Author: Vincent Diepeveen
Date: 00:56:34 09/17/01
Go up one level in this thread
On September 17, 2001 at 01:12:08, Uri Blass wrote: >On September 16, 2001 at 22:36:29, Slater Wold wrote: > >>On September 16, 2001 at 17:38:05, Uri Blass wrote: >> >>>On September 16, 2001 at 16:30:27, Bruce Moreland wrote: >>> >>>>On September 16, 2001 at 16:15:32, Vincent Diepeveen wrote: >>>> >>>>>On September 16, 2001 at 00:36:22, Bruce Moreland wrote: >>>>> >>>>>>Vincent emailed me and asked me to run these. I ran them on a quad 450 Xeon. >>>>>> >>>>>>[D]8/p4bpk/7p/3rq3/3Npp2/PPQ3P1/3R1PKP/8 w - - 0 1 >>>>>> >>>>>>I get gxf4 until ply 10, after which there is a switch to b4, which is a bad >>>>>>move. This fails low to -3.88 in ply 11, and gxf4 comes back with a score of >>>>>>-2.48. >>>>>> >>>>>>Later in ply 11, f3 pops up with a score of -2.33. >>>>>> >>>>>>Up to here takes 89 seconds. >>>>>> >>>>>>f3 sticks until ply 14, at which point if fails low to -4.24, and gxf4 comes >>>>>>back with a score of -3.21, resolving after about 1/2 hour. >>>>>> >>>>>>In ply 15, gxf4 fails low again, and the hour ended with no resolution. It was >>>>>><= -3.46. >>>>>> >>>>>>[D]8/p4bpk/7p/3rq3/3Npp2/PPQ2PP1/3R2KP/8 b - - 0 1 >>>>>> >>>>>>In this one, I have exf3+, with a score of +2.35 in ply 10. After 16 seconds, >>>>>>in ply 10, it finds Bh5, failing high to +2.80. >>>>>> >>>>>>Ply 11 was uneventful, but in ply 12, Bh5 failed high to +4.24. >>>>>> >>>>>>The score creeps up slightly, and the last score I got in the hour was +4.99, >>>>>>ply 15, after about 45 minutes. >>>>>> >>>>>>I'll run the first one again all night and see what happens. >>>>>> >>>>>>bruce >>>>> >>>>>thanks! >>>>> >>>>>It confirms already what i suspected. f3 is losing >>>>>way harder than alternatives! >>>>> >>>>>I do not understand why ferret needs a ply more to get -4.24 for f3 >>>>>than when after f3 is getting played. Possible to shine any light onto >>>>>this? >>>> >>>>I don't see that you can draw that conclusion at all, especially in light of >>>>this: >>>> >>>> ply milliseconds score line >>>>no 17 27657860 -433 -1221224784 gxf4 Qxf4 Kh1 Rh5 f3 exf3 Nxf3 Rh3 Rf2 >>>> Bd5 Rg2 g5 Qd3+ Qe4 Qxe4+ Bxe4 Nd2 Bxg2+ >>>> Kxg2 Rd3 Ne4 Rxb3 a4 Rb4 Nc5 Rb2+ Kg3 >>>> -- >>>>no 17 41753610 -499 666953819 gxf4 Qxf4 f3 Bh5 Kf1 e3 Re2 Rxd4 Rxe3 >>>> Rd1+ Re1 Qxf3+ Qxf3 Bxf3 Rxd1 Bxd1 b4 Kg6 >>>> Kf2 Bc2 Ke3 Kf5 b5 >>>>no 17 50832840 -494 -1178700567 f3 Bh5 g4 e3 Rd1 Bg6 Qc4 e2 Re1 Rxd4 Qxe2 >>>> Qxe2+ Rxe2 Rd3 b4 Rxa3 h4 Bd3 Rd2 Rc3 Kh3 >>>> >>>>Column 4 (node count) is broken and should be ignored. >>>> >>>>gxf4 failed low to -4.99 in ply 17, and f3 overtook it at -4.94. So mine would >>>>play f3 after 50,832 seconds. >>>> >>>>bruce >>> >>>I think the differece between f3 and gxf4 is a positional difference and we have >>>no logfiles of dep blue from that game so it may be interesting to look at the >>>following position >>> >>>[D]Rr6/5kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 w - - 0 1 >>> >>>Here is Deeper blue logfile >>> >>>--------------------------------------- >>>hash guess Rb7b8,Guessing Rb8 >>> 8(4) #[Ra6](156)[Ra6](156) 156^ T=0 >>>ra8a6 Qb6c6q pd5c6Q Kf7g8 be4d5 Kg8h7 ra6a7 Rb8c8 ra7b7 Ph6h5 rb7b5P Kh7h6 rb5b7 >>>Rc8c7 rb7b8 >>> 8(6) #[Ra6](128)########################## 128 T=3 >>>ra8a6 Qb6c6q pd5c6Q Kf7g8 be4d5 Kg8h7 ra6a7 Rb8c8 ra7b7 Ph6h5 rb7b5P Kh7h6 rb5b7 >>>Rc8c7 rb7b8 >>> 9(6) #[Ra6](128)###########<ch> 'rb8' >>>[183 sec (main.c:1847)][cont]############### 128 T=10 >>>ra8a6 Qb6c6q pd5c6Q Kf7g8 be4d5 Kg8h7 ra6a7 Rb8c8 ra7b7 Ph6h5 rb7b5P Kh7h6 rb5b7 >>>Rc8c7 rb7b8 >>>10(6) #[Ra6](158)[Ra6](158) 158^ T=17 >>>ra8a6 Qb6c6q pd5c6Q Kf7g8 be4d5 Kg8h7 ra6a7 Rb8c8 ra7b7 Ph6h5 rb7b5P Kh7h6 rb5b7 >>>Rc8c7 rb7b8 >>>10(6) #[Ra6](158)########################## 158 T=34 >>>ra8a6 Qb6c6q pd5c6Q Kf7g8 be4d5 Kg8h7 ra6a7 Rb8c8 ra7b7 Ph6h5 rb7b5P Kh7h6 rb5b7 >>>Rc8c7 rb7b8 >>>11(6) #[Ra6](156)########################## 156 T=99 >>>ra8a6 Qb6c6q pd5c6Q Kf7g8 be4d5 Kg8h7 ra6a7 Rb8c8 ra7b7 Ph6h5 rb7b5P Kh7h6 rb5b7 >>>Rc8c7 rb7b8 >>>12(6) #[Ra6](162)[TIMEOUT] 162 T=192 >>>ra8a6 Qb6c6q pd5c6Q Rb8c8 be4d5 Kf7e7 ra6a5 Bd6c7 ra5b5P Ke7d6 bd5f3 Kd6e7 rb5c5 >>>Rc8a8 bf3d5 Ke7d6 >>> >>>You can see that the evaluation of Deeper blue at depth 12(6) was 1.62 pawn for >>>white. >>> >>>I am interested to know the evaluation of Ferret at depth 18 in order to compare >>>with deeper blue. >>> >>>It may be interesting to see also the evaluation of Ferret when you increase the >>>value of pieces that are not pawns to 150% of their normal value(programs may >>>find quickly that black gets 2 pawns for the piece after Ra6 Qe3 so in order to >>>prevent finding Ra6 Qe3 for these reasons I suggest to increase the value of >>>pieces that are not pawns) >>> >>>I ask for Ferret's evaluation at depth 18 because other programs do not use >>>singular extensions. >>> >>>I suspect that Ferret can find Ra6 Qe3 in both cases at depth that is smaller >>>than 18 and that it can also cahnges it's mind later to Qd7+ >>> >>>Uri >> >>Here is Nimzo 8's eval; it also uses SE. >> >>Nimzo 8 - W,S >>Rr6/5kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 w - - 0 1 >> >>Analysis by Nimzo 8: >> >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ >> = (0.24) Depth: 3/14 00:00:00 >>1.Qd7+ Kg8 2.Ra7 Bf8 3.Qe6+ Kh8 4.Qxb6 >> ² (0.52) Depth: 3/14 00:00:00 >>1.Qd7+ Kg8 2.Ra7 Bf8 3.Qe6+ Kh7 4.Qxb6 Rxb6 >> ² (0.64) Depth: 5/17 00:00:00 >>1.Qd7+ >> ± (0.89) Depth: 6/19 00:00:00 >>1.Qd7+ Kg8 2.Ra7 Bf8 3.Qe6+ Kh7 4.Qxb6 Rxb6 5.g4 >> ± (1.05) Depth: 6/19 00:00:00 >>1.Qd7+ >> ² (0.60) Depth: 7/20 00:00:00 >>1.Qd7+ >> ² (0.61) Depth: 7/20 00:00:00 >>1.Qd7+ Kg8 2.Ra7 Bf8 3.Qe6+ Kh7 4.Qxe5 >> ² (0.60) Depth: 7/20 00:00:00 >>1.Ra6 >> ² (0.61) Depth: 7/20 00:00:00 >>1.Ra6 Qxc6 2.dxc6 Rd8 3.Ra7+ Kf8 4.g4 Ke8 5.c7 g6 >> ± (0.77) Depth: 7/20 00:00:00 >>1.Ra6 >> ± (1.02) Depth: 8/22 00:00:00 >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 h5 4.Rd7 Be7 5.g3 Ke8 >> ± (1.31) Depth: 8/22 00:00:01 >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rc8 4.Rb7 Rb8 5.Rd7 >> ± (1.36) Depth: 9/23 