A Piece Value Question
laocmo
Piece values, pawn=1, knight=3, etc. seem to have been developed as
approximate rules of thumb way before computers were invented. Has
modern chess computer theory modified these values any based on real
scientific principles?
pulgao
Not really. It's easier now to find out the exceptions to the general
rules (such as cases where a Bishop may be more valuable than a Rook)
but the general rules of thumb about piece valuations have remained a
constant despite the introduction of computers.
-- Steve Lopez
The Chess Kamikaze Home Page: http://www.geocities.com/ludekdudek/
[winner of the highly-coveted Van Dorp Web Site of the Whatever Award]
The Chess Kamikaze Club: http://clubs.yahoo.com/clubs/chesskamikazes
Mike Thelen
someone did various experiments by computer, using thousands of
Grandmaster games, to determine the "value" of each piece. I don't
remember the exact experiment, but the results were something like:
Pawn = 1.0, Knight = 2.9, Bishop = 3.1 (Bishop pair = +0.5),
Rook = 5.5, Queen = 9.7.... or something like that. I'll check
my magazines and see if I can find the article.
Mike
--
Michael W. Thelen | "I won't repeat myself again,
the...@cs.utah.edu | will not repeat myself again."
http://www.cs.utah.edu/~thelenm/ | -- Toad the Wet Sprocket
Today [Jan 12], laocmo postulated:
Robert Hyatt
> Piece values, pawn=1, knight=3, etc. seem to have been developed as
> approximate rules of thumb way before computers were invented. Has
> modern chess computer theory modified these values any based on real
> scientific principles?
Several have experimented with this. I have tried _many_ combinations
myself, but have always returned to the standard values. They were
derived over a long period of time, and it seems that they are pretty
accurate...
Of course different positions can modify the values a bit...
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Mike S.
Pine.GSO.4.21.000112...@ursa24.cs.utah.edu...
> I recall an article in Chess Life within the past year or so, where
> someone did various experiments by computer, using thousands of
> Grandmaster games, to determine the "value" of each piece. I don't
> remember the exact experiment, but the results were something like:
It's interesting to see the value of +0.5 for the bishop pair here,
derived from GM games. I have done an experiment with 8 computer
programs, which was published by "Computer-Schach & Spiele" No. 5/1996
(with Rebel 7, Fritz 4, M Chess Pro 5 etc.). This was done by letting
the programs play in autoplay mode from a set of 8 modified starting
positions, in which only one side each had the bishop pair. I was
espacially important to let the game run until the pair disappeared, so
that no bishop pair bonus could be included in the evaluations - only
the "real" success of the side which had this advantage.
The average resulting evaluation, out of 64 examples, was +0.46 pawn
units. So it seems, that computer programs handle the bishop pair quite
well.
Regards,
M.Scheidl
Robert Hyatt
> I recall an article in Chess Life within the past year or so, where
> someone did various experiments by computer, using thousands of
> Grandmaster games, to determine the "value" of each piece. I don't
> remember the exact experiment, but the results were something like:
> Rook = 5.5, Queen = 9.7.... or something like that. I'll check
> my magazines and see if I can find the article.
> Mike
That was IM Larry Kaufman. The problem is this: Many programs will
do silly things, like trade a knight/bishop for three pawns. And in
most cases that loses. So inflating the values of all the pieces is
one way to fix that. Or you can do as I do and simply evaluate that
as bad, in the positional evaluation code.
For programs that don't explicitly know that piece vs 3 pawns is a
bad trade, inflating everything but pawns will help. Ditto for a
program that doesn't understand that R+P vs two pieces is a bad
trade for the side giving up two pieces. Inflating the piece values
will fix this.
For some programs, higher piece values will therefore be better. For
others, it would be worse. Which is probably the expected answer when
you think about it. :)
> --
> Michael W. Thelen | "I won't repeat myself again,
> the...@cs.utah.edu | will not repeat myself again."
> http://www.cs.utah.edu/~thelenm/ | -- Toad the Wet Sprocket
> Today [Jan 12], laocmo postulated:
>>
>> Piece values, pawn=1, knight=3, etc. seem to have been developed as
>> approximate rules of thumb way before computers were invented. Has
>> modern chess computer theory modified these values any based on real
>> scientific principles?
