Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: test position

Author: Ron Murawski

Date: 14:18:37 07/22/02

Go up one level in this thread


On July 19, 2002 at 23:01:01, Robert Hyatt wrote:

>On July 19, 2002 at 12:46:57, Ron Murawski wrote:
>
>>On July 19, 2002 at 01:54:47, Steffen Jakob wrote:
>>
>>>On July 18, 2002 at 23:52:40, Ron Murawski wrote:
>>>
>>>>On July 18, 2002 at 22:24:13, K. Burcham wrote:
>>>>
>>>>>
>>>>>here GM Timman 2623 FIDE played 14.Nxf6
>>>>>is there any program that prefers to exchange Knights at f6, instead of the pawn
>>>>>capture at e7?
>>>>>
>>>>>kburcham
>>>>>
>>>>>
>>>>>
>>>>> [D] rnbq1r2/4ppkp/p2p1np1/3N4/1p1N4/4Q2P/PPP2PP1/2KR1B1R w - - 0 1
>>>>
>>>>On a 1.8GHz AMD, it takes the latest Horizon beta 14.12 seconds to find Nb6.
>>>>Horizon agrees with Chessmaster that this is the best move.
>>>>
>>>>Depth Score   Time      Nodes   pv
>>>>  4     47      44      16293   Nxe7 Re8 Re1 Kh8
>>>>  4    115      44      16673   Nxf6 exf6 Qe4 Ra7
>>>>  4    128      49      17552   Nxb4 Kh8 Qh6 Ra7
>>>>  4    137      49      19005   Bc4 Kh8 Nxe7 Ra7
>>>>  5     79?     55      30021   Bc4 Bb7 Ne6+ fxe6 Bxa6
>>>>  5    115      60      35197   Nxb4 Kh8 Qf3 Ra7 Bc4
>>>>  5    142      60      41915   Nxe7 Bb7 Ndf5+ Kh8 Qh6 d5
>>>>  6    128      82     103073   Nxe7 Re8 Re1 Kh8 Bc4 Ra7
>>>>  7    111     143     279955   Nxe7 Re8 Re1 Bb7 f3 Be4 Bd3 Rxe7 fxe4
>>>>  8     82?    247     652614   Nxe7 Re8 Re1 Qc7 Qf3 Ra7 Kd2
>>>>  8     82     478    1481058   Nxe7 Kh8 Qf3 d5 Nxc8 Ng8
>>>>  9     61     604    1889811   Nxe7 Re8 Qf3 Rxe7 Qxa8 Bb7 Qa7 Bxg2 Qxe7 Qxe7
>>>>                                Bxg2
>>>>  9    111!   1412    4493054   Nb6 Qxb6 Nf5+ Bxf5 Qxb6 Bxh3 Rxh3
>>>                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>>That is a weird PV!
>>>
>>>Greetings,
>>>Steffen.
>>>
>>
>>Yes, it *is* a weird pv. I have noticed weird lines in Horizon's pv from time to
>>time, but they tend to get replaced quickly. Most of the time the first move in
>>the pv is correct, but every once and a while Horizon can lose a game due to
>>this. I suppose it is caused by search instabilities. I've been searching for
>>this bug for a long time. When I again try and find/fix this bug I will attempt
>>to choose a "better" sequence of 64-bit random numbers for my Zobrist hashing.
>>I'm running out of ideas as to the cause of the occasional pv weirdness.
>>
>>Ron
>>
>>>>  9    114    1648    5265170   Nb6 Bb7 Nxa8 Bxa8 Be2 Bxg2 Rhg1 Be4 Qg5
>>>> 10     89    2422    7893348   Nb6 Bb7 Nxa8 Bxa8 Kb1 e5 Nf3 Bxf3 Qxf3 d5
>>>> 11    117    5471   17568997   Nb6 Ra7 Nxc8 Rb7 Nxd6 exd6 Bc4 Rc7 Bb3
>>>>
>>>>Ron
>
>Did you possibly get a fail high, and then run out of time before you got
>a score?  If so the PV is useless and depending on how you back it up, it
>could be corrupt even.

Hi Bob,

No, that is not the case. As you can see from the above example at depth 9 the
pv went weird and has a score associated with it. It resolved into a more
believable pv about 2 1/2 seconds later. This type of behavior is usually
associated with fairly substantial swing in the eval score and almost always on
a fail high. I'm still looking for the bug...

Ron



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.