Author: Andrew Williams
Date: 11:57:35 01/28/02
Go up one level in this thread
On January 28, 2002 at 14:18:14, Gerd Isenberg wrote: >From CCT4-10 IsiChessX - PostModernist 0-1 > >Does your program avoid Qc8 ? > >[D] 6k1/4bppp/2p3n1/5Q2/1qPB4/1P1R1BP1/r6P/7K w - - am Qc8 > My program can't avoid it in 15 minutes. >In this position after 34.Qc8+ Nf8, IsiChess failed even more with 35.Qxc6?? > >[D] 2Q2nk1/4bppp/2p5/8/1qPB4/1P1R1BP1/r6P/7K w - - am Qxc6 > >After 34.Qc8? Nf8 IsiChess got a FailLow with Qxc6 and a very bad search >behaviour. The first one at depth 10 from +130 to +95 after 0:10, a second one >at depth 13 after 1:45 resp. 2:06 from +106 ..via +65-- to -164. Few moments >later, i got an exploding search tree looking for 35.Bc3. > >After 11:10 it came back with a score of -0.88 and after 18:18 it found Rd1 with >-0.83. But unfortunally during the game, my maxTime exceeds (aprox. 3*averadge >time), signaled the Search thread to terminate unconditionally after aprox. 8 >Minutes - and IsiChess played Qxc6?? > >Seems that the FailLow behaviour of my Search needs some work, may be too many >extensions, but the wrong. Also i have to think about interrupting the Search in >these FailLow situations. Most often my branching factor seems OK, but sometimes >these terrible exceptions. I was just coming on to CCC to post about this very situation when I saw your message. I think this is simply an abominable position. PostModernist shows exactly the same behaviour here. I presume because of the mate threats on White's back rank. Indeed this is PM's search output from the game (I have truncated the PVs for clarity): --- Pondering (GWT=1), assuming move: Qxc6 --- Catching up clocks : 8085 --> 8893 MOVES SINCE BOOK: 23 1= -116 0 148 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 2= -91 0 333 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 3= -121 0 1521 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 4= -103 0 4801 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 5= -114 0 18803 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 6= -90 0 20720 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 7= -101 0 105672 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 8= -90 0 106789 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 9= -100 3 845483 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 10= -91 4 1085384 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 11= -102 13 3302402 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 SCRUBBING TT (on new guess): -302 [102] 98 12= 183 54 11534068 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 Score dropped 30+; extending time to investigate. Extending time by 70 percent Score dropped 60+; extending time to investigate. Extending time by 62 percent Score dropped 90+; extending time to investigate. Extending time by 18 percent 13= 36 181 39624177 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 SCRUBBING TT (on new guess): -164 [240] 236 14= 238 394 83089278 35...Qe1 36.Bg1 Ra1 37.Qb6 Qf1 38.Be4 Qe2 39.Bf5 --- Pondering finished --- Notice that the scores are from PostModernist's point of view. So it thought it was losing up to ply 11, then suddenly started to see some possibilities at depth 11. Then the score drops at depth 13, only to rise again at depth 14. So this is a position where my program also can't make up its mind what is going on. In fact during the game, seeing the "Score dropped ..." messages and the score of 183 at ply 12, I managed to confuse myself and thought the score had fallen to -183! > >Very impressed of PostModernist 33...Qb4! > I'd like to claim that PM was "fighting for the initiative" or "trying to generate counterplay", but there's no such code in my program. I'd even like to claim that it picks Qb4 and sticks with it forever, but it doesn't even do that. Although in the game that was the only move it considered, without the benefit of pre-filled hash tables it takes 40 seconds to discover it, then drops it at 198 seconds and only regains it at 616 seconds. Still, it *was* a nice move in a bad situation... >cheers, Gerd Best regards, Andrew
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.