>>
>>
Robert Hyatt
> Mike Thelen <the...@ursa24.cs.utah.edu> schrieb in im Newsbeitrag:
> Pine.GSO.4.21.000112...@ursa24.cs.utah.edu...
>> someone did various experiments by computer, using thousands of
>> Grandmaster games, to determine the "value" of each piece. I don't
>> remember the exact experiment, but the results were something like:
> It's interesting to see the value of +0.5 for the bishop pair here,
> derived from GM games. I have done an experiment with 8 computer
> programs, which was published by "Computer-Schach & Spiele" No. 5/1996
> (with Rebel 7, Fritz 4, M Chess Pro 5 etc.). This was done by letting
> the programs play in autoplay mode from a set of 8 modified starting
> positions, in which only one side each had the bishop pair. I was
> espacially important to let the game run until the pair disappeared, so
> that no bishop pair bonus could be included in the evaluations - only
> the "real" success of the side which had this advantage.
> The average resulting evaluation, out of 64 examples, was +0.46 pawn
> units. So it seems, that computer programs handle the bishop pair quite
> well.
.5 is not good. I do a similar bonus in Crafty. About a year ago or so,
Roman Dzhindi.. called and mentioned that this bonus was too high, that he
was able to force positions where Crafty would "bury" its bishop rather than
give it up. I slowly relaxed the score until he finally said "OK.. this
feels right.."
A big bonus can cause horrendous (and unexpected) results. Larry's problem
was that he tested computer vs computer. None of them knew how to do what
Roman was doing to Crafty. But Roman did. And he did it big-time until I
got it right...
Interesting thought... If you test comp vs comp, you miss a lot of problems,
because if one comp doesn't know how to exploit something bad the other program
did, then that 'bad' thing isn't so bad. But don't let a GM jump on that. :)
> Regards,
> M.Scheidl
Mike S.
85iifj$tql$2...@juniper.cis.uab.edu...
> .5 is not good. I do a similar bonus in Crafty. About a year ago or
so,
> Roman Dzhindi.. called and mentioned that this bonus was too high,
that he
> was able to force positions where Crafty would "bury" its bishop
rather than
> give it up. I slowly relaxed the score until he finally said "OK..
this
> feels right.."
> Interesting thought... If you test comp vs comp, you miss a lot of
problems,
> because if one comp doesn't know how to exploit something bad the
other program
> did, then that 'bad' thing isn't so bad. But don't let a GM jump on
that. :)
Yes, I think too that the bonus should be much smaller than 0.5. While a
"success" or evaluation increase of 0.4...0.5, due to the use of the
bishop pair, may be realistic between programs, I agree that between
human masters, the outcome of the experiment would probably have been
less clear. It's not likely that a computer will build a closed pawn
structure to play against the bishop pair. I would even say that he
shouldn't, because if the - human - opponent has the pair or not: The
computer will anyway have it's best chances in an open tactical game I
think.
To continue this logic, maybe it's even better for the computer program
to leave the pair to the opponent (if he's human), so that he will try
to open the game therefore, and things are more likely to get tactical.
Of course, this strategy would be foolish for program vs. program...
Regards,
M.Scheidl
Bruce Moreland
<sean_the...@my-deja.com> wrote in message
news:85jfbh$3fg$1...@nnrp1.deja.com...
> One day an artificial intelligence style program may be able
> to use variable piece values, determined on the fly, so to speak.
> So that when both sides have only a King, Bishop, and pawn, the
> 'puter thinks to itself- hey, this Bishop ain't worth nuttin'
> next to my only hope- my last pawn! :-)
This is done now. Maybe not by everyone, but it's obvious, not very hard,
and worthwhile.
bruce
sean_the...@my-deja.com
> >> I recall an article in Chess Life within the past year or so, where
> >> someone did various experiments by computer, using thousands of
> >> Grandmaster games, to determine the "value" of each piece. I don't
> >> remember the exact experiment, but the results were something like:
> >> Pawn = 1.0, Knight = 2.9, Bishop = 3.1 (Bishop pair = +0.5) (...)
>
> > It's interesting to see the value of +0.5 for the bishop pair here,
> > derived from GM games. I have done an experiment with 8 computer
> > programs, which was published by "Computer-Schach & Spiele" No.
