Author: Eelco de Groot
Date: 09:09:33 07/25/05
Go up one level in this thread
On July 25, 2005 at 03:46:01, Ed Schröder wrote:
>On July 24, 2005 at 21:08:43, Eelco de Groot wrote:
>
>>Hallo Ed,
>
>Hi Eelco,
>
>>Fruit 2.1 presents quite a difficult challenge I think for all the programmers
>>to see if they are able to distill some of the strong characteristics of Fruit
>>2.1 for their own programs. But great that the source is available from Fabien,
>>that should make a difference, in due time!
>>
>>Any luck with Windows 2000 and compiling Pro Deo with GCC (or was it another Gnu
>>compiler) Ed?
>
>I have moved to Windows 2000 now and able to run MVS 2005, it compiles a lot
>faster than my "Digital Mars" compiler.
>
Very good Ed! I hope Microsoft MVSC can tackle the assemblercode that you may
have still in Pro Deo?
It will be very interesting to see what another compiler produces for new code
for you and if the compiles/programs also behave differently, for instance in
different interfaces, or in relation to all the different processors from AMD or
Intel out there.
>
>>Also, I was wondering Ed do you have any idea if this (from
>>http://www.talkchess.com/forums/1/message.html?437706) could be a position where
>>Pro Deo 1.1 is doing a bit too much pruning? Or could it possibly be a history
>>reduction thing?
>>
>>Pro Deo 1.1 default is doing better here than Vulcan {9k9i} but also wants to
>>sac the knight at move 31 at first.
>>
>>7r/1kpr1p2/1p4p1/p6n/3pPP1N/3P1K2/PP1R3P/6R1 w - - 2 1
>>
>>Pro Deo 1.1 default
>>00:35:52.4 0,00 17 2840425533 f5 Nf6 Nxg6 fxg6 Rxg6 Re7 Rxf6 Rh4
>>
>>Vulcan {9k9n4}(={9k9i} with small end1 change) after 29..Rh8,
>>
>>I waited for the 18 ply result, finally it got in but unfortunately 30.f5 did
>>not remain best choice... there was no fail-low high, so I think one of the
>>extensions must have take so much time to get to 18 ply..
>
>I don't remember the Vulcan settings by head but sacs like these can be easily
>related to the following tweaks:
>
>1) You have giving the piece-values a too low setting, if you use:
>
> [Pawn value = 95]
> [Knight value = 95]
>
>then the engine will more easily sac a knight for 3 pawns as in this case.
>
>2) A too high passed pawn evaluation, if you use:
>
> [Passed Pawns = 111]
>
>then the engine will sac a knight/bishop in exchange for the passed passed (as
>in this case) more easily.
>
>3) The use of a too high value of the [Attractiveness] parameter. The
>[Attractiveness] parameter multiplies all positional components with the
>corresponding value. Say you are using a value of 120 then realize that in this
>specific position the 3 passed pawns created by the knight will receive an extra
>bonus of 20% which easily might be too much in this specific case.
>
>Last but not least, realize that all parameters are connected to each other. So
>if you are using:
>
> [Passed Pawns = 111]
> [Attractiveness = 120]
>
>then first the 3 passed pawns are upgraded with 11% and there after with 20%.
>
>Hope this makes some sense!
>
>Best,
>
>Ed
Thanks very much Ed! -Slightly redfaced:-, I had really no idea that
[Attractiveness = ] multiplies all the positional terms, just like lowering the
material values does that. I always just thought Attractiveness was a
combination of some positional terms, and that it also interacted with
Attacking. Now I can sure understand that lowering material to 90, upgrading
Passed Pawns to 110 and also Attractiveness an Attacking at 110 together may be
a bit too much in some cases... As they all do the same, probably the only small
difference is the new term [My Queen = ] when changing material values.
Especially in the endgame I think there is no escaping making some Master
Vulcan-end0, -end1 and -end2 engines with material values going back to 95 from
90, at least if I compare with what Stefan Meyer-Kahlen has done in his Gambit
Shredder and Kamikaze Shredder settings for the material values during the
endgame.
Thanks for the advice Ed!
Eelco
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.