash table: 2097152 entries >Recursive Null move: R=2 >History table for move ordering: On > > > >Depth Score Time_for_ply Time_PV PV >1 -223 0 0 Tc8-b8 >2 -196 0 0 Tc8-b8 Kc1-d2 e6xf5 Te1xe8+ Tb8xe8 Dd6xd7 >3 -192 2 1 g6xf5 Dd6xd7 >4 -154 3 0 g6xf5 Dd6-e5 f7-f6 De5xf6 >5 -195 14 8 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ c2-c4 Db4xc4+ Ka2-a3 >6 -289 9 1 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 Te1xe8+ >Tc8xe8 c2-c3 >7 -282 25 6 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 >8 -277 109 18 Pb4-a2+ Kc1-b2 Da5-b4+ Kb2xa2 Lg2-d5+ Dd6xd5 e6xd5 c2-c3 >Db4-c4+ Ka2-a3 d7-d6 Pc5-b7 > >total time 162 > > > >Bert van den Akker Very interesting but, to be sincere, I am a little bit lost in your figures. Reason - my programming, and maybe my game, is very far from be usual. I am able to compare only on the level of number of plys and the time. The one thing I would like only confirme with you: Is this true that hash table can give you in the most time consuming positions (beginning and middle game) advantage up to 100% in speed. If this is true I feel me doubtful if hash table is so useful. After all, make your game so swollen for so little is nothing. This could only help for the game that is very close in speed with everybody else. But if the hash tables are mainly for the moves ordering and this way is very efficent, that could be reason for having it. Until now my move ordering was very poor. This is something that make me loose speed against the best games in around 9 ply. Leonid.
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.