5/1996
> > (with Rebel 7, Fritz 4, M Chess Pro 5 etc.). This was done by
letting
> > the programs play in autoplay mode from a set of 8 modified starting
> > positions, in which only one side each had the bishop pair. I was
> > espacially important to let the game run until the pair
disappeared, so
> > that no bishop pair bonus could be included in the evaluations -
only
> > the "real" success of the side which had this advantage.
> > The average resulting evaluation, out of 64 examples, was +0.46 pawn
> > units. So it seems, that computer programs handle the bishop pair
quite
> > well.
>
so,
> Roman Dzhindi.. called and mentioned that this bonus was too high,
that he
> was able to force positions where Crafty would "bury" its bishop
rather than
> give it up. I slowly relaxed the score until he finally said "OK..
this
> feels right.."
>
problem
> was that he tested computer vs computer. None of them knew how to do
what
> Roman was doing to Crafty. But Roman did. And he did it big-time
until I
> got it right...
>
problems,
> because if one comp doesn't know how to exploit something bad the
other program
> did, then that 'bad' thing isn't so bad. But don't let a GM jump on
that. :)
A long time ago, when Marty Hirsch's Mchess was the top
rated PC program, I noticed that the most obvious difference
between it and others was his extraordinarilly high valuation
for Bishops vs. Knights. While other programs were voluntarilly
trading rather than lose a tempo, Mchess would retreat, and
later when the game opened up a little, punish its Bishopless
opponents mercilessly. He actually went too far in Mchess, but
revised the value downward for the introduction of Mpro- a good
move.
I'm not sure if optimizing a program's piece valuations for
GMs who are consciously trying to exploit the weakness of such
programs is wise. Unless of course, you merely wish to impress
the Gms. :-) Better to optimize the values for best results
overall, against all kinds of opponents, including humans, other
programs, and even GMs (not that they can't be human, mind you).
This would seem to imply tthat the most accurate values are best.
Thus, the simpleton 1,3,5,9 is much too simple. But if it speeds
things up any, there is no sense using very complex estimates
when these assigned numbers are, after all, only estimates.
One day an artificial intelligence style program may be able
to use variable piece values, determined on the fly, so to speak.
So that when both sides have only a King, Bishop, and pawn, the
'puter thinks to itself- hey, this Bishop ain't worth nuttin'
next to my only hope- my last pawn! :-)
Sean the non-Canadian
Sent via Deja.com http://www.deja.com/
Before you buy.
sean_the...@my-deja.com
> > One day an artificial intelligence style program may be able
> > to use variable piece values, determined on the fly, so to speak.
> > So that when both sides have only a King, Bishop, and pawn, the
> > 'puter thinks to itself- hey, this Bishop ain't worth nuttin'
> > next to my only hope- my last pawn! :-)
>
hard,
> and worthwhile.
Really? I had in mind a human-like process which changes the relative
values of the different pieces much the same way as a human would- based
on the attributes of the position. Let me give another example. Say
that one side, White, is up a pawn. A human would evaluate certain
continuations which lead to Rook endings as "good" for Black, while pure
King and pawn endings would be "bad." In essence, the Black Rook would
be a "good" piece, so to speak, because in that type of position its
continued existence helps Black's drawing chances. This would change
immediately if the extra pawn were lost- now the Rooks are "neutral."
A different example: in Garry's match with DeepBlue, he missed a draw
which existed because of two factors- DB's naked King, and Kasparov's
Queen- a powerhouse which often allows such draws to be extracted in
that type of position, all by its lonesome self. DeepBlue failed to
allow for proper King safety, while its human opponent failed to exploit
the one remaining strength of his position, and resigned. When losing,
it is almost always a good idea to keep the Queens on the board,
especially if your own King is reasonably safe. It gives the opponent
something to think about besides winning all your pawns, one by one. :-)
One more example: opposite colored Bishops. You know the deal. Some
computers still don't- even if their rating is in the over-2000 range
due to killer tactics and Pentium speed. In some positions, an extra
Bishop of the wrong color is entirely worthless. How many computers
know this, and show it accurately in their position score? Since Bob
is right here, let's use Crafty as an example- what score does Crafty
assign to K vs. K & B? Or K vs. K & 2 Bishops on the same color? Okay,
maybe he anticipated this, and TOLD Crafty the score in advance, so
let's try K vs. 9 Bishops, all on Black. I score it as 0.0, since the
extra Bishops serve only to cramp and restrict each other. :-) I bet
Crafty (and Fritz) would score it plus-something, unless specifically
told otherwise, in advance. In advance does not equate to on-the-fly.
