What about far plyes precision?
Stefano Gemma
From the series "thinked at night" :-) i have another strange question to
debate. Someone said that there must be a "tactical barrier" (i think it
was Vincent), someone else said that this is not true. I was talking about
"coherence", now i'm talking about "precision"... before that Bob can say
that someone in 33 BC talked about this before me! :-)))
Come back to first ply. We have our beautyfull ordered moves list. Any move
has its "as well as we can compute" score. What is the probability that a
score is the right score? At this ply, the probability is a little low, for
horizons effects do to captures. Ok... come down one ply. Another score,
another story. Now the probability that we have a good score it is a little
bigger... but still not bigger as we want. Come down and down and down...
finally we have the best that we can compute, in the two minutes before the
human player shut down our PC, tyred to wayting for a never coming move ;-)
Now we can suppose to have almost eliminated any horizons effect, maybe
this is true... we have reached ply 30 on a Intel 986 at 1000MHz. No one
can beat us. Ours leaves's scores are almost perfect! True? No! False...
false at all! We can eliminate or minimize horizons effect by doying a 30
ply search (we cannot do it for now, but we will do it in the future) but
we loose for probability.
Suppose to not consider horizons effects. We can rely only on goodness of
our evaluation function. Can we do a perfect evaluation function? I think
no, so our evaluation function is not perfect. Maybe we can have a very
good evaluation. Maybe we can compute the true score of 999/1000 positions
and just one bad value on 1000 positions. What it means? It means that at
ply 1 we have 1/1000 probability (i never remember where to put "y" and "i"
in english words :-) and how to coniugate verbs :-) ) to have computed the
right score. If we do some forward pruning based on the position's score
(or any other position parameters) at intermediate plyes, we still have
1/1000 (or less or more) probability to do the right thing. This is true
for any ply. What happens if we look so far as 30 plyes? I think that the
probability to have a bad values will increase... maybe exponentially. I
have not enough mathematical knowledge to compute this probability, i think
that we should multiply the goodness of our evaluation function at any ply.
Supposing to reach the precision of 999/1000 good score at any ply, maybe
the probability to have a good score at ply n should be:
(999/1000)^n
this would means that at ply 30 we have a probability for a good score of
"only" 97%, instead of 99.9%. The probability will diminish at any next
ply, but not so dramatically.
What happens if we have a bad evaluation function and we rely only on
speed? The probability to have good score is lesser than before, maybe 90%?
Trying with 90% we find this:
(90/100)^30 = 4% !!!!!!!!
so if we do forward pruning and we have only a 90% of good evaluations, it
is better not to go so far... because we will reach a very incoherent value
at the last ply. (Obviously this is not important if we don't use forward
pruning).
We can compute some others precisions:
99%^30 -> 74%
98%^30 -> 54%
97%^30 -> 40%
Maybe i'm wrong, but (if not) this means that forward pruning it is a
suicide if we go too far. Just brute force (even in conjunction with
alfabeta) cannot help us to reach ply 30 and more, so we can say that there
is a tactical barrier.
So we have a big hope to beat Deep Blue on our future Pentium 986 :-)
because long distance plyes could not be reached nor from DB nor for us
without forward pruning. Forward pruning needs a good evaluation function.
Good evaluation function needs good programmers. Good programmers don't
needs IBM's $$$ to be good... just a lot of time and the suggestions of
Bob! :-)
Goodnight from Italy.
Ciao!
Komputer Korner
Stefano Gemma wrote:
>
snipped
> Maybe i'm wrong, but (if not) this means that forward pruning it is a
> suicide if we go too far. Just brute force (even in conjunction with
> alfabeta) cannot help us to reach ply 30 and more, so we can say that there
> is a tactical barrier.
>
> So we have a big hope to beat Deep Blue on our future Pentium 986 :-)
> because long distance plyes could not be reached nor from DB nor for us
> without forward pruning. Forward pruning needs a good evaluation function.
> Good evaluation function needs good programmers. Good programmers don't
> needs IBM's $$$ to be good... just a lot of time and the suggestions of
> Bob! :-)
>
> Goodnight from Italy.
>
> Ciao!
Well, the way that you are defining tactical barrier is a real barrier.
