Computer Chess Club Archives


Search

Terms

Messages

Subject: Bugreport (in hashtables?)

Author: Eelco de Groot

Date: 03:12:22 09/12/03

Go up one level in this thread


Hello Ed (or Lex),

I don't know if it has anything to do with he Singular Extensions or the way
they are represented in the hash tables, but I noticed already before, with the
prerelease version, that it is as if the contents of the hash table don't all
get cleared sometimes, when you load a new personality. I only saw a clear sign
of it in the position that Richard Harrison posted a while ago here, with the
Bishop sacrifice on g6.

[D]rq3r2/3bp1kp/5pp1/2pPP3/ppP2P1Q/n2B2R1/P3N1PP/3R2K1 w - - 0 28

If I have tried a personality that finds the winning move I always stop the game
and enter "new game" in ChessPartner but I don't reload the engine every time
trying a new style. But now with a version that has SEX_HIST_DEPTH_7 and that
I'm trying to tune a bit, I can get results that I cannot reproduce. For
instance I got

Personality hist7_4

00:00:00.3	2,62	1	252	e6
00:00:00.3	2,87	2	524	e6 Be8
00:00:00.5	2,78	3	3642	e6 Be8 Re1
00:00:00.7	2,28	4	11264	e6 Be8 f5 Rh8 fxg6 Bxg6
00:00:00.8	2,30	4	31316	Be4 Nxc4 d6 exd6 Bxa8
00:00:00.9	2,40	4	42406	f5 Qxe5 fxg6 h5 Rf1
00:00:01.1	2,56	5	56862	f5 Qxe5 fxg6 h5 Re1
00:00:01.7	2,31	6	122498	f5 Qxe5 fxg6 h5 Nf4 Qd4+ Kh1
00:00:01.8	2,34	6	143989	e6 Be8 Bxg6 Bxg6 f5 Nxc4 fxg6
00:00:03.9	2,36	7	498259	e6 Be8 Bxg6 Bxg6 f5 Rh8 Nf4 Qe8
00:00:07.1	2,36	8	1081969	e6 Be8 Bxg6 Bxg6 f5 Rh8 Nf4 Qe8
00:00:16.1	1,96	9	2784261	e6 Be8 Bxg6 Bxg6 d6 exd6 f5 Nxc4
00:00:34.6	2,06	9	6323403	Rf1 fxe5 f5 Qd6 fxg6 Rxf1+
00:01:10.1	2,06	9	13130016	Bxg6
00:01:11.1	2,06	9	6323403	Rf1 fxe5 f5 Qd6 fxg6 Rxf1+
00:02:18.1	2,15	10	26219508	Rf1 Be8 exf6+ Rxf6 Rh3
00:02:28.2	2,15	10	28146090	Bxg6
00:02:41.2	2,15	10	26219508	Rf1 Be8 exf6+ Rxf6 Rh3
00:04:35.8	1,89	11	53918205	Rf1 fxe5 Qxe7+ Rf7 Qxe5+
00:04:37.2	1,89	11	54222326	Bxg6
00:08:04.2	1,89	11	96319466	Bxg6
00:20:35.6	6,07	11	252779863	Bxg6 hxg6 Rxg6+
00:21:49.4	6,07	12	268376077	Bxg6 hxg6 Rxg6+ Kxg6 Rd3 Rh8 f5+ Kf7

which is strange in the sense that Bxg6 already appears at ply nine but gets a
fail high error. Reloading the engine, and testing another postion in between to
flush the hash, I try again the same version (All 28 Mb hash tables):