Thus, we presently have artificial un-intelligence, so to speak.
Bruce Moreland
<sean_the...@my-deja.com> wrote in message
news:85js88$c83$1...@nnrp1.deja.com...
Right. Let's talk about KBB vs K, with bishops on the same color. It is
not reasonable to expect that a computer chess program would understand that
it is impossible to construct a mate under these circumstances. So you add
knowledge for this case, if you wish.
Likewise the case where one side has KB vs K. You add knowledge for that
case, as well.
With more complicated endings, you can cover many of the cases with general
purpose knowledge, but there are specifics that can still benefit from
application of knowledge.
I don't see how this varies much from general-case middlegame positional
knowledge.
But I think I realize what you are getting at. You're talking about
knowledge derived automatically, either on the fly or from past experience.
Some programs do this, I think KnightCap is one, but I don't think this is
going to be an optimal strategy with regard to computer chess for quite a
while.
There is a lot of fairly subtle knowledge in computer chess programs now,
but for the most part they didn't put it there themselves.
bruce
Robert Hyatt
>> > One day an artificial intelligence style program may be able
>> > to use variable piece values, determined on the fly, so to speak.
>> > So that when both sides have only a King, Bishop, and pawn, the
>> > 'puter thinks to itself- hey, this Bishop ain't worth nuttin'
>> > next to my only hope- my last pawn! :-)
>>
>> This is done now. Maybe not by everyone, but it's obvious, not very
> hard,
>> and worthwhile.
> Really? I had in mind a human-like process which changes the relative
> values of the different pieces much the same way as a human would- based
> on the attributes of the position. Let me give another example. Say
> that one side, White, is up a pawn. A human would evaluate certain
> continuations which lead to Rook endings as "good" for Black, while pure
> King and pawn endings would be "bad." In essence, the Black Rook would
> be a "good" piece, so to speak, because in that type of position its
> continued existence helps Black's drawing chances. This would change
> immediately if the extra pawn were lost- now the Rooks are "neutral."
This seems backward. It is far easier to draw most endings with rooks
on the board. A pawn advantage is much easier to cash in without any
pieces at all on the board. Then you make a passed pawn, use that to decoy
the king, and while the king is busy stopping that pawn, the other king has
a feast somewhere on the board.
> A different example: in Garry's match with DeepBlue, he missed a draw
> which existed because of two factors- DB's naked King, and Kasparov's
> Queen- a powerhouse which often allows such draws to be extracted in
> that type of position, all by its lonesome self. DeepBlue failed to
> allow for proper King safety, while its human opponent failed to exploit
> the one remaining strength of his position, and resigned. When losing,
> it is almost always a good idea to keep the Queens on the board,
> especially if your own King is reasonably safe. It gives the opponent
> something to think about besides winning all your pawns, one by one. :-)
This is normal. "When ahead, trade pieces but _not_ pawns. When behind,
trade pawns but not pieces..." This appears in many chess books, and is a
good idea in general, although there are exceptions.
> One more example: opposite colored Bishops. You know the deal. Some
> computers still don't- even if their rating is in the over-2000 range
> due to killer tactics and Pentium speed. In some positions, an extra
> Bishop of the wrong color is entirely worthless. How many computers
> know this, and show it accurately in their position score? Since Bob
> is right here, let's use Crafty as an example- what score does Crafty
> assign to K vs. K & B? Or K vs. K & 2 Bishops on the same color?
draw.
> Okay,
> maybe he anticipated this, and TOLD Crafty the score in advance, so
> let's try K vs. 9 Bishops, all on Black. I score it as 0.0, since the
> extra Bishops serve only to cramp and restrict each other. :-) I bet
> Crafty (and Fritz) would score it plus-something, unless specifically
> told otherwise, in advance. In advance does not equate to on-the-fly.
> Thus, we presently have artificial un-intelligence, so to speak.
I score it as totally stupid by the side that made all those bishops. :)
For normal positions, it definitely knows which are drawn, and which or
not. It doesn't include special cases that would normally be considered
stupidity on the part of the opponent. :) IE Crafty will _never_ promote
to two same-color bishops. If the opponent does, ho-humm.. :)
> Sean the non-Canadian
> Sent via Deja.com http://www.deja.com/
> Before you buy.