Vincent and Brian Sheppard defined it the exact opposite way. I agree
with your concept of tactical barrier with the one proviso that if
you search full width deep enough the evaluation does not matter,
only the materiel count in the end. The evaluation score is an
indicator to the program where and what to search next. That is why
asymmetric evaluations are playing with fire.
--
Komputer Korner
The komputer that kouldn't keep a password safe from
prying eyes, kouldn't kompute the square root of 36^n,
kouldn't find the real Motive and variation tree in
ChessBase, kouldn't compute the proper time in 2 variation
mode, missed the Hiarcs functionality in Extreme
and also misread the real learning feature of Nimzo.
Tom C. Kerrigan
Komputer Korner (kor...@netcom.ca) wrote:
> you search full width deep enough the evaluation does not matter,
> only the materiel count in the end. The evaluation score is an
Material is nothing. Winning is what matters. To quote somebody I know, "I
lost, but I still have the better position..."
> indicator to the program where and what to search next. That is why
It is? If I go through mental contortions, I can see how this might be
possible, but at no point in my program do I have code that says, "Hey!
Let's use the evaluation function to determine what to search!"
> asymmetric evaluations are playing with fire.
Explain.
Cheers,
Tom
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Tom C. Kerrigan <kerr...@merlin.pn.org> wrote in article
<5dfhr6$1...@merlin.pn.org>...
What ????
Tom, are you crazy, you want us have to wade through a KK explanation ?
Typikal ekample (well, anyroads, how it appears to me):
At20pliesdeep,howmanyplieswillthemostknowledgeablekomputer
program'sknowledgebeworthinextrasearkhpliesInotherwords
willyouletmebuildaDEEEEEEPBluethatwillsearkh30plydeep
WillyourknowledgestillbeenoughtowinOrareyousayingthat
superiorknowledgeisworthonly5plyattopsDon'tforgetwhile
youwillbeaddingknowledge,sowillDEEEEEPBlueIthinkthat
tomakeupfor5ply,youwillneedalltheknowledgeofalltheGM's
Theproblemisthatyourprogramwon'tbeabletoassimilateallthat
knowledgekorrektlyKnowledgekan'tmakeupkompletelythedifferenke
atverydeepsearkhdepthsMynextstatementwillhitbelowthebelt,
buthereitisVinkentonkesaidthatthemoreasymmetrikaprogram
isthehardertimeithasatverydeepsearkhdepthsIdon't
understandwhybutifyouagreewiththatstatementthenwouldn't
yousaythataddingknowledgetoanasymmetrikprogramisliketrying
toteakhadrunktowalkastraightlineIfyoudisagreewith
Vinkent'sstatementthenthatisonemoreVinkentismwehavetoklear
upAnywaytogetbakktotheoriginalargumentEvenifyouareright
aboutknowledgemakingupfor5pliesofextrasearkh,thatdoesn'tmean
that10plyextrawon'tbeanygoodandthattheknowledgewill
preventyourprogramfromfallingintoataktikalholeTaktikslurk
everywhereevenverydeeplySayingthataprogramkanapply
hugeknowledgetomakeupfordeepersearkhingisarguingtheroot
prokessorargumentandIdidn'tknowthatKhrisWhittingtonhad
jumpedshipintotherootprokessorkrowd
--
KomputerKorner
Thekomputerthatkouldn'tkeepapasswordsafefrom
pryingeyes,kouldn'tkomputethesquarerootof36^n,
kouldn'tfindtherealMotiveandvariationtreein
KhessBase,kouldn'tkomputethepropertimein2variation
mode,missedtheHiarksfunktionalityinExtreme
andalsomisreadthereallearningfeatureofNimzo
>
> Cheers,
> Tom
>
Komputer Korner
Tom C. Kerrigan wrote:
>
> Komputer Korner (kor...@netcom.ca) wrote:
>
> > you search full width deep enough the evaluation does not matter,
> > only the materiel count in the end. The evaluation score is an
>
> Material is nothing. Winning is what matters. To quote somebody I know, "I
> lost, but I still have the better position..."
>
> > indicator to the program where and what to search next. That is why
>
> It is? If I go through mental contortions, I can see how this might be
> possible, but at no point in my program do I have code that says, "Hey!
> Let's use the evaluation function to determine what to search!"
>
> > asymmetric evaluations are playing with fire.
>
> Explain.
