Topic Rybka Support & Discussion / Rybka Discussion / Radical revolutionary idea - two evaluation numbers!
![[gb]](/mwf/v12/flags/gb.png "United Kingdom")
Hi all
I know I never posted here before, but I think I have an interesting idea, which maybe Rybka could pioneeer.
Okay this may sound close to a joke but bear with me.
I think there should be two evaluation numbers for the aspiring OTB players.
The 1st evaluation number is as currently used in engine software. The 2nd one should be based on the probability of human errors. So positions which are "easy to play" even for someone less than 2300, would be given more weighting positively under this second number. It would be a measure perhaps of positional pressure on the opponent and the possibility of the opponent blundering as a result of that pressure. And vice-versa.
There is already an interesting discussion about this concept with reference to something very concrete in a video I did earlier today:
http://www.youtube.com/watch?v=MztUc8zhmfU
Please check out the video and the accompanying comments. I don't think in particular engines would rate Kasparov's b5 early gambit as very good technically. But it was enough to get good pressure against GM Christiansen, and helped Kasparov as a result of winning the Interzonal in 1982 to qualify to play Karpov in 1984.
I actually believe the current evaluation numbers are close to useless when the precision needed to play "only moves" is say several moves in a row. Here is an example - dangerous passed pawns. You might be a knight up, but the possibility of you blundering and allowing the dynamic potential of the passed pawns to get "swindled" is very high.
Best wishes
K
I know I never posted here before, but I think I have an interesting idea, which maybe Rybka could pioneeer.
Okay this may sound close to a joke but bear with me.
I think there should be two evaluation numbers for the aspiring OTB players.
The 1st evaluation number is as currently used in engine software. The 2nd one should be based on the probability of human errors. So positions which are "easy to play" even for someone less than 2300, would be given more weighting positively under this second number. It would be a measure perhaps of positional pressure on the opponent and the possibility of the opponent blundering as a result of that pressure. And vice-versa.
There is already an interesting discussion about this concept with reference to something very concrete in a video I did earlier today:
http://www.youtube.com/watch?v=MztUc8zhmfU
Please check out the video and the accompanying comments. I don't think in particular engines would rate Kasparov's b5 early gambit as very good technically. But it was enough to get good pressure against GM Christiansen, and helped Kasparov as a result of winning the Interzonal in 1982 to qualify to play Karpov in 1984.
I actually believe the current evaluation numbers are close to useless when the precision needed to play "only moves" is say several moves in a row. Here is an example - dangerous passed pawns. You might be a knight up, but the possibility of you blundering and allowing the dynamic potential of the passed pawns to get "swindled" is very high.
Best wishes
K
![[us]](/mwf/v12/flags/us.png "United States")
I think that lots of different evaluation numbers can have value.
There is the evaluation based on search by some engine at some time control. The stronger the engine and the deeper the search, the more reliable the number.
This is the mechanical measure.
There is the evaluation based on actual outcome of games (e.g. 10000 games played, how did white and black fare? I use this formula:
(white_wins - black_wins)* 1.0/(white_wins+black_wins+draws)
Which is negative when it favors black and positive when it favors white.
This measure is "the proof of the pudding" measure.
Then there is the measure of GM analysis such as that found in MCO, NCO, ECO, etc.
This is the expert measure.
There are probably ten times as many good measures as these, and all of them have value.
There is the evaluation based on search by some engine at some time control. The stronger the engine and the deeper the search, the more reliable the number.
This is the mechanical measure.
There is the evaluation based on actual outcome of games (e.g. 10000 games played, how did white and black fare? I use this formula:
(white_wins - black_wins)* 1.0/(white_wins+black_wins+draws)
Which is negative when it favors black and positive when it favors white.
This measure is "the proof of the pudding" measure.
Then there is the measure of GM analysis such as that found in MCO, NCO, ECO, etc.
This is the expert measure.
There are probably ten times as many good measures as these, and all of them have value.
It's interesting. It would take some serious effort to do it properly though.
It sounds to me like this feature already exists: Aquarium trees
![[de]](/mwf/v12/flags/de.png "Germany")
that idea isn't new :) there is also some idea which is similar and proposes a "drawishness" score.
![[ch]](/mwf/v12/flags/ch.png "Switzerland")
It's been discussed before, including in this forum; the problem is how to propagate such “uncertainty” values up the game tree. It's not immediately obvious how to do that, and until somebody comes up with a sound extension of minimax (and alpha-beta!), it's not likely to gain much following. :-)
/* Steinar */
/* Steinar */
Morning all
I did a video yesterday which I think I should mention on this thread - a particularly "tricky" past pawn scenario which for a human is "difficult" to play, but for a computer is clear of course:
http://www.youtube.com/watch?v=dKy5E1YgVG8
Maybe somehow there can be some greater "empathy" for humans playing the position, for in particular the aspiring Over the board players who may want to use engines for opening and middlegame preparation.
Best wishes
Tryfon
I did a video yesterday which I think I should mention on this thread - a particularly "tricky" past pawn scenario which for a human is "difficult" to play, but for a computer is clear of course:
http://www.youtube.com/watch?v=dKy5E1YgVG8
Maybe somehow there can be some greater "empathy" for humans playing the position, for in particular the aspiring Over the board players who may want to use engines for opening and middlegame preparation.
Best wishes
Tryfon
Topic Rybka Support & Discussion / Rybka Discussion / Radical revolutionary idea - two evaluation numbers!
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill