Author: Robert Hyatt
Date: 06:12:59 12/02/02
Go up one level in this thread
On December 02, 2002 at 01:33:00, Anthony Cozzie wrote: >1. e4 b6 2. exd5 exd5 3. Rb1 > = (0.25) Depth: 5 00:00:00.04 21kN >1. e4 b6 2. Qe2 dxe4 3. Bxe4 Nxe4 > = (0.25) Depth: 6 00:00:00.09 42kN >1. e4 b6 2. Qe2 dxe4 3. Bxe4 Nxe4 4. Nxe4 > = (0.25) Depth: 7 00:00:00.23 112kN >1. e4 b6 2. exd5 exd5 3. Qe1 Re8 4. Ne5 Nxe5 > = (0.24) Depth: 8 00:00:01.21 541kN >1. Ne5 Nxe5 2. dxe5 Nd7 3. Qg4 Kh8 4. Rfd1 b6 5. Bg3 > = (0.21) Depth: 9 00:00:09.75 4667kN >1. e4 b6 2. Qe2 Re8 3. Rfe1 Bb7 4. Nd2 a5 5. Bb5 Kh8 > = (0.15) Depth: 10 00:00:24.49 11115kN >1. e4 b6 2. Qe2 Re8 3. Rfe1 Bb7 4. Nd2 a5 5. Bb5 Bb4 6. f3 > = (0.15) Depth: 11 00:00:58.47 27207kN > >Time: 60093ms, Nodes: 27844k, Quiescence Nodes: 13650k [49%] >Table Hits: 4073k [15%], Table Pings: 5799k [21%], NPS: 463k, Null Cuts: 295k > >Recently I have noticed that Zappa sometimes changes its mind for 1 move in a >most annoying way. Its pretty clear that the score is not changing much, but it >still has to work out an entirely new PV. I expect that as I expand on the >rather rudimentary evaluation this problem will only get worse. So A) does >anyone else view this as a problem, and B) what did you guys do to mitigate it? > >anthony There is one published way to handle this. When you search, and get the score for the first root move (say X) then search the remaining root moves with an offset window. IE alpha=X+.02, beta=X+.03 or something similar. Now a move has to be .02 better than "best" before it will fail high and work its way into a real PV move. You can adjust the offset to whatever you want...
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.