Author: Eelco de Groot
Date: 09:16:58 05/06/05
Go up one level in this thread
On May 05, 2005 at 23:41:09, Eelco de Groot wrote: >On May 05, 2005 at 21:04:42, Eelco de Groot wrote: > > >> >>I was not entirely happy with what Betazoid T90 plays here in the testposition >>from the Ed schroder-Rebel 12 game: >> >>[D]rnbq1rk1/pp1n1p1p/4p1p1/3pP1NP/1b3P2/1PpB4/P1P3P1/R1BQK2R b KQ - 0 1 >> >>Best choices moves are I think Nc6, or Nxe5 which leads to 0.00 score. >> >>T90 over Ed-Rebel 12, met 40 Mb transposition tables >> >>00:00:01.5 3,14 4 7232 h6 Nxf7 Kxf7 Qg4 g5 fxg5 >>00:00:01.6 0,94 5 23971 h6 hxg6 Nxe5 fxe5 hxg5 g7 Kxg7 >>00:00:01.7 1,86 5 37584 Bc5 Nxh7 Nxe5 Nxf8 Nxd3+ Qxd3 >>00:00:01.8 2,30 5 45422 Qb6 hxg6 hxg6 Qe2 Nc6 Be3 >>00:00:01.9 2,34 5 52523 Nc5 hxg6 hxg6 Rh7 Nxd3+ Qxd3 Qxg5 Rg7+ >>00:00:01.9 2,98 5 60858 Be7 hxg6 hxg6 Nf3 Qb6 a4 >>00:00:02.4 3,25 6 103759 Be7 hxg6 hxg6 Nf3 Qb6 a3 Nc6 >>00:00:03.6 2,48 7 215877 Be7 Nxh7 Bh4+ Kf1 Kxh7 hxg6+ Kg7 gxf7 Nxe5 Qh5 Nxd3 >>Rxh4 Nxc1 >>00:00:07.7 2,48 8 694891 Be7 Nxh7 Bh4+ Kf1 Kxh7 hxg6+ Kg7 gxf7 Nxe5 Qh5 Nxd3 >>Rxh4 Nxc1 >>00:00:42.5 0,65 9 5143236 Be7 Nxh7 Na6 hxg6 Bh4+ Kf1 fxg6 Nxf8 >>00:00:50.5 0,86 9 6144608 Bc5 Nxh7 Re8 hxg6 f6 Qh5 >>00:02:09.7 1,03 9 16190937 Qc7 hxg6 fxg6 Nxe6 >>00:05:20.0 0,20 10 42047386 Qc7 Nxh7 Nxe5 >>00:05:42.6 0,27 10 44901986 Bc5 Nxh7 Qe8 Nxf8 Nxf8 hxg6 >>00:06:13.4 0,64 10 49010035 Be7 Nxh7 Bh4+ Kf1 Kxh7 Qg4 Kg7 Rxh4 Rh8 >>00:08:41.7 0,69 10 67141584 Nf6 exf6 Qxf6 hxg6 hxg6 Be3 >>00:09:57.5 -1,12 11 77186856 Nf6 exf6 Qxf6 Nxh7 Qh8 hxg6 e5 gxf7+ Rxf7 Bg6 Rd7 >>00:12:49.6 -0,49 11 100198126 Be7 hxg6 fxg6 Bxg6 >>00:19:37.0 0,00 11 157734728 Qc7 hxg6 fxg6 Rxh7 Qc5 >>00:20:19.3 0,30 11 163233980 Bc5 Nxh7 Na6 hxg6 fxg6 Bxg6 >>00:32:37.4 0,28 12 261283350 Bc5 Nxh7 Nc6 hxg6 fxg6 Bxg6 > >I thought that Betazoid T8 would do better in this particular position, but I >cannot reproduce the earlier results! Nc6 is chosen, but only after 34 minutes, >later Nxe5 comes back. In an earlier result, Nc6 was chosen much sooner. Why >this could have happened, only thing I can think of is maybe not all hashtable >results were gone. I have no learning switches so there should be no position >learning at all. There is no Clear hashtables Always switch in Betazoid, so that >could be a partial cause. Maybe I forgot to set 'Clear hash' under Advanced >properties that time, when loading a new personality. > >What I said then in my post before above about the effect of different hashtable >size also does not hold, as that referred to testing this position with 40 Mb >HT. But the 28 MB results are not reproducible. > >Conclusion then