Author: Albert Silver
Date: 09:43:04 08/04/04
Go up one level in this thread
On August 04, 2004 at 11:50:33, Karl-Johan Olsen wrote:
>On August 04, 2004 at 11:25:17, Albert Silver wrote:
>
>>On August 04, 2004 at 11:04:36, Karl-Johan Olsen wrote:
>>
>>>Do such program exits ?
>>>I was thinking of making one over a decade ago, but unfortunately time...
>>>The idea was human tree pruning to an chessprograms tree, thereby cutting
>>>opening-, middle- and endgame together and marking nodes win/draw/lost.
>>>Have anybody seen such or even made one ?
>>
>>If I understand corrcetly, you want to see the chess trees the program is
>>thinking, and then manually mark them? This would be an impossible task IMO as
>>the number of lines and moves analyzed is going to be too large for a human to
>>analyze in a game. The best option, provided I understood what you mean, would
>>be to use the learning functions of the top programs and play them in Advanced
>>Chess/Centaur mode (Centaur being how it is called in Playchess):
>>
>>Play a move in the analysis, watch the results. Try a few moves ahead, and see
>>the results. Then go back, and see if the engine changes it's opinion or
>>evaluation thanks to the hash of the forward analysis. As a rule, short
>>analysis, but using intelligent forward and backward analysis, is the best way
>>to play this form of chess, as opposed to letting the engine think a long time
>>on a single move.
>>
>>If you meant something else, please explain.
>>
>> Albert
>
>You have the idea right ...
>This engine is only supposed to calculate as to begin with what a normal engine
>do 'today' where the 'tree' is human edited (some would call it a book), but the
>engine will to start with have no book.
>This is where the 'human tree' edition is human tree edited...
>The other way, seen from endgame view, is that it should be able to calculate
>back (some will call it retrograde analysis), you will set methods on some
>positions with a primary plan, this would help on the modifying the engines
>horizon effect...
>With time it will become better.
>Yes you are right, to start with the 'tree is empty' and will be time consuming,
>but what is not ?
>
>Kaali Ujuaat
I think something is not quite understood, either by you or by me. If you want
to see the entire tree of analysis that a program is building while examining a
position and then edit this by hand, it is an impossible task. Today's programs
analyze millions of positions in a couple of minutes, and order these into a
tree. To edit this by hand would be no help.
You can get the same benefits of improving its vision beyond the horizon effect,
by manually entering the moves in infinite mode and letting it analyze and see
the results. When you go back in the variation, it will remember what it learned
from the forward analsyis, and modify its evaluations. For example, suppose the
program is ignoring a sacrifice that you think is strong: 1.Nxf7! Because the
program can't see deeply enough, it thinks the move is only 5th best. You enter
the first moves of the line, and suddenly its evaluation goes up fast. You now
go back to the first position before the sacrifice. It will now have learned the
consequences and choose the sacrifice as best move. This can be done by the top
professional engines and even some of the freeware, such as Crafty, and possibly
Pro Deo.
Albert
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.