Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: mate threat extension/null move

Author: Michael Henderson

Date: 17:12:26 09/29/04

Go up one level in this thread


On September 29, 2004 at 19:32:27, Stuart Cracraft wrote:

>On September 29, 2004 at 18:52:45, Michael Henderson wrote:
>
>>On September 29, 2004 at 17:29:57, Stuart Cracraft wrote:
>>
>>>On September 29, 2004 at 16:06:36, Uri Blass wrote:
>>>
>>>>On September 29, 2004 at 14:51:16, Dan Honeycutt wrote:
>>>>
>>>>>On September 29, 2004 at 14:16:23, Stuart Cracraft wrote:
>>>>>
>>>>>[snip]
>>>>>>
>>>>>>WAC 141 takes me a shade over 24,000,000 nodes, a fully completed 7 ply search,
>>>>>>to find Qxf4 with all the above improvements, still the same amount of
>>>>>>time as without:
>>>>>>
>>>>>>Alpha=-1332 Beta=-532 Maxdepth=9999999 MaxTime=1410065407
>>>>>> 1/11  g2f1  0.00 -953      945 g2f1 f4d5
>>>>>>                                g2f1 f4d5
>>>>>> 2/12  g2f1  0.01 -953     1653
>>>>>>                                g2f1 f4d5 c1g5
>>>>>> 3/12  g2f1  0.02 -953     5073
>>>>>>                                g2f1 f4d5 c1g5 d5f6
>>>>>> 4/20  g2f1  0.09 -953    19924
>>>>>>                                g2f1 f4d5 b3d5 c6d5 c1c7 d6c7 f1g1
>>>>>> 5/22  g2f1  0.71 -953   170357
>>>>>>                                g2f1 b5b4 b3a4 f4d5 f6g5 d5e7
>>>>>> 6/26  g2f1  2.85 -953   617222
>>>>>>                                g2f1 b5b4 mtmt
>>>>>> 7/32> g2f1 56.60 -553 13644478 g2f1 e8c8 c1b1 f4d5 b1e4 d6b4 f6e5 c7b6
>>>>>>                                g2f1 e8c8 c1b1 f4d5 b1e4 d6b4
>>>>>> 7/34  c1f4 97.33 5113 24161497 c1f4 e8e6 f4g5 d7e7 b3e6 e7e6 h1d1 d6e7 f6e7
>>>>>>                                c1f4 e8e6 f4g5 d7e7 b3e6 e7e6 h1d1
>>>>>> 8/34  c1f4 115.59 5113 27838968
>>>>>>                                c1f4 e8e6 f4g5 d7e7 b3e6 e7e6 h1d1
>>>>>> 9/34> c1f4 197.37 5513 48705672
>>>>>>                                c1f4 e8e6 f4g5 d7e7 b3e6 e7e6 h1d1 mtmt
>>>>>> 9/... still searching
>>>>>>
>>>>>>Stuart
>>>>>
>>>>>Doesn't look like the program has actually found Qxf4.  e8e6 is not the next
>>>>>move.
>>>>>
>>>>>Dan H.
>>>>
>>>>The next move is not important
>>>>
>>>>Every move is losing and the question what is the next move in the program pv is
>>>>dependent on extensions.
>>>>
>>>>It is clear that too many extensions were used because the program needs so many
>>>>nodes to finish depth 7 and the selctive depth is too high(depth 7/34).
>>>>
>>>>I believe that the recapture extension that is used is bad if it helps to solve
>>>>the problem at depth 7 with so many nodes but the only test to be sure is games.
>>>>
>>>>Uri
>>>>Uri
>>>
>>>I considered your last point as well and removed RECAP. The problem cannot
>>>be resolved without it after 10+ minutes. With it, it solves at 95 seconds
>>>at ply 7.
>>>
>>>My main search escape-from-checks and quiescence escape-from-checks are
>>>unlimited.
>>>
>>>I do not search checking moves except in main search or unless a move
>>>out of check in quiescence is a checking move.
>>>
>>>Stuart
>>
>>I think there is something wrong with your search.  -953 is too large of a
>>losing score for g2f1, it should be around -100 to -300.  Can you post the
>>complete variation that leads to this score? with Triangular array you can by
>>not cutting off with a PV entry. h1d1 in the PV looks strange also.
>>
>>Recapture extension should not be helping you here, I believe.  I don't think
>>even captures are hiding anything useful.
>>
>>Michael
>
>I measure in millipawns so -953 is just under a pawn down for Kf1
>which is about the higher end of your range.
>
>In the printout, the first line is the triangular array variation
>and the line immediately under it is the walk-the-pv-in-hash variation.
>So for ply 7:
>
>>>>>> 7/34  c1f4 97.33 5113 24161497 c1f4 e8e6 f4g5 d7e7 b3e6 e7e6 h1d1 d6e7 f6e7
>>>>>>                                c1f4 e8e6 f4g5 d7e7 b3e6 e7e6 h1d1
>
>The c1f4 e8e6 f4g5 d7e7 b3e6 e7e6 h1d1 d6e7 is triangular array pv.
>
>I have heard that said about recaptures but if I don't have it on it
>fails. My recapture extension is:
>
>#ifdef RECAP
>  // Extend if a recapture
>  if (!extended && histptr >= 2 && hist[histptr-1].cap != 0 &&
>        hist[histptr-2].cap != 0 && hist[histptr-1].to == hist[histptr-2].to
>#ifdef NEVER
>        && ABS(hist[histptr-2].pc) == ABS(hist[histptr-1].cap)
>        && ABS(hist[histptr-2].cap) == ABS(hist[histptr-1].cap)
>#endif
>    ) {
>
>        recapext++;
>        extend+=1.00;
>  }
>
>I don't have the NEVER on as that limits the recaptures too severely.
>By just limiting to two captures by any pieces to the same square
>in the last two moves, without regards to the piece or the material,
>that RECAP is what gets me WAC 141 at all. If I don't have it or have
>it in that exact way as above, I don't get WAC 141 at all. With it
>I get WAC 141 with or without my mate-threat at the same time interval
>of 95-97 seconds.
>
>Stuart

I suggest not doing any more testing until you find out why you can't solve WAC
at all with reasonable, good extensions.  I honestly think that is a serious
bug.  You need to strip down the search to something simple and easy to work
with, to find out what is wrong.  For example an engine with a captures only
qsearch, plain alpha beta and mate threat extension only--easy enough to make
and debug :)

Michael



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.