Author: Miguel A. Ballicora
Date: 13:08:32 08/28/01
Go up one level in this thread
On August 28, 2001 at 15:34:30, Peter Berger wrote:
>On August 28, 2001 at 11:58:02, Dieter Buerssner wrote:
>
>>
>>However, I don't totally agree, that this is a King safety problem. For many
>>depths, Kd6 and Ke8 have almost the same score (or have the same score). Sure,
>>fiddling a little bit at KS-values will make Yace "find" the correct move
>>muchearlier. I actually lowered some time ago some "K-advance-penalty". But I
>>had the feeling, that this makes Yace weaker after some time, so I put back the
>>old values.
>>
>>Also, this position may be a bit unusual, and not fitting some heuristics used
>>in many chess programs. Usually programs come out of book, in castled position,
>>or with a closed center. Neither is the case here.
>>
>>BTW. An material only eval, never switches away from Kd6.
>
>I agree and never said anything else. You won't want your engine to play Ke8
>here though and lose in about 5 moves if it can be avoided - it definitely is a
>king safety problem ( why else would the engine prefer Ke8 anyway ? ) It is an
>exception of the usual rule - and it obviously can happen in a game ( as it did
>) . Maybe it isn't important as it doesn't happen often.
I tried this with Gaviota because it has almost no king safety evaluation terms,
so it has a big penalty for going to the center when there is a lot
of pieces around. That kills it (yes, I know, my poor engine sucks).
It takes 229 seconds (PIII 700mhz) to fail low and then 10 times more to find
Kd6! I think this is a good test position to make sure that penalties do not get
too big, or at least are balanced with other terms. I haven't worked on that
yet.
It finds rather easily f3 and g3 though...
I have the strange feeling that move ordering could be critical in this
position. What happen if we mirror the position?
+-----------------+
| . . . . . . . r |
| Q . . k . x x x |
| . . b . x . . . |
| . . . . . . . . |
| . . . . . . q . |
| . . B . . . . . |
| . . . . . o o o |
| . . R . . . K . |
+-----------------+
5: 0.1 -0.52 Kd7-e8 f2-f3 Qg4-f4 Rc1-c2 Ke8-f8 Bc3xg7
Kf8xg7 Rc2xc6
6 0.2 :-( Kd7-e8
6 0.3 :-(
6 0.4 -1.60 Kd7-c8 f2-f3 Qg4-g5 Bc3-b2 Qg5-b5 Bb2xg7
6 0.6 -0.93 Kd7-e8 f2-f3 Qg4-g5 Bc3-b2 Bc6-d7 Rc1-d1
6: 0.6 -0.93 Kd7-e8 f2-f3 Qg4-g5 Bc3-b2 Bc6-d7 Rc1-d1
7 1.0 :-) Kd7-e8
7 1.1 +0.00 Kd7-e8 f2-f3 Bc6xf3 Qa7-b8 Ke8-e7 Qb8-c7
Ke7-e8 Qc7-c8 Ke8-e7 Qc8-c7
7: 1.3 +0.00 Kd7-e8 f2-f3 Bc6xf3 Qa7-b8 Ke8-e7 Qb8-c7
Ke7-e8 Qc7-c8 Ke8-e7 Qc8-c7
8 3.7 -0.10 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-g5 Rc1-f1
Bf3-h5 Rf1-f4
8: 3.9 -0.10 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-g5 Rc1-f1
Bf3-h5 Rf1-f4
9 8.0 -0.38 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-g5 Rc1-f1