>
> Tom
I should have qualified the statement of materiel count matters as long
as
there is no immediate checkmate. Long term positional pressure is never
worth more than 4 pawns and even in these positions most are decided by
win of materiel.
I only meant that the evaluation algorithm enables the program to score
the moves so that the highest is looked at first. Vincent once made a
comment that asymmetric evaluations became a larger problem as the
search
gets to very deep levels. I don't know why this would be true but
perhaps
someone will explain the validity of this.
Chris Whittington
--
http://www.demon.co.uk/oxford-soft
Komputer Korner <kor...@netcom.ca> wrote in article
<32FEBE...@netcom.ca>...
> Tom C. Kerrigan wrote:
> >
> > Komputer Korner (kor...@netcom.ca) wrote:
> >
> > > you search full width deep enough the evaluation does not matter,
> > > only the materiel count in the end. The evaluation score is an
> >
> > Material is nothing. Winning is what matters. To quote somebody I know,
"I
> > lost, but I still have the better position..."
> >
> > > indicator to the program where and what to search next. That is why
> >
> > It is? If I go through mental contortions, I can see how this might be
> > possible, but at no point in my program do I have code that says, "Hey!
> > Let's use the evaluation function to determine what to search!"
> >
> > > asymmetric evaluations are playing with fire.
> >
> > Explain.
> >
> > Cheers,
> > Tom
>
> I should have qualified the statement of materiel count matters as long
> as
> there is no immediate checkmate. Long term positional pressure is never
> worth more than 4 pawns
This is the eighteenth kommandment ?
Long term positional pressure < 4 pawns
I'll put that in my program then, because right now it oftens thinks long
term positional pressure > 4 pawns.
Thanks for the tip
Chris Whittington
Vincent Diepeveen
I have a nice position for KK with more than a rook up in material...
...and you may guess who wins in the longterm, after 45 checks or something.
Vincent
>Long term positional pressure < 4 pawns
>I'll put that in my program then, because right now it oftens thinks long
>term positional pressure > 4 pawns.
If Diep transposes to a won ending it automatically adds +20 pawns
to its evaluation.
>Thanks for the tip
>Chris Whittington
>
>
>> and even in these positions most are decided by
>> win of materiel.
>> I only meant that the evaluation algorithm enables the program to score
>> the moves so that the highest is looked at first. Vincent once made a
>> comment that asymmetric evaluations became a larger problem as the
>> search
>> gets to very deep levels. I don't know why this would be true but
>> perhaps
>> someone will explain the validity of this.
>> --
>> Komputer Korner
>> The komputer that kouldn't keep a password safe from
>> prying eyes, kouldn't kompute the square root of 36^n,
>> kouldn't find the real Motive and variation tree in
>> ChessBase, kouldn't compute the proper time in 2 variation
>> mode, missed the Hiarcs functionality in Extreme
>> and also misread the real learning feature of Nimzo.
>>
--
+----------------------------------------------------+
| Vincent Diepeveen email: vdie...@cs.ruu.nl |
| http://www.students.cs.ruu.nl/~vdiepeve/ |
+----------------------------------------------------+
brucemo
Stefano Gemma wrote:
> Supposing to reach the precision of 999/1000 good score at any ply, maybe
> the probability to have a good score at ply n should be:
>
> (999/1000)^n
>
> this would means that at ply 30 we have a probability for a good score of
> "only" 97%, instead of 99.9%. The probability will diminish at any next
> ply, but not so dramatically.
I think this is not an accurate assessment of the situation.
As near as I can tell, what you are saying is the longer a line is, the more
chance is that there's a bug in it.
What is missing is that the lines aren't just long, they're wide, and the deeper
you go, the wider they get.
So a bug very deep in a PV is often no big deal, because there's usually a point
of profitable digression at some earlier point. The deeper you go, the more
chance that such a digression (either an improvement in your predicted
line, or something that will save you if you made a mistake) exists.
Did I read your post correctly?
bruce
Komputer Korner
Vincent Diepeveen wrote:
>
snipped
snipped
>
> >
> >> and even in these positions most are decided by
> >> win of materiel.
> >> I only meant that the evaluation algorithm enables the program to score
> >> the moves so that the highest is looked at first. Vincent once made a
> >> comment that asymmetric evaluations became a larger problem as the
> >> search
> >> gets to very deep levels. I don't know why this would be true but
> >> perhaps
> >> someone will explain the validity of this.
