Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: About qsearch...

Author: Uri Blass

Date: 16:17:25 12/29/01

Go up one level in this thread


On December 29, 2001 at 15:05:22, Sune Fischer wrote:

>On December 29, 2001 at 13:59:25, Ed Schröder wrote:
>
>>On December 28, 2001 at 08:14:02, Uri Blass wrote:
>>
>>>if you extend the qsearch too much you may get
>>>often irrelevant positions that you analyze.
>>
>>So?
>>
>>What you get back is an accurate score of the leaf, the only thing that counts.
>>
>>Limiting QS to 6,7... plies is a bad thing because it will return an unreliable
>>score too many times.
>>
>>Ed
>
>It is not as bad as it sounds really, not if you sort them by MVV-LVA.
>If I limit QS to e.g. 3 plies, then I can afford 1 ply extra for the whole tree.
>OK, I don't use SEE now, maybe that will speed up QS enough so that is
>affordable, but for right now my program is the overall strongest by using QS of
>only 3 plies deep (I tested against itself). And actually it almost never begins
>a losing capture sequence.
>If I do a full QS then >80% of the time is spent in there, that is a complete
>overkill IMO, since chess is not 80% capture moves.
>It is very much a trade off, and dependent on the rest of the program.
>
>-S.

My qsearch is not close to include >80% of the nodes and I suspect that you do
something wrong in the qsearch.

Here are the results of my program when searching depth 7 from the initial
position:
limit of qsearch in plies is the first number in every line   and  nodes for
depth 7 is the second number:

Note that it is using static evaluation when it finish the qsearch and it is not
a smart thing to do

I think that limiting the qsearch too much may do chess program slower in
getting the same depth thanks to bad order of moves
0   2868842
1   3611982
2   997968
3   1564826
4   1196247
5   1214999
6   1184129
7   1153555
8   1180282
9   1153618
>=10  1153584

I also tried modified TSCP that use similiar techniques and got the following
0 1329080
1 2107731
2 1495776
3 1341166
4 1106628
5 1090297
6 1142271
7 1173772
8 1156489
9 1174336
10 1174570
11 1174435
12 1174560
>=13 1174526

I believe that you do something wrong if you get 80% of the nodes in the
qsearch.
note that my program and TSCP have no SEE.

Uri



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.