Author: Christophe Theron
Date: 17:58:28 08/11/00
Go up one level in this thread
On August 11, 2000 at 02:12:05, stuart taylor wrote:
>On August 11, 2000 at 01:03:49, Ed Schröder wrote:
>
>>On August 10, 2000 at 21:36:23, Robert Hyatt wrote:
>>
>>>On August 10, 2000 at 17:22:17, Dann Corbit wrote:
>>>
>>>>On August 10, 2000 at 16:17:47, Christoph Fieberg wrote:
>>>>
>>>>>Unbelievable, but true:
>>>>>I composed a position which is mate in 1 and can easily found, but Fritz6 on
>>>>>Pentium III, 500 Mhz and 32 MB Hash needs more than 1 hour to show it (excatly
>>>>>1:03:17)!!!
>>>>>How do other computers react?
>>>>>
>>>>>The position is
>>>>>FEN: 8/8/pppppppK/NBBR1NRp/nbbrqnrP/PPPPPPPk/8/Q7 w - - 0 1
>>>>>
>>>>>It is possible to reach this position from starting position:
>>>>>1.Sa3 Sa6 2.Sc4 Sc5 3.Sa5 Sa4 4.b3 b6 5.La3 La6 6.Lc5 Lc4 7.h4 h5 8.Th3 Th6
>>>>>9.Td3 Te6 10.Td5 Te4 11.Sh3 Sh6 12.e3 e6 13.Le2 Le7 14.Lf3 Lf6 15.d3 d6 16.Ke2
>>>>>Ke7 17.Dd2 Lc3 18.Te5 Lb4 19.Th1 Td4 20.Lc6 Dd7 21.Lb5 Th8 22.Td5 Kf6 23.Sg1 Sg4
>>>>>24.Th3 Se5 25.Tg3 Kf5 26.Tf3+ Kg4 27.Tf5 a6 28.g3 c6 29.Sf3 Kh3 30.Sg5+ Kh2
>>>>>31.Se4 Sg6 32.Kf3 Se5+ 33.Kf4 Sg6+ 34.Kg5 Se5 35.a3 Th6 36.c3 Tf6 37.De2 g6
>>>>>38.Kh6 Sg4+ 39.Kh7 De7 40.Tf4 Tf5 41.Sg5 Df6 42.Df3 De5 43.Kg8 De4 44.Sh3 Tg5
>>>>>45.Tf6 Sh6+ 46.Kh7 Sg4 47.Df5 Kg2 48.Kg7 Sh2 49.De5 Tg4 50.Sg5 Sf3 51.Sh7 Sg1
>>>>>52.Df4 [52.Tf5 Se2 53.Sf6 Sf4 54.Se8 Kh2 55.Tg5 Kg2 56.Kh7 Kh2 57.Sg7 Kg2 58.Sf5
>>>>>f6 59.f3 Kh2 60.Kh6 Kh3 Test it! What time do computers need to reach a certain
>>>>>search depths?] 52...Se2 53.Df3+ Kh2 54.Tff5 Sf4 55.Dd1 Kg2 56.Tg5 Kh2 57.Sf6
>>>>>Kg2 58.Sg8 Kh2 59.Se7 Kg2 60.Sf5 Kh2 61.Da1 Kg2 62.f3 f6 63.Kh6 Kh3 Mate in 1!
>>>>>(64. Dh1#). What time do different programms need to show it in the display? 1-0
>>>>>
>>>>>52.Tf5 leads to a position with White Queen on e5 (instead on a1).
>>>>>FEN: 8/8/pppppppK/NBBRQNRp/nbbrqnrP/PPPPPPPk/8/8 w - - 0 61
>>>>>
>>>>>How long do computers need to reach for example depth 9?
>>>>>Fritz 6 reached depth 5 after 9 hours!! (on Pentium III, 500 MHz, 32 MB Hash) He
>>>>>showed 61. Dxe4 Txe4 62. dxe4 Lxc5 3. Txg4 hxg4 as best moves (+1.56 for White).
>>>>>What is the best move?
>>>>
>>>>chest finds the solution after examination of exactly one position (the first
>>>>one it tried):
>>>>
>>>>E:\chest-3.19>chest319 -b epd.epd
>>>>[D] 8/8/pppppppK/NBBR1NRp/nbbrqnrP/PPPPPPPk/8/Q7 w - - acn 1; acs 0; bm Qh1#; ce
>>>>32766; dm 1; pv Qh1#;
>>>>
>>>>
>>>>Crafty found it at ply 2:
>>>>E:\PROGRA~2\winboard\Crafty>crafty
>>>>EPD Kit revision date: 1996.04.21
>>>>found computer opening book file [e:/crafty/release/bookc.bin].
>>>>hash table memory = 24M bytes.
>>>>pawn hash table memory = 4M bytes.
>>>>EGTB cache memory = 2M bytes.
>>>>draw score set to 0.00 pawns.
>>>>choose from book moves randomly (using weights.)
>>>>choose from 5 best moves.
>>>>book learning enabled
>>>>result learning enabled
>>>>position learning enabled
>>>>threshold set to 9 pawns.
>>>>4 piece tablebase files found
>>>>1302kb of RAM used for TB indices and decompression tables
>>>>
>>>>Crafty v17.11
>>>>
>>>>White(1): epdpfga epd.epd epd.out
>>>>PFGA: EPD record: 1
>>>>middle-game phase
>>>> clearing hash tables
>>>> time surplus 0.00 time limit 30.00 (3:00)
>>>> nss depth time score variation (1)
>>>> 1 43.87 7.19 1. dxe4 Rxd5 2. Rxg4 hxg4 3. exd5 Bxa5
>>>> 4. Bxa4 dxc5 5. gxf4 gxf5 6. bxc4 cxd5
>>>> 7. cxd5 exd5
>>>> 1 50.87 ++ 1. Qh1#!!
>>>> 1 50.87 Mate 1. Qh1#
>>>> 1-> 50.87 Mate 1. Qh1#
>>>> 2 50.87 Mate 1. Qh1#
>>>> 2-> 50.89 Mate 1. Qh1#
>>>> time=51.17 cpu=99% mat=0 n=5928139 fh=100% nps=115851
>>>> ext-> checks=11 recaps=0 pawns=0 1rep=1 thrt:0
>>>> predicted=0 nodes=5928139 evals=4531128
>>>> endgame tablebase-> probes done=0 successful=0
>>>> hashing-> trans/ref=1% pawn=92% used=0%
>>>>
>>>>I had a lot of stuff in memory, so it took crafty a long time to start up.
>>>
>>>
>>>That wasn't the problem. the first move searched (the one a simple eval/
>>>material ordering suggests) takes forever, and millions of nodes to search.
>>>
>>>Programs that like the check first will solve it instantly. programs that
>>>look at something else first may well get lost in a huge tree of captures,
>>>and take a lot of time to get off the captures and onto the simple mating
>>>move.
>>
>>Here is another one that might blow up the tree on the first ply (and next
>>iterations) in case you don't limit q-search.
>>
>>[d]4K3/PPPPPPPP/8/8/8/8/pppppppp/4k3 w - -
>>
>>Lots of fun.
>>
>>Ed
>
>So how can we ever hope for computers to make GREAT discoveries, if we don't
>give them a free hand to look for everything?
>S.Taylor
I think you don't understand what the problem is.
The programs do look at every possible move, and that's why it takes so long to
some of them to find the right answer.
In the positions given, there are many possible captures and promotions. Even if
you only look at your next possible move, are you going to ignore the captures
or promotions that your opponent can do, just after your have moved?
Even a beginner would not ignore the possible captures of his opponents. But
when there are too many of them, a human player simply cannot look at them all.
So it plays a little bit at random.
Some chess programs will not do that, and will try to evaluate all the possible
captures or promotions in order to not play at random. Unfortunately, in these
artifical positions, this strategy fails because it takes an incredible long
time to search all the captures and recaptures.
Don't blame the chess program for MISSING the mate. They don't MISS it. The
problem is that before they look at the mating move they first look at other
moves, and unfortunately the consequences of these moves are terribly complex to
evaluate.
Christophe
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.