> >> --
>snipped
Well that isn't really a positional compensation question with 45
checks. The position isn't stable with 45 checks. Mating
sequences with 45 checks are not what I meant by positional
compensation. I did a long term study of all the sacs in the
opening that were unclear, and found that 4 points positional
compensation was the max. Of course you will find some weird
positions where a queen might = a knight or bishop but these usually
are not quiescent either.
Tom C. Kerrigan
Chris Whittington (chr...@demon.co.uk) wrote:
> Tom, are you crazy, you want us have to wade through a KK explanation ?
> Typikal ekample (well, anyroads, how it appears to me):
> At20pliesdeep,howmanyplieswillthemostknowledgeablekomputer
> program'sknowledgebeworthinextrasearkhpliesInotherwords
[blah]
Chris... not quite... I think you need to run this thru the fax a few
times to get the desired lack of legibility... :)
Cheers,
Tom
Tom C. Kerrigan
Komputer Korner (kor...@netcom.ca) wrote:
> > > asymmetric evaluations are playing with fire.
> > Explain.
> the moves so that the highest is looked at first. Vincent once made a
> comment that asymmetric evaluations became a larger problem as the
> search
> gets to very deep levels. I don't know why this would be true but
> perhaps
> someone will explain the validity of this.
Ooh, good to know you're propogating second hand speculation from none
other than Vincent...
"Would somebody please explain why I posted this??"
Nice.
Cheers,
Tom
Vincent Diepeveen
>Vincent Diepeveen wrote:
>>
>snipped
>snipped
>>
>
>> >
>> >> and even in these positions most are decided by
>> >> win of materiel.
>> >> I only meant that the evaluation algorithm enables the program to score
>> >> the moves so that the highest is looked at first. Vincent once made a
>> >> comment that asymmetric evaluations became a larger problem as the
>> >> search
>> >> gets to very deep levels. I don't know why this would be true but
>> >> perhaps
>> >> someone will explain the validity of this.
>> >> --
>>snipped
>
>Well that isn't really a positional compensation question with 45
>checks. The position isn't stable with 45 checks. Mating
>sequences with 45 checks are not what I meant by positional
>compensation. I did a long term study of all the sacs in the
>opening that were unclear, and found that 4 points positional
>compensation was the max. Of course you will find some weird
>positions where a queen might = a knight or bishop but these usually
>are not quiescent either.
You still say that no way positional things compensate for 4 pawns,
so there is no position where having a rook up doesn't compensate?
Go play chess instead of just writing email!
Caro-Kann, rook sacrafice, long term. Black is rook up, but still
looses without chances.
something like:
e4,c6 d4,d5 nc3,de ne4,bf5 ng5,bg6 h4,h6 nf3,nd7 h5,bh7 bd3,bd3
Qd3,Qc7 Rh4,e6 Bf4.. look into BCO II
there is a variation here where White plays quickly Rg4 xg7 xf7,
sacraficing a rook. A full rook!
The rook sacrafice is in the book!
Look into the first chessbook one should buy
, ECO B or BCO II, there are lots of rooksacrafices over there.
Queen sacrafices are little harder, but also available.
Even in an openingsbook, written and played by GM's!
No artificial positions. Quiescent positions.
What about a position where one threatens to get a queen in the long term?
For example test position II in the win at chess positions.
Vincent
>--
>Komputer Korner
>The komputer that kouldn't keep a password safe from
>prying eyes, kouldn't kompute the square root of 36^n,
>kouldn't find the real Motive and variation tree in
>ChessBase, kouldn't compute the proper time in 2 variation
>mode, missed the Hiarcs functionality in Extreme
>and also misread the real learning feature of Nimzo.
Komputer Korner
Vincent Diepeveen wrote:
>snipped
>
"Caro-Kann, rook sacrafice, long term. Black is rook up, but still
looses without chances.
something like:
e4,c6 d4,d5 nc3,de ne4,bf5 ng5,bg6 h4,h6 nf3,nd7 h5,bh7 bd3,bd3
Qd3,Qc7 Rh4,e6 Bf4.. look into BCO II
>
there is a variation here where White plays quickly Rg4 xg7 xf7,
sacraficing a rook. A full rook!
