Author: Bernd Nürnberger
Date: 02:32:10 02/04/06
Go up one level in this thread
On February 04, 2006 at 04:42:14, Vasik Rajlich wrote: >On February 04, 2006 at 04:18:58, Bernd Nürnberger wrote: > >>On February 04, 2006 at 03:56:09, Vasik Rajlich wrote: >> >>>On February 04, 2006 at 03:21:12, Bernd Nürnberger wrote: >>> >>>>Hello, >>>> >>>>I have a question concerning the rate of beta cuts. My program prunes the tree >>>>by doing a beta cut at about 5-10% of all nodes. This seems rather low to me. >>>>Thus most moves that were generated are actually made. Nethertheless I am >>>>getting rather low node counts and a mediocre first cut rate: for WAC (pos >>>>1-300) at depth 8, my program searches about 69 M nodes with a first cut rate of >>>>88%. >>>> >>>>Now to the question: is a cut rate of 5..10% common?? I often heard that most >>>>moves generated are nether done because of beta cuts. That's not the case in my >>>>program. >>>>For WAC (depth 8) 88 M nodes are generated and 69 M nodes are done ! >>>> >>>>Configuration: >>>>- Hash Table (1 M entries, deep/new two-level replacement) >>>>- Razoring/(Extended) Futility Pruning >>>>- Null Move (rather aggressive) >>>>- Move Ordering: >>>> x hash move (IID) >>>> x winning + equals caps (by SEE, only x-rays capturer) >>>> x killer 1 / killer 2 >>>> x non-caps (history heuristics) >>>> x losing caps >>>>- PVS / aspiration search >>>>- QS (prunes rather aggresively using SEE) >>>>- plain vanilla eval (mat + pc-sq + boni/penality for castlings) >>>> >>>>Any comments appreciated :-) >>>> >>>>Bernd >>> >>>In a classical search without razoring-type stuff you'll fail high at well over >>>90% of nodes, and will need much fewer than 69M nodes to do 8 iterations. >>> >>>It's possible you have the alpha-beta logic mixed up. You might want to step >>>through a sample variation move by move. For this the "Chant" tool (advertised >>>here from the search engine page) is quite handy, it's what I use. >>> >>>Vas >> >>Hello Vas, >> >>I don't think that 69 M nodes for 300 positions to depth 8 is way too much. >>After disabling null move and razoring/futility pruning, I am getting >>98% average first cuts at WAC positions 1-10 (depth 8) while getting >>24% beta cuts on average. With null move the it's 91% / 4% and with >>null move and razoring/fut. it's 90% / 7%. >> >>So the first cut rate is good with null moves disabled, but the total beta cut >>rate is somewhat low. Did you mean, that the total beta cut rate should be >>over 90% or just the first cuts?? >> >>Bernd > >Aha, now I understand what you mean. > >69M for all 300 positions sounds ok. > >Your first cut rates sound right. They will drop further if you are more >selective at the leaves (ie. more razoring). > >Re. what you call beta cut rate, I have never measured this. Offhand, I'd guess >that it should be way over 90%. The vast majority of program nodes where any >moves are generated are depth == 1 fail-high nodes where the first move causes a >cut. > >Vas I see. Probably I have some really nasty bug. Even in Q-search I only get a beta cut at about 1/3 of the nodes. Curiously most PVs looks reasonable and I get about 270 WAC positions solved at 5s. So alpha-beta is not all mixed up. Thanks for your comments... Bernd
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.