00:00:00.3	2,62	1	276	e6
00:00:00.3	2,87	2	544	e6 Be8
00:00:00.5	2,79	3	3602	e6 Be8 Rg4 Rg8 Be4 Kh8 d6 f5
00:00:00.7	2,28	4	10988	e6 Be8 f5 Rh8 fxg6 hxg6 Qe4
00:00:00.9	2,40	4	34446	f5 Qxe5 fxg6 h5 Kh1
00:00:01.1	2,56	5	47643	f5 Qxe5 fxg6 h5 Re1
00:00:01.6	2,31	6	111606	f5 Qxe5 fxg6 h5 Nf4 Qd4+ Kh1 Bb5
00:00:01.7	2,34	6	124126	e6 Be8 Bxg6 Bxg6 f5 Nxc4 fxg6 hxg6
00:00:03.2	1,97	7	374057	e6 Be8 Rg4 Kh8 Bxg6 Bxg6 Rxg6 Nxc4
00:00:03.5	2,36	7	410794	f5 Qxe5 fxg6 h5 Nf4 Bg4 Ne6+ Bxe6
00:00:06.9	1,99	8	1056261	f5 Qxe5 fxg6 h5 h3 b3 Nf4
00:00:11.9	2,01	8	1997886	Kh1 Be8 Bxg6 Bxg6 f5 Qxe5 fxg6
00:00:25.5	2,03	8	4635756	d6 exd6 f5 dxe5 fxg6 h6 Be4 Ra7
00:00:31.7	1,87	9	5853238	d6 exd6 Be4 Nxc4 Bd5 Bb5 exf6+ Rxf6
00:00:38.0	1,87	9	6994322	e6
00:00:38.9	2,36	9	7158674	e6 Be8 Bxg6 Bxg6 f5 Qe8 Nf4
00:01:13.3	2,87	10	13862333	e6 Be8 Bxg6 Bxg6 f5 Qe5 fxg6 h5
00:01:50.1	2,87	11	21952070	e6 Be8 Bxg6 Bxg6 f5 Qe5 fxg6 h5
00:04:24.4	2,87	11	53000944	Bxg6
00:04:27.7	2,87	11	21952070	e6 Be8 Bxg6 Bxg6 f5 Qe5 fxg6 h5
00:06:19.5	2,57	12	76173649	e6 Be8 Bxg6 Bxg6 f5 Qe5 fxg6 h5
00:06:44.2	2,57	12	81100833	Bxg6
00:15:17.3	4,17	12	184690543	Bxg6 hxg6 Rxg6+ Kxg6 Rd3 Rh8 f5+ Kf7
00:25:42.2	4,17	13	315204856	Bxg6 hxg6 Rxg6+ Kxg6 Rd3 Qc8

Notice that Bxg6 comes later (and that there is a strange line at ply eight
where suddenly the White King makes a quiet move but that is still possible)
Also I can get a result with another version that I cannot reproduce:

 hist7_5 kan niet kloppen

00:00:00.2	2,06	1	114	exf6+ Rxf6
00:00:00.2	2,55	1	295	e6
00:00:00.3	2,79	2	543	e6 Be8
00:00:00.4	2,72	3	3677	e6 Be8 Rg4 Rg8 Bxg6 Bxg6 f5 Rh8
00:00:00.7	2,23	4	10947	e6 Be8 f5 Rh8 fxg6 hxg6 Qf4
00:00:00.9	2,35	4	42406	f5 Qxe5 fxg6 h5 Qh3
00:00:01.1	2,51	5	52529	f5 Qxe5 fxg6 h5 Re1
00:00:01.6	2,27	6	114685	f5 Qxe5 fxg6 h5 Nf4 Qd4+ Kh1
00:00:03.1	2,31	7	362566	f5 Qxe5 fxg6 h5 Nf4 Bg4 Ne6+ Bxe6
00:00:03.2	2,31	7	379297	e6
00:00:03.4	2,31	7	362566	f5 Qxe5 fxg6 h5 Nf4 Bg4 Ne6+ Bxe6
00:00:06.0	2,31	8	876309	f5 Qxe5 fxg6 h5 Nf4 Bg4 Ne6+ Bxe6
00:00:11.1	2,31	8	1894440	Bxg6
00:00:15.2	2,31	8	2639982	Bxg6
00:00:23.6	5,50	8	4307478	Bxg6 hxg6 Rxg6+ Kxg6 Rd3 Qc8 Rg3+ Bg4
00:00:25.8	5,50	9	4709094	Bxg6 hxg6 Rxg6+ Kxg6 Rd3 Qc8 Rg3+ Bg4
00:03:33.4	4,85	10	43985226	Bxg6 hxg6 Rxg6+ Kxg6 Rd3 Rh8 f5+ Kf7
00:04:21.1	4,85	11	54423645	Bxg6 hxg6 Rxg6+ Kxg6 Rd3 Rh8 f5+ Kf7

It is not confined to the settings of the history depth, with older Betazoid
versions where I did not use that option I also saw this (in this position). Has
this been reported already before Ed? I hope you can reproduce some of the
behaviour at least and try to repair it then?

Groeten, Eelco



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.