--
michael adams
>
> > > One day an artificial intelligence style program may be able
> > > to use variable piece values, determined on the fly, so to speak.
> > > So that when both sides have only a King, Bishop, and pawn, the
> > > 'puter thinks to itself- hey, this Bishop ain't worth nuttin'
> > > next to my only hope- my last pawn! :-)
> >
> > This is done now. Maybe not by everyone, but it's obvious, not very
> hard,
> > and worthwhile.
>
> Really? I had in mind a human-like process which changes the relative
> values of the different pieces much the same way as a human would- based
> on the attributes of the position. Let me give another example. Say
> that one side, White, is up a pawn. A human would evaluate certain
> continuations which lead to Rook endings as "good" for Black, while pure
> King and pawn endings would be "bad." In essence, the Black Rook would
> be a "good" piece, so to speak, because in that type of position its
> continued existence helps Black's drawing chances. This would change
> immediately if the extra pawn were lost- now the Rooks are "neutral."
>
> which existed because of two factors- DB's naked King, and Kasparov's
> Queen- a powerhouse which often allows such draws to be extracted in
> that type of position, all by its lonesome self. DeepBlue failed to
> allow for proper King safety, while its human opponent failed to exploit
> the one remaining strength of his position, and resigned. When losing,
> it is almost always a good idea to keep the Queens on the board,
> especially if your own King is reasonably safe. It gives the opponent
> something to think about besides winning all your pawns, one by one. :-)
>
> computers still don't- even if their rating is in the over-2000 range
> due to killer tactics and Pentium speed. In some positions, an extra
> Bishop of the wrong color is entirely worthless. How many computers
> know this, and show it accurately in their position score? Since Bob
> is right here, let's use Crafty as an example- what score does Crafty
> maybe he anticipated this, and TOLD Crafty the score in advance, so
> let's try K vs. 9 Bishops, all on Black. I score it as 0.0, since the
> extra Bishops serve only to cramp and restrict each other. :-) I bet
> Crafty (and Fritz) would score it plus-something, unless specifically
> told otherwise, in advance. In advance does not equate to on-the-fly.
> Thus, we presently have artificial un-intelligence, so to speak.
>
threaten oppression the prescribed play is kill the oppressive
muvver,sometimes this is not so easy,so as the good chap that I know you
are,simply resign this position,you will be able to come back,and make
sure of the born opprressor. 'gards Mick.
michael adams
> to use variable piece values, determined on the fly, so to speak.
> So that when both sides have only a King, Bishop, and pawn, the
> 'puter thinks to itself- hey, this Bishop ain't worth nuttin'
> next to my only hope- my last pawn! :-)
>
messes the average Canadian to high heaven. 'gards Nick.
sean_the...@my-deja.com
> > Really? I had in mind a human-like process which changes the
relative
> > values of the different pieces much the same way as a human would-
based
> > on the attributes of the position. Let me give another example.
Say
> > that one side, White, is up a pawn. A human would evaluate certain
> > continuations which lead to Rook endings as "good" for Black, while
pure
> > King and pawn endings would be "bad." In essence, the Black Rook
would
> > be a "good" piece, so to speak, because in that type of position its
> > continued existence helps Black's drawing chances. This would
change
> > immediately if the extra pawn were lost- now the Rooks
are "neutral."
> This seems backward. It is far easier to draw most endings with rooks
> on the board. A pawn advantage is much easier to cash in without any
> pieces at all on the board. Then you make a passed pawn, use that to
decoy
> the king, and while the king is busy stopping that pawn, the other
king has
> a feast somewhere on the board.
Okay, I wasn't crystal clear. What I meant was that a human
would automatically evaluate Rook endings as good for Black,
when COMPARED TO pure pawn endings, because of the improvement
in drawing chances when behind by a pawn in Rook endings.
> > A different example: in Garry's match with DeepBlue, he missed a
draw
> > which existed because of two factors- DB's naked King, and
Kasparov's
> > Queen- a powerhouse which often allows such draws to be extracted in
> > that type of position, all by its lonesome self. DeepBlue failed to
> > allow for proper King safety, while its human opponent failed to
exploit
> > the one remaining strength of his position, and resigned. When
losing,
> > it is almost always a good idea to keep the Queens on the board,
> > especially if your own King is reasonably safe. It gives the
opponent
> > something to think about besides winning all your pawns, one by
one. :-)
> This is normal. "When ahead, trade pieces but _not_ pawns. When
behind,
> trade pawns but not pieces..." This appears in many chess books, and
is a
> good idea in general, although there are exceptions.
