Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: cutoff rates

Author: Robert Hyatt

Date: 10:17:45 10/11/04

Go up one level in this thread


On October 11, 2004 at 08:53:30, Rick Bischoff wrote:

>Hello,
>
>I have been trying to get my move ordering right, and am running into the
>following anomalies-- was wondering if your engine has the same rates;
>basically, if the score is above alpha (even if it is above beta), I increment a
>counter for the type of move.
>
>I currently count hash moves, winning captures, =captures, losing captures,
>killers, history and everything else.
>
>I ran the full WAC test at a fixed depth of 6 and the following happens;
>
>for every position:
>
>hash cutoffs are around 15-40%
>winning captures are around 40-73%
>even captures are ALWAYS 0% (but I still have some, just not enough to be >= 1%)
>losing captures are around 10%
>killers are around 1%
>history are around 0%
>and other are 3%
>
>I *must* have a bug-- it makes sense to me that hash cutoffs should be much
>bigger than winning captures shoud be much bigger than even captures should be
>bigger than losing captures, etc.
>
>One scenario that I could think of was that the search kept changing its mind
>during the next iteration, but surely this shouldn't happen enough to account
>for the large difference.
>
>From the opening position with a fixed depth of 8 with material only eval:
>
>1,197,470 nodes
>242162 cutoffs
>9% hash (23260)
>34% winning captures (82780)
>0% even captures (61)
>22% losing captures (54077)
>31% killers (75561)
>0% history (71)
>2% other (6352)
>
>with full eval:
>
>1,492,098 nodes
>295234 cutoffs
>8% hash (25776)
>37% winning captures (111206)
>0% even captures (109)
>24% losing captures (70898)
>25% killers (76224)
>0% history (290)
>3% other (10731)


I don't see anything strange in your numbers.  Winning captures will always be
at the top, since a full-width search drops pieces everywhere and the best
refutation is to simply take the piece.  losing captures seems odd but I haven't
given it much thought.  Killers looks reasonable.  Hash will always be low as
there are so few hash hits where the entry has a best move...




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.