The rook sacrafice is in the book!
Look into the first chessbook one should buy
, ECO B or BCO II, there are lots of rooksacrafices over there.
Queen sacrafices are little harder, but also available.
Even in an openingsbook, written and played by GM's!
No artificial positions. Quiescent positions.
What about a position where one threatens to get a queen in the long
term?
For example test position II in the win at chess positions."
: Vincent
Hi Vincent,
I would like to see an unclear positional Queen sacrifice where the
sacrificing side doesn't get any materiel back in return. They just
don't exist and all of the ones where it is for 1 minor piece
disintegrate into either materiel equality or a win for one side
within a tactical horizon of a deep searcher.
As for Rook sacrifices, there are rare positions where it is a Rook
for 1 pawn, but most are 2 or 3 pawns. Your example is not in ECO.
It seems that white is just lost here. The most plausible line I found
based on your analysis is the following:
1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5 5. Ng5 Bg6 6. h4 h6 7. N5f3
Nd7
8. h5 Bh7 9. Bd3 Bxd3 10. Qxd3 Qc7 11. Rh4 e6 12. Bf4 Bd6 13. Bxd6 Qxd6
14. Rg4 Qb4+ 15. c3 Qxb2 16. Rb1 Qxa2 17. Rxg7 b6 18. Rxf7 Kxf7
I could not find this line in either ECO or my Genius 6 million move
opening book. I also checked my 750,000 game database in CA's tree.
It had the game Bykhovsky A.- Grabarczyk 1996 Capelle La Grande which
went 10....e6 11.Bd2 which is not the line you have. There are 5 game
transpositions further up the tree including the stem game
Timman-Ivanovic Niksic 1978 which ended up in a double rook endgame
so no Rook sac here. I don't have BCO 2 , so can you please quote the
game reference or the exact finishing moves?
--
Komputer Korner
The inkompetent komputer.
Wolfgang Kuechle
Komputer Korner wrote:
> Hi Vincent,
> I would like to see an unclear positional Queen sacrifice where the
> sacrificing side doesn't get any materiel back in return. They just
> don't exist and all of the ones where it is for 1 minor piece
> disintegrate into either materiel equality or a win for one side
> within a tactical horizon of a deep searcher.
Wasn't there a spectacular queen sacrifice in a game of Kotov in the
Zurich 1953 tournament where he only got a pawn and a horrible attack
against the king for his queen ?
Regards,
Wolfgang Kuechle
Vincent Diepeveen
>Vincent Diepeveen wrote:
>>snipped
>>
> "Caro-Kann, rook sacrafice, long term. Black is rook up, but still
> looses without chances.
>
> something like:
> e4,c6 d4,d5 nc3,de ne4,bf5 ng5,bg6 h4,h6 nf3,nd7 h5,bh7 bd3,bd3
> Qd3,Qc7 Rh4,e6 Bf4.. look into BCO II
>>
> there is a variation here where White plays quickly Rg4 xg7 xf7,
> sacraficing a rook. A full rook!
>
> The rook sacrafice is in the book!
>
> Look into the first chessbook one should buy
> , ECO B or BCO II, there are lots of rooksacrafices over there.
>
> Queen sacrafices are little harder, but also available.
>
> Even in an openingsbook, written and played by GM's!
>
> No artificial positions. Quiescent positions.
>
> What about a position where one threatens to get a queen in the long
>term?
>
> For example test position II in the win at chess positions."
>: Vincent
>
>Hi Vincent,
>I would like to see an unclear positional Queen sacrifice where the
>sacrificing side doesn't get any materiel back in return. They just
>don't exist and all of the ones where it is for 1 minor piece
>disintegrate into either materiel equality or a win for one side
>within a tactical horizon of a deep searcher.
So you changed your mind. 4 pawns is not the maximum amount of
positional compensation?
That deep searcher definition sucks by the way. If i can search whole game,
just detecting the king capture and repetition and stalemate would be enough.
No evaluation needed. This is however not practical possible.
Why don't you define it:
"There does not exist a sacrafice without getting compensation?"
