quiescence search

[flok]

Hi,

What should I expect from QS? I think I read somewhere that 80% of a search is in QS. And what about the depth?
In the initial position for a depth of 9, my program calculates 83% in qs (346096 nodes for qs, 70662 for non-qs) and with a maximum depth (9 + qs depth) of 30 (and 22502 non-qs tt hits and 5428 qs-tt-hits, 256MB tt, entry size: 15 bytes, 8 entries per bucket).

[jdart]

That doesn't look too unusual. The exact numbers are going to vary depending on how much you prune/extend and how good your move ordering is.

--Jon

[Luis Babboni]

My first attempt with QS in Soberango down its usual non QS depth from 6 or 7 to 4 or 5 when I spected directly no signal of the few more nodes (at least I thought will be few) need to be visited.
I still think was just cause a buggy implement of it, you say it could be just cause it could be like this with QS?

In any case my implementation was bugy but it is an interesting thing to know.

[Luis Babboni]

That doesn't look too unusual. The exact numbers are going to vary depending on how much you prune/extend and how good your move ordering is.

--Jon

Thanks! I wrote my question before see yours.

[cdani]

Current Andscacs go depth 25 from startpos. Number of times it enters alpha_beta or quiesce:

Code: Select all

depth	alpha_beta	quiesce	% of quiesce
1	0	23	100,0%
2	40	115	74,2%
3	62	300	82,9%
4	111	395	78,1%
5	246	459	65,1%
6	1188	3401	74,1%
7	952	1432	60,1%
8	1068	2272	68,0%
9	3358	5252	61,0%
10	6817	12909	65,4%
11	19011	37764	66,5%
12	29238	50274	63,2%
13	41241	70769	63,2%
14	170346	313987	64,8%
15	47370	79711	62,7%
16	367541	661094	64,3%
17	128405	193801	60,1%
18	258702	431044	62,5%
19	824571	1452299	63,8%
20	1125471	1962028	63,5%
21	510736	797560	61,0%
22	1874693	3427381	64,6%
23	2019958	3553644	63,8%
24	4259991	7529464	63,9%
25	11154820	20078191	64,3%
Total	22845936	40665569	64,0%