00:00:01 532kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rc8 4.Rb7 Ra8 5.Rxb5 Ra3 6.Rd5 >> ± (1.30) Depth: 10/25 00:00:02 532kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rd8 4.Rb7 Rc8 5.Rxb5 Ke7 6.Kf2 >> +- (1.44) Depth: 11/26 00:00:03 2522kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rd8 4.Rb7 Rc8 5.Rxb5 Ke7 6.Ke2 >> +- (1.68) Depth: 12/28 00:00:29 30325kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rc8 4.g4 h5 5.Rb7 Ra8 6.Rd7 >> +- (1.76) Depth: 13/29 00:01:32 96689kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rc8 4.g4 h5 5.gxh5 Bc7 6.Rb7 >> +- (2.00) Depth: 14/31 00:02:39 166527kN >> >>(W, 16.08.2001) > >I see that you stopped it at depth 14 when nimzo could find Qe3 in the main line >at depth 15. > >I also doubt if it is using singular extensions because the number of plies >here suggest that it did not search more than 31 plies in this search and not >more than 37 plies in the second search > >Note that I doubt if the second number means the number of plies in the >selective search because I saw shredder gives a mian line of 10 plies at depth >1/2 so at least I know that the second number in shredder is not the selective >search depth. > >> >> >>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ >> >>And then I editing the piece value to 150% of it's original value: >> >> >>Nimzo 8 - W,S >>Rr6/5kp1/1qQb1p1p/1p1PpP2/1Pp1B3/2P4P/6P1/5K2 w - - 0 1 >> >>Analysis by Nimzo 8: >> >>1.Qd7+ >> = (-0.08) Depth: 1/11 00:00:00 >>1.Qxb6 Rxb6 2.g4 >> = (-0.03) Depth: 1/11 00:00:00 >>1.Qd7+ >> = (0.10) Depth: 1/11 00:00:00 >>1.Qd7+ >> = (0.10) Depth: 1/11 00:00:00 >>1.Qd7+ >> = (0.10) Depth: 1/11 00:00:00 >>1.Qd7+ >> = (0.10) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 1/11 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 5/17 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 5/17 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 5/17 00:00:00 >>1.Qd7+ Kg8 2.Rxb8+ Qxb8 3.g4 >> = (0.19) Depth: 5/17 00:00:00 >>1.Qd7+ >> ² (0.44) Depth: 5/17 00:00:00 >>1.Qd7+ Kg8 2.Ra7 Bf8 3.Qe6+ Kh7 4.Qxb6 Rxb6 >> ² (0.64) Depth: 5/17 00:00:00 >>1.Qd7+ >> ± (0.89) Depth: 6/19 00:00:00 >>1.Qd7+ Kg8 2.Ra7 Bf8 3.Qe6+ Kh8 4.Qxb6 Rxb6 5.g4 >> ± (1.08) Depth: 6/19 00:00:00 >>1.Qd7+ >> ² (0.63) Depth: 7/20 00:00:00 >>1.Qd7+ Kg8 2.Ra7 Bf8 3.Qe6+ Kh8 4.Qxb6 Rxb6 5.g4 Bd6 >> ² (0.58) Depth: 7/20 00:00:00 >>1.Ra6 >> ² (0.59) Depth: 7/20 00:00:00 >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rc8 4.Rb7 Rb8 5.c7 >> ± (0.78) Depth: 7/20 00:00:00 >>1.Ra6 >> ± (1.03) Depth: 8/22 00:00:01 >>1.Ra6 Qxc6 2.dxc6 Kg8 3.Ra7 Rd8 4.g4 h5 5.Ke2 >> ± (1.04) Depth: 8/22 00:00:01 >>1.Qd7+ >> ± (1.05) Depth: 8/22 00:00:01 >>1.Ra6 Qxc6 2.dxc6 Kg8 3.Ra7 Rd8 4.g4 h5 5.Ke2 >> ± (1.04) Depth: 8/22 00:00:01 >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rc8 4.Rb7 h5 >> ± (1.23) Depth: 9/23 00:00:01 861kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rd8 4.Rb7 Bxb4 5.cxb4 c3 >> ± (1.29) Depth: 10/25 00:00:02 861kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rc8 4.Rb7 