Yes, but I specifically emphasised the Queen as the most drawish
piece (for humans anyway) in the middlegame. In the endgame, it
could be the Rook. The reason is the rather large number of
possibilities for a one-piece perpetual check or forced repetition.
The Queen is unique in that it can suddenly leap in, all by itself,
and secure the draw. If only Garry had known about this... :-)
> > One more example: opposite colored Bishops. You know the deal.
Some
> > computers still don't- even if their rating is in the over-2000
range
> > due to killer tactics and Pentium speed. In some positions, an
extra
> > Bishop of the wrong color is entirely worthless. How many computers
> > know this, and show it accurately in their position score? Since
Bob
> > is right here, let's use Crafty as an example- what score does
Crafty
> > assign to K vs. K & B? Or K vs. K & 2 Bishops on the same color?
> draw.
Wow. I thought computers used numbers to represent the
score of positions. Your Crafty uses "win," "draw," and
"loss" instead?!! I'm not trying to pick on Crafty- I
would assume that all the others do the same exact thing.
I can check this in Fritz, or maybe with the old Crafty
engine in ChessbaseLight, but not the latest and greatest
Crafty.
> > Okay,
> > maybe he anticipated this, and TOLD Crafty the score in advance, so
> > let's try K vs. 9 Bishops, all on Black. I score it as 0.0, since
the
> > extra Bishops serve only to cramp and restrict each other. :-) I
bet
> > Crafty (and Fritz) would score it plus-something, unless
specifically
> > told otherwise, in advance. In advance does not equate to on-the-
fly.
> > Thus, we presently have artificial un-intelligence, so to speak.
> I score it as totally stupid by the side that made all those
bishops. :)
Hee hee. In fact, I wasn't thinking very clearly when I
thought that example up- it would be very difficult to make
that many underpromotions all on Black squares- four of the
pawns would have to have switched files so they, too, could
become Black Bishops. It might require more than a few
power surges and sunflares for Crafty to play that badly. :-)
> For normal positions, it definitely knows which are drawn, and which
[are]
> not. It doesn't include special cases that would normally be
considered
> stupidity on the part of the opponent. :) IE Crafty will _never_
promote
> to two same-color bishops. If the opponent does, ho-humm.. :)
Just as I said then, either you specifically TOLD Crafty what
to think in advance, or it wrongly assesses this type of position.
Granted that this would have no real effect on its results, since it
only makes the bad evaluation against idiot opponents who have
all their Bishops on one color. In fact, if you told Crafty what
to think in advance, then it will never be wrong, right? :-)
But this is mere parrotting (Bob: "All Rook and pawn ending are drawn."
Crafty: "Bwaawk- All Rook and pawn endings are drawn.") or mindless
regurgitation, if you prefer, not intelligence. For that, Crafty (or
Fritz or ChessmasterBazillion) would have to go BEYOND what it is
specifically told, ala HAL in 2001, a Space Odyssey.
Garry, to Crafty: I'm going to shut you down!
Crafty, to Garry: I don't see how you can do that,
Garry, without your Queen. Mate
in seventy-three, per my endgame
tablebase. Care to play again?
Robert Hyatt
> Wow. I thought computers used numbers to represent the
> score of positions. Your Crafty uses "win," "draw," and
> "loss" instead?!! I'm not trying to pick on Crafty- I
> would assume that all the others do the same exact thing.
> I can check this in Fritz, or maybe with the old Crafty
> engine in ChessbaseLight, but not the latest and greatest
> Crafty.
It is a number. But the number varies. IE if crafty is rated lower
than you, draw could be +.20, if your ratings are similar, it would use
0.00 for draw, and if it is rated higher than you, it would use some
negative number between -.20 and -.50 depending on the rating difference.
That is a definition of programming... "telling crafty what to think
in advance." :)
> Garry, to Crafty: I'm going to shut you down!
> Crafty, to Garry: I don't see how you can do that,
> Garry, without your Queen. Mate
> in seventy-three, per my endgame
> tablebase. Care to play again?
