Author: blass uri
Date: 10:29:22 07/30/98
Go up one level in this thread
On July 30, 1998 at 11:58:00, Steven J. Edwards wrote: >On July 29, 1998 at 20:25:54, blass uri wrote: > >>How much time did you use to compute ply 7? > >25322 seconds -> about 7 hours -> 7.92 microseconds per move -> 126 KHz; run on >a Motorola PowerPC 604e at 200 MHz using offset (not bitboard) generation. Full >legality checking is included as is incremental tracking of the material >balance, a 64 bit hash code, and other items. > >>Did you use hash tables to save time? > >It is not clear how transposition tables would be of much help. I think you can store all the positions after 5 plies and how many times you can reach to every position in hash tables. If we look in the game 1.a3 a6 2.b3 b6 3.c3 the position after 3.c3 can be reached by 12 games so the number of positions is significantly lower than the number of games. It is possible to compute for every position after 5 plies how many games of 3 plies there are beginning from the position and multiply the result by the number of times you can reach the position. I think if you use this method you can save at least 80% of the time (I believe the number of positions after 5 plies is less than 20% of the number of games). In order to store a position it is enough to store the difference between it and the initial position and it is not many bits Uri > >>can you compute the number of different positions at every ply? > >The reported numbers are only for trajectories (move pathways). A rather large >amount of storage would be needed for an accurate and precise enumeration of >distinct positions. > >I am running ply=8 now. Some partial results are available; the seven ply >subtree grown from Nc3 has 3,926,684,340 pathways and the subtree from Na3 has >3,193,522,577 pathways. > >Estimating the 10 ply count to be around 200 trillion, it would take my machine >about 50 years to get the corresponding result. But if we could get a few >thousand networked machines running the OCD toolkit or similar software with >pathway enumeration, we could have the results much sooner. > >-- Steven (sje@mv.mv.com)
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.