Ra8 5.Rd7 Be7 6.c7 >> +- (1.45) Depth: 11/26 00:00:04 3896kN >>1.Ra6 Qxc6 2.dxc6 Kf8 3.Ra7 Rd8 4.Rb7 Be7 5.Rxb5 Bxb4 6.cxb4 >> +- (1.69) Depth: 12/28 00:00:12 13364kN >>1.Ra6 >> +- (1.94) Depth: 13/29 00:00:28 29837kN >>1.Ra6 Qxc6 2.dxc6 Rc8 3.Ra7+ Rc7 4.Ra5 h5 5.g4 hxg4 6.Ke2 >> +- (1.95) Depth: 13/29 00:00:41 43514kN >>1.Ra6 Qxc6 2.dxc6 Rc8 3.Ra7+ Rc7 4.Ra5 Rc8 5.Rxb5 Ke7 6.Ra5 >> +- (2.12) Depth: 14/31 00:04:35 303184kN >>1.Ra6 Qe3 2.Qd7+ Kg8 3.Qxd6 Rf8 4.Qe6+ Kh8 5.Qe7 Rg8 6.Bf3 >> +- (2.32) Depth: 15/32 00:11:30 769135kN >>1.Ra6 Qe3 2.Qd7+ Kg8 3.Qxd6 Rf8 4.Qe6+ Kh8 5.Qe7 Rg8 6.Bf3 >> +- (2.32) Depth: 16/34 00:26:19 1792190kN >>1.Ra6 Qe3 2.Qd7+ Kg8 3.Qxd6 Rf8 4.Qe6+ Kh8 5.Qe7 Rg8 6.Bf3 >> +- (2.22) Depth: 17/35 01:22:39 5740900kN >>1.Ra6 >> +- (1.77) Depth: 18/37 03:32:49 14934641kN >> >>(W, 16.08.2001) >> >>~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ >> >>Slate > >I am really surprised that when the pieces are 150% of their original value it >can find Qe3 at depth 15 without draw evaluation. > >Another surprising thing is that I see that the score for Ra6 Qxc6 is lower then >the score for Ra6 Qxc6 when the pieces have their original value. > >I expected the score for Ra6 Qxc6 not to be changed by increasing the value of >pieces because I believe that programs cannot see that white wins a piece in >that line but for some reason it does get changed. In fact they all see that white wins 2 pawn quick. But what it's all about is the fact that there are opposite bishops on the board then. Some programs give this a huge penalty. I am very sure DBII gave this a major penalty. We can see that because it played Be4 instead of winning Qb6 for example at move 37. It played axb5 for that reason some moves before that too. DIEP 1997 also had a major penalty for opposite bishops. Causing it to draw an opposite bishop endgame in world championship versus ferret with 3 pawns less for DIEP. DIEP 1997 also played Be4 there first 10 ply. DIEP 2001 doesn't give a major penalty for the right reasons here and therefore doesn't play Be4 because it sees Qb6 to be winning. Also here with the 1.62 score we must take into account that DB most likely saw the pawns lost, Singular extensions are made to extend that, but most likely the Qe3 line where it did not give the opposite bishop penalty (because queens on the board) it of course got the full 'piece down' penalty. From the 1.62 score we can easily see how much penalty db already got from opposite bishops easily. Compare scores for Qxc6 with other programs! Note that increasing piece values is not a smart idea, depeninding upon how a program has been made. Most do material incremental in makemove and hardcode piece values there. An assembly program like nimzo might do this. Also default piece values of nimzo are not exactly what it is with other programs. Lineair scaling will not work very well because penalties for small exchange and such will go wrong then. >1.76/13 1.95/13 >2.00/14 2.12/14 > >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.