Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Cutoff Statistics

Author: Robert Hyatt

Date: 14:22:57 12/18/97

Go up one level in this thread


On December 18, 1997 at 15:13:18, Willie Wood wrote:

>
>On December 18, 1997 at 07:02:57, Roland Pfister wrote:
>
>>In  the development/debug version of my program Patzer I have built in
>>some statistics about cutoffs. I ran it on the Bratko-Kopec test
>>with search depth 8. Here some results of the cutoff efficiency of move
>>types:
>>
>>BK  null hash capt kil1 kil2 misc
>> 2:  67   60   46   31   13   0
>> 3:  79   61   35   45   12   0
>> 4:  87   64   41   38   11   1
>> 5:  76   62   44   25   10   0
>> 6:  74   69   63   30    6   0
>> 7:  78   61   39   36   13   0
>> 8:  72   55   48   45   13   0
>> 9:  79   63   40   24    9   0
>>10:  80   60   45   34   18   0
>>11:  75   68   45   28    9   0
>>12:  72   66   48   40   18   0
>>
>>that means e.g. in Bratko-Kopec #12  72% of all null moves produced
>>a cutoff, 66% of all hashmoves produced a cutoff etc.
>>
>>the statistics was done in the normal search not quiescense and my
>>move ordering/generation is from left to right ( nullmove first,
>>hash move, good captures, first killer, second killer, up to 5
>>history moves, rest ).
>>
>>Has anyone else statistics about cutoffs or will produce them?
>
>
>That's pretty interesting.  Can you tell me what's the difference
>bewteen killer 1 and 2, and do you allow captures in the killers?
>
>WW

Generally you wouldn't, because you try captures *before* the killer
moves.  If there were no good captures before killers, it is unlikely
a killer would be a capture...  Also, you'd wind up with wasted killer
entries because every capture that causes a cutoff would be entered in
the killer list, but would be tried *before* the killer move next time
anyway...



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.