With that definition i would agree :)
What about Nolot position 11.
r1b3k1/p2p1nP1/2pqr1Rp/1p2p2P/2B1PnQ1/1P6/P1PP4/1K4R1 w - - bm Rxh6;
r = b = - = k =
p - = p = n P -
- = p q r = R p
= p = - p - = P
- = B = P n Q =
= P = - = - = -
P = P P - = - =
= K = - = - R -
white to move
1.Rxh6! just taking 1 pawn. SACRAFICING A ROOK
1 ... nxh6
2.Qg5
2 ... nf7
3.Qd8!! SACRAFICING a FULL QUEEN without capturing anything, just
to remove the knight from f7! Brilliant!
refuting the Komputer Korners false/true algorithm completely!
>As for Rook sacrifices, there are rare positions where it is a Rook
>for 1 pawn, but most are 2 or 3 pawns. Your example is not in ECO.
>It seems that white is just lost here. The most plausible line I found
>based on your analysis is the following:
>1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5 5. Ng5 Bg6 6. h4 h6 7. N5f3
>Nd7
>8. h5 Bh7 9. Bd3 Bxd3 10. Qxd3 Qc7 11. Rh4 e6 12. Bf4 Bd6 13. Bxd6 Qxd6
>14. Rg4 Qb4+ 15. c3 Qxb2 16. Rb1 Qxa2 17. Rxg7 b6 18. Rxf7 Kxf7
Look into your database, and search for a
correspondence game: .. - Diepeveen 1993, master qualification group from NBC.
Against this opponent i also played a game in Dieren 1995 or 1996.
Of course it is not in ECO. ECO is older than BCOII.
Line was under development and also in Informator long after ECO-B was written.
>I could not find this line in either ECO or my Genius 6 million move
>opening book. I also checked my 750,000 game database in CA's tree.
>It had the game Bykhovsky A.- Grabarczyk 1996 Capelle La Grande which
>went 10....e6 11.Bd2 which is not the line you have. There are 5 game
>transpositions further up the tree including the stem game
>Timman-Ivanovic Niksic 1978 which ended up in a double rook endgame
>so no Rook sac here. I don't have BCO 2 , so can you please quote the
>game reference or the exact finishing moves?
It should be in the million base (?)
There are some grandmasters who played and analyzed this Rh4-g4xg7xf7 line,
before it was put in the books.
>Komputer Korner
>The inkompetent komputer.
Vincent
Robert Hyatt
Vincent Diepeveen (vdie...@cs.ruu.nl) wrote:
Nope... none of that is a sacrifice, because all the material comes
back along with the king soon. A sacrifice is typically defined as
giving up material for a positional advantage *only*. Such as the
Rxc3 rook sac (for a knight) in the Sicilian, to name one. Black
gets a lot of counterplay by removing that knight, but he doesn't get
a knight, and at least two pawns for it, so he's really down material.
Another example is to sac a pawn, to disrupt your opponent's pawn structure,
so he has no way to produce a passed pawn on that side, even though he's a
pawn ahead. This *might* not be a sac if you can see that you can make one
and can figure out that you are going to win as a result... but you get
the idea. Win at Chess 141 tosses the queen on the first move, but you
end up with a +2 eval after 8 plies or so. So that isn't a sacrifice by
anybody's standard... it's a *combination*...
: refuting the Komputer Korners false/true algorithm completely!
Ronald de Man
Wolfgang Kuechle <kue...@theochem.uni-stuttgart.de> writes:
>Komputer Korner wrote:
>> Hi Vincent,
>> I would like to see an unclear positional Queen sacrifice where the
>> sacrificing side doesn't get any materiel back in return. They just
>> don't exist and all of the ones where it is for 1 minor piece
>> disintegrate into either materiel equality or a win for one side
>> within a tactical horizon of a deep searcher.
>Wasn't there a spectacular queen sacrifice in a game of Kotov in the
>Zurich 1953 tournament where he only got a pawn and a horrible attack
>against the king for his queen ?
>Regards,
>Wolfgang Kuechle
That was the following position:
1r4nk/1p1qb2p/3p1r2/p1pPp3/2P1Pp2/5P1P/PP1QNBRK/5R2 b
Kotov played Qxh3 and won. According to Bronstein, a human can only play
this move based on intuition. But for computers it's not so difficult
to see that you win back a lot. Unfortunately, my program doesn't really
see a win and thinks it's doing fine after Rh6, so it won't play Qxh3.
