Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Wierd hash problems

Author: James Robertson

Date: 02:15:26 08/20/99

Go up one level in this thread


On August 20, 1999 at 05:01:55, Bruce Moreland wrote:

>
>On August 20, 1999 at 04:34:49, James Robertson wrote:
>
>>My program played a game in which this position arose:
>>
>>r3r1k1/3b1p1p/2p4q/p2p4/3P2p1/1PNPPpP1/P1RQ1P1P/5RK1 w - -
>>
>>It then played, after an 8 ply search, 1. e4?? which loses immediately to Qh3
>>and the loss of the queen after 1. e4?? Qh3 2. Qg5+ 3. Qf8 Qh6+ 4. Qxh6, an
>>incredibly simple tactical sequence. I later ran my program on this position and
>>it switched to the much much better move Kh1 after 6 plies.
>>
>>The only differece I can think of between the two searches would be the hash
>>table? During the game it had entries from previous searches, and afterwards I
>>ran it as a testsuite position with a clean table. I cannot fathom why else my
>>program would miss such a thing in an actual game. Is there any reason why a
>>hash table would do this? Is there some depth/draft thing I am messing up with
>>my table? Is my null move causing problems? Any help anyone can provide is
>>greatly appreciated.
>>
>>James
>>
>>P.S. Here are the two searches:
>>
>>a) during the game.
>>
>>Depth  Score     Time       Nodes  PV
>>   1     216        0          35  Rfc1
>>   2     200        0         250  e4 Qxd2 Rxd2
>>   3     202       60        1263  Rfc1 Qh3 e4
>>   4     203      170        9783  e4 Qg7 Qe3 Rab8
>>   5     207      220       12885  e4 Qxd2 Rxd2 Be6 Rc1
>>   6     202      330       20163  e4 Qxd2 Rxd2 Be6 Rc1 Rab8 Rdc2
>>   7     199      550       40726  e4 Qxd2 Rxd2 Be6 Rc1 Rab8 Rdc2 Rb7
>>   8     199     1050       89464  e4 Qxd2 Rxd2 Be6 Rc1 Rab8 Nb1 Rb6 Nc3
>>
>>b) after the game with a fresh hash:
>>
>>Depth  Score     Time       Nodes  PV
>>   1     216        0          32  Rfc1
>>   2     200        0         255  e4 Qxd2 Rxd2
>>   3     202      270        1457  Rfc1 Qh3 e4
>>   4     203      380       10068  e4 Qg7 Qe3 Rab8
>>   5     199      440       14046  e4 Qxd2 Rxd2 Be6 Rc1 Rab8
>>   6>   -160      660       38006  Ne2 fxe2 Qxe2 a4 Rb1 Qh3 e4 axb3 Rxb3
>>   6>   -159      770       46432  b4 axb4 Nxd5 cxd5 Qxb4 Qh3
>>   6>     50      770       47608  h4 gxh3 Kh2
>>   6>    176      820       51711  Kh1 Qh3 Rg1 f5 Rcc1 Rab8
>>   6     176      820       51949  Kh1 Qh3 Rg1 f5 Rcc1 Rab8
>>   7     178     1100       76625  Kh1 Qh3 Rg1 f5 Rcc1 Rab8 Rge1
>>   8     184     2250      191340  Kh1 Qh3 Rg1 f5 Qd1 Rab8 Qf1 Qxf1 Rxf1
>>   9     184     5110      452120  Kh1 Qh3 Rg1 f5 Qd1 Rab8 Qf1 Qxf1 Rxf1
>>  10     176     9060      845582  Kh1 Qh3 Rg1 f5 Rcc1 Rab8 Na4 Rb4 Nc5 Be6
>
>I think you are flailing a bit.
>
>There can be any number of things going on differently in the real game case.
>You can have some problems involving transition from one position to the next,
>for instance you could be leaving things in a bogus state when you terminate a
>search.
>
>Also, when you start the program and immediately start a search, there are a lot
>of memory locations that are zeroed, which might not be zeroed the second time
>you start a search, and perhaps this is causing problems.
>
>A pure hash table problem, which would mean that you are totally off the hook
>and have to do nothing, is probably not what is causing this.
>
>This is an important bug for you to understand.
>
>bruce

Ok, I will go through carefully and see if there is any data not reset before a
new search.

James



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.