Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: move_generation + hash

Author: Robert Hyatt

Date: 15:00:01 05/30/00

Go up one level in this thread


On May 30, 2000 at 14:54:28, Tom Kerrigan wrote:

>On May 30, 2000 at 14:36:24, Gian-Carlo Pascutto wrote:
>
>>On May 30, 2000 at 12:50:33, Tom Kerrigan wrote:
>>
>>>If you're not hashing during quiescence, how are you getting hit rates around
>>>40%? That's more than my percentage of non-quiescent nodes, period...
>>
>>I remember seeing my hitrate DROP by adding quiescence hashing...a lot
>>more stuff goes in, the tables are probed a lot more, but q-positions
>>do not seem to be searched as often as normal ones are, hence less hits.
>>
>>--
>>GCP
>
>I just ran a test and over 90% of my nodes are qsearch() nodes. So it's not
>possible for me to have a hash table hit rate over 10%. So a 40% hit rate is
>pretty amazing.
>



Your q-search is out of control.  We had this discussion here a few years ago
with myself, Bruce and Ed doing the discussing.  If your q-search is > 50% of
the total nodes searched, it is probably too big...  There may be other things
in my code that improve hash hits too, like internal iterative deepening, and
so forth.

>I assume that when Bob says he expects a 25% hit rate, he means 25% of the times
>that the hash table is probed, and not 25% of the nodes. Which means that more
>work needs to be done to estimate the speedup from the optimization that started
>this thread, i.e., the equation becomes:
>
>speedup = (% hit rate) * (% hash probes) * (% hash move cutoffs) * (% of time
>spent doing move generation)
>
>Throwing in some extremely optimal numbers, the result is:
>
>0.5 * 0.2 * 0.5 * 0.2 = 1%
>
>So I think the best you can hope for is 1%. Seems like too much work for too
>little, to me.
>
>-Tom



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.