But when I make the move Qxh3 and let it search, it almost immediately
sees a draw by repetition. Then at 10 plies the evaluation starts to
drop for white, but not that much (only -0.260 after 13 plies).
Crafty v9.30 finds Qxh3 in less than a second at 4 plies, with evaluation
-0.333 (so apparently crafty doesn't like the position after Rh6 as much
as my program). After 11 plies it is still at -0.210 (for black).
It would be interesting to hear of programs that really see a win of
material for black (in the position before Qxh3).
Ronald de Man
Harald Faber
quoting a mail from deman # wsdw10.win.tue.nl
Hello Ronald,
RdM> From: de...@wsdw10.win.tue.nl (Ronald de Man)
RdM> Subject: Re: What about far plyes precision?
RdM> Organization: Eindhoven University of Technology, The Netherlands
RdM> That was the following position:
RdM>
RdM> 1r4nk/1p1qb2p/3p1r2/p1pPp3/2P1Pp2/5P1P/PP1QNBRK/5R2 b
RdM>
RdM> Kotov played Qxh3 and won. According to Bronstein, a human can only play
RdM> this move based on intuition. But for computers it's not so difficult
RdM> to see that you win back a lot. Unfortunately, my program doesn't really
RdM> see a win and thinks it's doing fine after Rh6, so it won't play Qxh3.
RdM> But when I make the move Qxh3 and let it search, it almost immediately
RdM> sees a draw by repetition. Then at 10 plies the evaluation starts to
RdM> drop for white, but not that much (only -0.260 after 13 plies).
RdM>
RdM> Crafty v9.30 finds Qxh3 in less than a second at 4 plies, with evaluation
RdM> -0.333 (so apparently crafty doesn't like the position after Rh6 as much
RdM> as my program). After 11 plies it is still at -0.210 (for black).
RdM>
RdM> It would be interesting to hear of programs that really see a win of
RdM> material for black (in the position before Qxh3).
RdM> Ronald de Man
No Problem for MChessPro6 (immediately 0.00, after 20sec +0.60 or so, 40MB
hash), Genius5 (immediately 0.0, +0.42 after 2 sec, 32MB hash) and Hiarcs
3 (immediately +046 or so, 47MB hash) and Rebel8 (starting 0.00, after 2
sec +0.23, 28MB hash) on AMD5x86-133 P75
Harald
--
Komputer Korner
Vincent Diepeveen wrote:
>
snipped
>
> It should be in the million base (?)
>
> There are some grandmasters who played and analyzed this Rh4-g4xg7xf7 line,
> before it was put in the books.
>
> >Komputer Korner
> >The inkompetent komputer.
>
> Vincent
> --
> +----------------------------------------------------+
> | Vincent Diepeveen email: vdie...@cs.ruu.nl |
> | http://www.students.cs.ruu.nl/~vdiepeve/ |
> +----------------------------------------------------+
Okay Vincent again you have led us on a wild goose chase. The line you
meant was in B19 and it is not a whole rook sac. White gets 2 pawns for
the rook and in the main line, in the end gets a knight back in trade
for one of the pawns.
[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5 5. Ng3 Bg6 6. h4 h6 7. Nf3 Nd7
8. h5 Bh7 9. Bd3 Bxd3 10. Qxd3 Qc7 11. Rh4 e6 12. Bf4 Qa5+ 13. Bd2 Qb6
14. O-O-O Be7 15. Rg4 Ngf6 16. Rxg7 Bf8 17. Rxf7 Kxf7 *
There are 7 games in this line in my database with the stem game
Hassell-Smith 1984 corr. Great Britain
The main line runs
18. Ne5+ Nxe5 19. dxe5 Nd5 20. Qg6+ Ke7 21. Ne4 Kd7 22. c4 Qd8
23. cxd5 cxd5 and white certainly has most of the materiel back.
It is not much of a sac anymore ( 1 pawn for the exchange). What I
originally said was that within a reasonable horizon or certainly
within the horizon of PC programs, there are no positions in opening
theory and probably none in the middlegame where one side is up a full
5 points with the position being quiescent. In fact there are very few
where the difference is 4 points. So for all practical purposes, if
your program comes up with positional compensation of more than 3 points
then it is overestimating the term. All positions with more than that
amount of materiel difference will degenerate to either mate or gain of
the materiel back or it is simply lost for the inferior materiel side.