Bf3-e2 Qa7xf7 Ke8-d8 Rf1-f4 Qg5-e7 Qf7xg7
Qe7xg7 Bc3xg7
9: 8.5 -0.38 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-g5 Rc1-f1
Bf3-e2 Qa7xf7 Ke8-d8 Rf1-f4 Qg5-e7 Qf7xg7
Qe7xg7 Bc3xg7
10 9.8 :-( Kd7-e8
10 24.3 :-(
10 152.7 -1.73 Kd7-d6 f2-f3 Bc6xf3 Rc1-c2 Bf3xg2 Qa7-a3
Kd6-d7 Rc2xg2 Qg4-d1 Kg1-f2 Qd1-d6 Qa3-a4
Qd6-c6
10 164.7 -1.27 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-h3 Qa7-f2
Qh3-g4 Rc1-f1 Ke8-d7 Qf2xf3 Qg4xf3 Rf1xf3
10: 166.1 -1.27 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-h3 Qa7-f2
Qh3-g4 Rc1-f1 Ke8-d7 Qf2xf3 Qg4xf3 Rf1xf3
11 202.7 -1.05 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-h3 Qa7-f2
Bf3-c6 Bc3-b2 Bc6-b7 Bb2xg7 Rh8-g8
11: 211.4 -1.05 Kd7-e8 f2-f3 Bc6xf3 g2-g3 Qg4-h3 Qa7-f2
Bf3-c6 Bc3-b2 Bc6-b7 Bb2xg7 Rh8-g8
12 229.7 :-( Kd7-e8
12 2143.2 -1.42 Kd7-d6 f2-f3 Bc6xf3 Rc1-c2 Bf3xg2 Qa7-a3
Kd6-d7 Rc2xg2 Qg4-d1 Kg1-f2 Qd1-c2 Kf2-f1
Qc2-d3 Kf1-e1 Qd3-e4 Rg2-e2 Rh8-a8
12: 2157.7 -1.42 Kd7-d6 f2-f3 Bc6xf3 Rc1-c2 Bf3xg2 Qa7-a3
Kd6-d7 Rc2xg2 Qg4-d1 Kg1-f2 Qd1-c2 Kf2-f1
Qc2-d3 Kf1-e1 Qd3-e4 Rg2-e2 Rh8-a8
Regards,
Miguel
>>I think, this position tests more the search of the engine. When Yace sees the
>>problem, it gets a fail low to a score < -3. I wondered, why this is seen so
>>late.
>>
>
>Yes, again agreed . Look at Andrew William's confirmation for PM - it seems the
>same happens to Yace here.
>
>
>>First of all, after 1...Ke8, 2.f3 must be found. To see, that f3 is really good,
>>probably (after 2...Bxf3, which is no problem of course), 3. g3 must be found.
>>All those moves (and more in the mainline) don't trigger any extensions in Yace.
>>OTOH in other lines many extensions are triggered (and are mainly useless). So
>>the search depth increases too slowly, and Yace needs too much time.
>>
>>I think, perhaps finding f3 after Ke8 would be another good test position.
>
>f3 with a good line and eval seems to be hard indeed for some of the engines
>that won't find Kd6 .
>
>>
>>And - Yace thinks even after Kd6, it will be difficult to hold the game.
>>After I have analyzed some lines, and gone back (what you called "cheating" :-)
>>I get the following after Kd6:
>>
>> 16702914 1:27.8 1.55 10t 2.f3 Bxf3 3.Rc2 Bxg2 4.Be1H Be4+H 5.Bg3+H
>> Qxg3+H 6.hxg3H Bxc2H 7.Qd4+H Ke7H 8.Qc5+H Kf6H
>> 9.Qxc2H h5H 10.Qd1H g5H 11.Kf2H Rh7H 12.Qd8+H
>> Kg6H 13.Qd3+H Kg7H 14.Qd4+H {HT} {261}
>> 29572054 2:33.8 1.55 10. 2.f3 Bxf3 3.Rc2 Bxg2 4.Be1 Be4+ 5.Bg3+ Qxg3+
>> 6.hxg3 Bxc2 7.Qd4+ Ke7 8.Qc5+ Kf6 9.Qxc2 h5
>> 10.Qd1 g5 11.Kf2 Rh7 12.Qd8+ Kg6 13.Qd3+ Kg7
>> 14.Qd4+ {HT} {261}
>
>I don't get your point here. I agree to your analysis but this is a line where
>black is alive and fighting still although its position isn't nice. Ke8 ?? is
>dead though.
>
>pete
>>
>>Regards,
>>Dieter
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.