> Sean the non-Canadian
> Sent via Deja.com http://www.deja.com/
> Before you buy.
--
Vincent Diepeveen
<the...@ursa24.cs.utah.edu> wrote:
>I recall an article in Chess Life within the past year or so, where
>someone did various experiments by computer, using thousands of
>Grandmaster games, to determine the "value" of each piece. I don't
>remember the exact experiment, but the results were something like:
>Pawn = 1.0, Knight = 2.9, Bishop = 3.1 (Bishop pair = +0.5),
>Rook = 5.5, Queen = 9.7.... or something like that. I'll check
>my magazines and see if I can find the article.
This is an incredible conclusion because queen is in 90% of the times
outperforming a rook, where here even a queen+passed pawn
is not worth 2 rooks.
of course in GM games you only get exceptions on the board.
in the average game a queen is a lot better than this.
Vincent
> Mike
sean_the...@my-deja.com
> > Just as I said then, either you specifically TOLD Crafty what
> > to think in advance, or it wrongly assesses this type of position.
> > Granted that this would have no real effect on its results, since it
> > only makes the bad evaluation against idiot opponents who have
> > all their Bishops on one color. In fact, if you told Crafty what
> > to think in advance, then it will never be wrong, right? :-)
> > But this is mere parrotting (Bob: "All Rook and pawn ending are
drawn."
> > Crafty: "Bwaawk- All Rook and pawn endings are drawn.") or mindless
> > regurgitation, if you prefer, not intelligence. For that, Crafty
(or
> > Fritz or ChessmasterBazillion) would have to go BEYOND what it is
> > specifically told, ala HAL in 2001, a Space Odyssey.
>
> That is a definition of programming... "telling crafty what to think
> in advance." :)
Not quite. You must tell a computer HOW to "think," or
evaluate positions, etc., in advance. But you needn't tell
it EXACTLY how to score K vs. K & x-Bishops in advance. By
taking that approach, you are limiting Crafty to what you
have specifically told it in advance. This means that you
are in a never-ending battle to keep up with- or stay ahead
of- the Joneses, or in this case, the Fritzes.
HAL never would have become intelligent if he had been so
limited. :-) Do you maintain that he was taught IN ADVANCE
how to read lips, so he could defend himself against those
evil, conspiring humans? I think not. :-)
I recently downloaded a couple of files containing brief
discussions of artificial intelligence and the programming
techniques presently employed to implement them. What they
said was in direct contradiction to some of what you have
often posted here to rgc. Perhaps you are not quite the
expert you think you are- except of course, when it comes
to chess programming as it has been done and still is at
present. Anyone interested in the basics of these ideas
can get some descriptive files from (if I remember) zifdavis.
One of the basic types of artificial intelligence programs
souded as if it were tailor-made for chess. Of course, they
have a very long row to hoe before an a.i. chess program will
ever beat Crafty, but that is all the more reason to start
as soon as possible! :-) While unintelligent programs
follow a linear curve of improvement based upon hardware
speed improvements, an a.i. program might start out as truly
pathetic, but gradually improve itself until one day, it
surpasses the relatively dumb speed demons. Food for thought.
Or deep thought. Hey- that would be a good name for.... no!
Robert Hyatt
> On Wed, 12 Jan 2000 08:47:29 -0700, Mike Thelen
> <the...@ursa24.cs.utah.edu> wrote:
>>I recall an article in Chess Life within the past year or so, where
>>someone did various experiments by computer, using thousands of
>>Grandmaster games, to determine the "value" of each piece. I don't
>>remember the exact experiment, but the results were something like:
>>Pawn = 1.0, Knight = 2.9, Bishop = 3.1 (Bishop pair = +0.5),
>>Rook = 5.5, Queen = 9.7.... or something like that. I'll check
>>my magazines and see if I can find the article.
> This is an incredible conclusion because queen is in 90% of the times
> outperforming a rook, where here even a queen+passed pawn
> is not worth 2 rooks.
> of course in GM games you only get exceptions on the board.
> in the average game a queen is a lot better than this.
> Vincent
what am I overlooking? he said queen=9.7, which is is right at 1 pawn
less than two rooks in the above scheme (2*5.5 = 11, Q+P = 9.7+1=10.7,
which is very close to 'normal...