that maybe Betazoid T90 will not differ much from Betazoid T8 >and could be worth further testing. I would rather have some version that does >not play Bc5, but that may not be so easy.. > >Betazoid T8 with 28 Mb HT not like first results > >00:00:01.6 0,94 5 24372 h6 hxg6 Nxe5 fxe5 hxg5 g7 Kxg7 >00:00:01.8 1,86 5 40105 Bc5 Nxh7 Nxe5 Nxf8 Nxd3+ Qxd3 >00:00:01.8 2,30 5 47966 Qb6 hxg6 hxg6 Qe2 Nc6 Be3 >00:00:01.9 2,34 5 55120 Nc5 hxg6 hxg6 Rh7 Nxd3+ Qxd3 Qxg5 Rg7+ >00:00:02.0 2,44 5 63221 Nc6 Nxh7 Ncxe5 Nxf8 Nxd3+ Qxd3 >00:00:02.2 2,98 5 70733 Be7 hxg6 hxg6 Nf3 Qb6 a4 >00:00:02.6 3,25 6 93710 Be7 hxg6 hxg6 Nf3 Qb6 a3 Nc6 >00:00:03.7 2,48 7 201797 Be7 Nxh7 Bh4+ Kf1 Kxh7 hxg6+ Kg7 gxf7 Nxe5 Qh5 Nxd3 >Rxh4 Nxc1 >00:00:08.9 2,07 8 828009 Be7 Nxh7 Bh4+ Kf1 Kxh7 hxg6+ Kg7 gxf7 Nxe5 Qh5 Nxd3 >Rxh4 Qxh4 Qxh4 Kxf7 >00:00:42.5 0,73 9 5074778 Be7 Nxh7 Nc6 Nxf8 Nxf8 g3 Nb4 >00:01:27.0 0,80 9 10818820 Qb6 Nxh7 Qd4 hxg6 fxg6 Nxf8 Bc5 Qf3 >00:01:46.5 1,03 9 13489670 Qc7 hxg6 fxg6 Nxe6 >00:06:03.5 0,23 10 48807141 Qc7 Nxh7 Nxe5 fxe5 Qxe5+ Kf1 >00:06:22.6 0,60 10 51342908 Qb6 Nxh7 Qd4 hxg6 fxg6 Nxf8 Bc5 Qf3 Nxf8 Qg3 >00:06:40.2 0,71 10 53604700 Be7 Nxh7 Nc6 Nxf8 Nxf8 g3 Nb4 >00:09:49.9 0,83 10 77978287 Nf6 exf6 Qxf6 hxg6 hxg6 Be3 >00:11:39.8 -1,12 11 93383561 Nf6 exf6 Qxf6 Nxh7 Qh8 hxg6 e5 Rf1 >00:20:04.4 -0,23 11 165036661 Be7 hxg6 fxg6 Bxg6 >00:25:50.5 0,00 11 212952428 Qc7 hxg6 fxg6 Rxh7 Qc5 >00:32:38.9 0,23 11 269069774 Nxe5 fxe5 Qc7 Nxh7 Qxe5+ Kf1 >00:34:56.7 0,30 11 288216946 Nc6 hxg6 fxg6 Nxh7 Qe7 Nxf8 Nxf8 >00:47:15.4 0,03 12 386966333 Nc6 hxg6 fxg6 Nxh7 Ne7 >00:51:26.5 0,23 12 424254610 Nxe5 fxe5 Qc7 Nxh7 Qxe5+ Kf1 Okay, it gets more complicated. I ran this postion again but now with even less HT, 13 Mb. Now I get two a bit optimistic scores for Nc6 and Nxe5 respectively, at ten ply. This somehow seems to stabilize the search a bit, around the point where hashtables are full. Eventually Nc6 or Nxe5 is chosen. In a way this is a pity because in this particular position then with bigger hashtables the results are just worse in finding Nc6 or Nxe5. And you would like to use bigger hashtables if you can! In games between two programs sofar I always used 13 Mb, otherwise the harddisk starts swapping. Even if this presumed effect of hashtable size happens in very few positions, it can still cause very different results. All this assuming I got the hashtables cleared this time, I still cannot reproduce the earliest Betazoid T8 lines in this position (certainly not with 28 Mb hashtables). They are not very different, but just a little. Nogmaals deze stelling, Pro Deo 1.1 Betazoid T8 met 13 Mb HT 00:00:01.1 