In the endgame it is easy to find exceptions to this rule but it should
hold for middlegame and opening positions. Now I know that immediately
there will be a lot of webbers gunning for this latest of komputerisms
and they no doubt will come up with some examples, but I would plead
before you do please check out that the difference is really 5 points
and that it is really quiescent. I would also be interested in positions
with 4 points difference because those are exceedingly rare.
Webbers, it is important that you prove me wrong on this or else this
will lead to another Kommandment and we all don't want another one
do we?
--
Ed Schroder
From: Harald...@p21.f2.n1.z1001.fidonet.org (Harald Faber)
RdM> That was the following position:
RdM>
RdM> 1r4nk/1p1qb2p/3p1r2/p1pPp3/2P1Pp2/5P1P/PP1QNBRK/5R2 b
: No Problem for MChessPro6 (immediately 0.00, after 20sec +0.60 or
: so, 40MB hash), Genius5 (immediately 0.0, +0.42 after 2 sec, 32MB
: hash) and Hiarcs 3 (immediately +046 or so, 47MB hash) and Rebel8
: (starting 0.00, after 2 sec +0.23, 28MB hash) on AMD5x86-133 P75
: Harald
Here is the (winning?) analyze I found in a book.
The analyze was a result of a contest in the "Die Welt" magazine 1953
30.. Qxh3+!! 31.Kxh3 Rh6+ 32.Kg4 Nf6+ 33.Kf5 Ng4!!
I am of course not sure if this is correct.
- Ed Schroder -
brucemo
> RdM> 1r4nk/1p1qb2p/3p1r2/p1pPp3/2P1Pp2/5P1P/PP1QNBRK/5R2 b
>
> : No Problem for MChessPro6 (immediately 0.00, after 20sec +0.60 or
> : so, 40MB hash), Genius5 (immediately 0.0, +0.42 after 2 sec, 32MB
> : hash) and Hiarcs 3 (immediately +046 or so, 47MB hash) and Rebel8
> : (starting 0.00, after 2 sec +0.23, 28MB hash) on AMD5x86-133 P75
> Here is the (winning?) analyze I found in a book.
> The analyze was a result of a contest in the "Die Welt" magazine 1953
>
> 30.. Qxh3+!! 31.Kxh3 Rh6+ 32.Kg4 Nf6+ 33.Kf5 Ng4!!
>
> I am of course not sure if this is correct.
My own got an instant draw with Qxh3+ as well. I let it sit for a minute or two
and this didn't get better.
I played out the line ending with 33. .. Ng4, and after a few minutes it was -2
from white's perspective, so I believe it likely that black is winning.
bruce
Harald Faber
## Nachricht vom : 23.02.97 weitergeleitet
## Ursprung : /FIDO/REC.GAMES.CHESS.COMPUTER
## Ersteller war : rebchess # xs4all.nl@1001:1/0
ES> From: Ed Schroder <rebc...@xs4all.nl>
ES> Subject: Re: What about far plyes precision?
ES> Organization: Schroder BV
ES>
ES> From: Harald...@p21.f2.n1.z1001.fidonet.org (Harald Faber)
RdM>> 1r4nk/1p1qb2p/3p1r2/p1pPp3/2P1Pp2/5P1P/PP1QNBRK/5R2 b
ES> : No Problem for MChessPro6 (immediately 0.00, after 20sec +0.60 or
ES> : so, 40MB hash), Genius5 (immediately 0.0, +0.42 after 2 sec, 32MB
ES> : hash) and Hiarcs 3 (immediately +046 or so, 47MB hash) and Rebel8
ES> : (starting 0.00, after 2 sec +0.23, 28MB hash) on AMD5x86-133 P75
ES> : Harald
ES>
ES> Here is the (winning?) analyze I found in a book.
ES> The analyze was a result of a contest in the "Die Welt" magazine 1953
ES>
ES> 30.. Qxh3+!! 31.Kxh3 Rh6+ 32.Kg4 Nf6+ 33.Kf5 Ng4!!
ES>
ES> I am of course not sure if this is correct.
ES>
ES> - Ed Schroder -
So do you want me to have a look at the times they need for finding this
solution line?
Tschuessikowski
Harald
--