Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Mr. Iacovoni's idea: How to program it.

Author: Robert Hyatt

Date: 11:13:55 08/29/98

Go up one level in this thread


On August 28, 1998 at 21:50:09, Serge Desmarais wrote:

>On August 28, 1998 at 08:20:03, Alessio Iacovoni wrote:
>
>>On August 28, 1998 at 07:32:07, Ilya P. Kozachenko wrote:
>>
>>>
>>>On August 28, 1998 at 03:02:02, Alessio Iacovoni wrote:
>>>
>>>>On August 28, 1998 at 01:18:14, Jeff Anderson wrote:
>>>>
>>>>>
>>>>>>
>>>>>>   Another point is that 2 engines running simultaneously could receive only 50%
>>>>>>of the CPU time each (or so) and so the tactical engine would not reach the same
>>>>>>depth as if it was running alone (at least one full ply of even more shorter)
>>>>>>and so, the evaluation of BOTH programs would be less reliable? Like when you
>>>>>>make one engine play another in Fritz 5, they are both weaker than when running
>>>>>>alone.
>>>>>>
>>>>>>
>>>>>>Serge Desmarais
>>>>>
>>>>>Well as Mr. Iacovoni suggested, this would require two processors.
>>>>>Jeff
>>>>
>>>>
>>>>By the way Jeff.. i'm not a programmer but I just read an article on the thread
>>>>of alt.computers.chess concerning "hsu" and deep blue... apparently his strategy
>>>>seems very similar... all of the moves are passed to a "unit" of some sort
>>>>before beeing played. The article is not very clear on what this "unit" actually
>>>>does. So my idea is not that original.. but at least tha fact that hsu is
>>>>working in the same direction does show that, at least to some extent, my idea
>>>>is not to stupid.
>>>
>>>BTW, how you would decide, which move - produced by tactical or positional
>>>engine - it's better to select ?
>>>It was the point, when discussing of the same idea I posted, finished.
>>
>>Ok... the tactical engine comes up with it's best line which it believes will
>>give it a +0.60. The line (say 10 plies) are passed on to the positional
>>"blunder check" engine which determins if that line would bring to a postional
>>weakness.. (doubled pawns for example).. if it would, then it subtracts the -
>>value of doubled paws (I dont know how much) from the positive value of the
>>tactical engine (+0.60 - 0.60 for example = 0 so the line is not satisfactory).
>>The process is repeated as many times as possible untill a line is found which
>>will reach the highest overall value (tactical and strategical). The
>>coefficients could be weighted in such a way as to give more importance to the
>>tactics (I would give tactical value at least 50% more importance than
>>position... but that has to be decided with trial and error).... In such a way
>>very strong tactical lines would pass even though they receive a negative value
>>from the blunder check module (a checkmate +99.9 for example would pass
>>regardless of the - value attributed by the blunder check module). It seems very
>>simple to me.
>
>
>
>   I would sound better to me that the tartical engine would do a blunder check
>on the positionnal engine's move!
>
>Serge Desmarais



the problem with this is that the "blunder check" can't be done until *after*
the positional engine has produced its move.  Which probably means that the
positional search gets 1/2 the time, the tactical search 1/2 the time.  I'd
bet that a single combined search produces better results almost *every* time.




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.