3,14 2 975 h6 a3 00:00:01.2 3,35 3 1839 h6 Nf3 gxh5 Rxh5 00:00:01.2 3,14 4 7232 h6 Nxf7 Kxf7 Qg4 g5 fxg5 00:00:01.3 0,94 5 24372 h6 hxg6 Nxe5 fxe5 hxg5 g7 Kxg7 00:00:01.5 1,86 5 40105 Bc5 Nxh7 Nxe5 Nxf8 Nxd3+ Qxd3 00:00:01.5 2,30 5 47966 Qb6 hxg6 hxg6 Qe2 Nc6 Be3 00:00:01.6 2,34 5 55120 Nc5 hxg6 hxg6 Rh7 Nxd3+ Qxd3 Qxg5 Rg7+ 00:00:01.7 2,44 5 63221 Nc6 Nxh7 Ncxe5 Nxf8 Nxd3+ Qxd3 00:00:01.8 2,98 5 70733 Be7 hxg6 hxg6 Nf3 Qb6 a4 00:00:02.2 3,25 6 93710 Be7 hxg6 hxg6 Nf3 Qb6 a3 Nc6 00:00:03.3 2,48 7 201797 Be7 Nxh7 Bh4+ Kf1 Kxh7 hxg6+ Kg7 gxf7 Nxe5 Qh5 Nxd3 Rxh4 Nxc1 00:00:08.5 2,07 8 830948 Be7 Nxh7 Bh4+ Kf1 Kxh7 hxg6+ Kg7 gxf7 Nxe5 Qh5 Nxd3 Rxh4 Qxh4 Qxh4 Kxf7 00:00:41.4 0,73 9 5054804 Be7 Nxh7 Nc6 Nxf8 Nxf8 g3 Nb4 00:01:23.3 0,80 9 10547236 Qb6 Nxh7 Qd4 hxg6 00:01:42.6 1,03 9 13257609 Qc7 hxg6 fxg6 Nxe6 Qc6 00:05:39.2 0,33 10 46287951 Qc7 Nxh7 Nc6 Nxf8 Nxf8 Qg4 00:06:12.3 0,60 10 50747032 Qb6 hxg6 fxg6 Nxh7 Qd4 Nxf8 Bc5 Qf3 Nxf8 Qg3 00:06:44.4 0,65 10 55096524 Be7 Nxh7 Na6 Nxf8 Nxf8 hxg6 Bh4+ 00:07:21.3 0,66 10 60158369 Nc6 Nxh7 Re8 hxg6 f6 Nxf6+ Nxf6 g7 Kf7 exf6 00:13:15.2 0,30 11 107677884 Nc6 hxg6 fxg6 Nxh7 Qe7 00:34:17.8 0,03 12 281794255 Nc6 hxg6 fxg6 Nxh7 Ne7 Qg4 00:51:28.4 0,23 12 428346574 Nxe5 fxe5 Qc7 Nxh7 Qxe5+ Kf1 Nd7 hxg6 Betazoid T90b with 13 Mb. Nxe5 fails low at 11 ply, and that takes some time. Probably have to run this again to see if I can reproduce this. But glad that Nc6 or Nxe5 are chosen (I'm assuming they are better than Nf6 or Be7). 00:00:01.0 1,83 1 26 gxh5 Bxh7+ Kh8 Qxh5 00:00:01.0 3,71 1 28 Nc6 00:00:01.0 3,78 1 48 Qb6 00:00:01.1 2,76 2 194 Qb6 hxg6 hxg6 00:00:01.1 2,86 2 302 Bc5 hxg6 hxg6 00:00:01.1 3,14 2 975 h6 a3 00:00:01.1 3,35 3 1855 h6 Nf3 gxh5 Rxh5 00:00:01.2 3,14 4 7298 h6 Nxf7 Kxf7 Qg4 g5 fxg5 00:00:01.3 0,94 5 24413 h6 hxg6 Nxe5 fxe5 hxg5 g7 Kxg7 00:00:01.4 1,86 5 38251 Bc5 Nxh7 Nxe5 Nxf8 Nxd3+ Qxd3 00:00:01.5 2,10 5 51761 Qb6 Nxh7 Kxh7 hxg6+ Kg8 g7 Rd8 Rh8+ 00:00:01.6 2,34 5 58903 Nc5 hxg6 hxg6 Rh7 Nxd3+ Qxd3 Qxg5 Rg7+ 00:00:01.7 2,98 5 67231 Be7 hxg6 hxg6 Nf3 Qb6 a4 00:00:02.2 3,25 6 106818 Be7 hxg6 hxg6 Nf3 Qb6 a3 Nc6 00:00:04.5 2,17 7 378767 Be7 Nxh7 Bh4+ Kf1 Kxh7 hxg6+ Kg7 Qh5 00:00:10.6 1,76 8 1069190 Be7 Nxh7 Bh4+ Kf1 Nxe5 Nxf8 Nxd3 Qxd3 Kxf8 Ba3+ Kg7 00:00:25.7 1,56 9 2902544 Be7 hxg6 hxg6 Bxg6 Bxg5 Qh5 00:01:22.2 0,89 10 10086240 Be7 Nxh7 Bh4+ Kf1 Kxh7 Qg4 Kg7 Rxh4 00:02:52.6 1,02 10 22500670 Nxe5 fxe5 Qc7 Qe2 f6 exf6 Rxf6 hxg6 Qg3+ 00:11:47.2 0,14 11 100072939 Nxe5 fxe5 Qc7 Nxh7 Qxe5+ Kf1 00:19:59.2 0,30 11 167584346 Nc6 Nxh7 Bc5 Nxf8 Nxf8 hxg6 fxg6 Qg4 Just trying to be complete regards, 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.