Author: Dieter Buerssner
Date: 08:58:10 06/16/03
Go up one level in this thread
On June 16, 2003 at 02:48:35, Uri Blass wrote:
>On June 16, 2003 at 00:57:16, Harald Faber wrote:
>
>>On June 15, 2003 at 15:59:25, Dieter Buerssner wrote:
>>
>>>On June 15, 2003 at 13:00:40, Harald Faber wrote:
>>>
>>>>8/8/8/5k2/1pK4P/6P1/8/4b3 b - - 0 1
>>>But I can prove (needed perhaps 20 minutes) that with Kg4 black can announce a
>>>mate in 24 (it could be a faster mate):
>>>
>>> 252678 0.967 Mat24 17t 1...Kg4 2. Kb3 Bd2H 3. Ka4H Bc3H 4. Kb3H Kf3H
>>> 5. Ka4H Ke3H 6. h5H Kd2H 7. Kb3H Kd3H 8. h6H
>>> Kd4H 9. h7H Kc5H 10. g4H Bh8H 11. g5H Bd4H 12.
>>> Ka4H Bg1H 13. h8=QH Be3H {HT}
>>>
>>>Note that most moves are out of hash tables (the ones with H attached). Those
>>>are not necessarily reliable (however the further up ones should be rather
>>>reliable). The score must be reliable nevertheless.
>>Did you notice that in this line the idea with the king is OK but that white
>>gets a queen with 13.h8-Q and black cannot capture this piece?
>>Maybe you should play through this line, I am sure Yace will find the right
>>moves then too, because up to move 9 everything is fine.
Yace already found the right moves. It just cannot not display them reliably.
The score shown can be considered reliable (shorter mates are possible, but a
mate in 24 is proven).
After 11...Bd4 I get:
2133358 26.261 -Mat13 17t 12. g6 Bc3 13. Kc2 Kc4 14. Kc1 b3 15. Kd1 b2
16. Kc2 Bh8 17. g7 Bxg7 {EGTB} 18. Kd2! b1=Q!
19. Ke3! Qf1 20. Ke4! Bh8 21. Ke3 Bf6! 22. Ke4!
Bg5! 23. Ke5! Qf6+ 24. Ke4 Qf4#! {-1271}
And later
41246901 5:12.2 -Mat12 21t 12. g6 Bc3 13. Kc2 Kc4 14. Kc1 b3 15. Kd1 b2
16. Kc2 Bg7 17. Kd2 b1=Q 18. Ke2H Qf5H 19.
h8=QH Bxh8H {EGTB} 20. g7 Bxg7! 21. Ke3 Bc3!
22. Ke2 Qf4! 23. Kd1 Qd2#! {-1351}
Possibly, when I started here again (I did not try), I could get mate in 23 from
the root.
After 12. Ka4:
46798 0.653 Mat10 10t 12...Kc4 13. Ka5 b3 14. Ka6 b2 15. Kb7 b1=Q+
16. Kc6 Qxh7 {EGTB} 17. g6 Qe7 18. g7 Bxg7! 19.
Kb6 Qd7 20. Ka6! Bd4 21. Ka5 Qa7# {1351}
>Did you notice the letter H?
>I understood that it means that the moves were constructed from hash tables and
>cannot be trusted.
>
>Note that I have no idea what is the way of chess programs to use hash tables to
>construct pv lines and the only clear thing is that it leads often to wrong
>results(I am sure that yace does not remember mate score for black after
>13.h8=Q).
Correct. Indeed in the case of mate scores, one could make the displayed PV more
reliable, by looking at the scores stored. I think in more general cases this
would not be the case (at least not that easy). Until now, I was too lazy to
implement the idea. Actually it would only be really useful for showing PVs
somewhere like here in a forum. I got the score by some back analysis. When I
have the mate score at the root, Yace is more or less in a state, where it can
show very fast the correct continuation everywhere in the line and in the side
lines.
I think, the status quo in Yace (where it warns by the "H") is not that bad and
impractical. It would of course be very easy, to only let it show very short
lines (that really are reliable), like regularily seen in some other programs.
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.