In article .com,
raylopez99 wrote:
On Apr 28, 8:45 am, David Richerby
wrote:
raylopez99 wrote:
Goodbye, duffer.

Oh. Well, if you put it like that, goodbye.

Dave.

--
David Richerby Swiss Apple (TM): it's like a tastywww.chiark.greenend.org.uk/~davidr/ fruit but it's made in Switzerland!

Geez, don't be so sensitive, I was only flaming you.

That's OK, I'm only killfiling you. Goodbye.

--
Christopher Mattern

NOTICE
Thank you for noticing this new notice
Your noticing it has been noted
And will be reported to the authorities

On Apr 26, 5:24 pm, raylopez99 wrote:
On Apr 25, 10:26 pm, help bot wrote:

You know, if you took the games of a typical (1300) rated
player and checked them with a dumbed-down-Crafty (1500),
you might get some useful information, but not nearly as
much as hoped for. But when you take the games of the
world champions and check them with a program which is
short of 2800, you get mainly garbage, combined with many
instances where a tactical oversight is correctly pinpointed.

But chess is 99% tactics help bot.

That old adage is very useful for instructing newbies, but
every world chess champion is competent in this area, so
the multitude of positional misjudgments comes to the fore.


You also penalize those players who *deliberately* chose
to play what they knew to be sub-optimal moves, for
whatever reason. I just did this myself at RedHotPawn,
choosing to grab a Knight rather than leap in with another
piece to set up a 95%-certain mating net. Why? Because
while the mating net was around 95% certain, the capture
of the free piece was 100% certain (unless I have lost my
mind)! When I spot another mating net, things should be
simple enough for me to get the 100% certainty I desire,
and having captured yet another piece, this is all but
inevitable, barring my opponent's resignation.

But you risk the chance of letting your opponent escape--remember the
maxim: "always check, since the next move may be mate".

What you are missing is this: the reason hopping another
Knight into the fracas was not a 100%-certain mate was that
it wasn't a check, but a quiet move. OTOH, capturing the
free piece was a 100% certain massive gain. With my own
King safe, there was virtually no risk of "escape". BTW, I
was not able to execute a mate because my opponent quickly
resigned after giving up (in addition to the free piece) the
exchange to slow me down a bit. He had no counter play
and the material deficit was continuing to mount.


Just recently
I did not follow this move and instead of winning a pawn against my PC
I drifted and eventually lost.

That was *you*. You are a drifter, a patzer, while
I am a "star".


Another item which these statistical analyses overlook
is the deliberate gift of, say, a half-point. These have
been known to occur in world championship level play,
and of course the "nice guys" will be penalized for not
being "tough players", despite clinching the match
with their action.

Keep in mind this was not a statistical analysis of the kind Sonas is
famous for, but a different kind. Also over time the "nice guys"
penalty will statistically average out.

No, it won't. As someone once said: there are nice
guys and there are tough players. The "nice guys" tend
to remain "nice", while the tough players tend to go insane,
getting meaner and even more self-obsessed.


In short, what can be learned is who was least prone
to tactical blunders, and apparently, whose style leans
most toward a sizable gap between what the program
sees as the #1 optimal move, and #2 -- something I
think may be termed the sharpness of play. For one
example, I am playing a game at RedHot now where
I had to decide whether to develop my QB "normally"
via ...d6 and then B-moves somewhere, or fianchetto
via ...b6 and B-b7. It was a toss-up, since it makes
no difference whatever to the outcome. I expect a
computer would see both moves as being nearly
equal, weighing them in such a way as to slightly
favor the move which gives the Bishop immediate
control of squares, though this immediacy is quite
irrelevant to the true value of the moves.

Again, over time this will "wash out" or "average out".

No, it won't. Chess programs are written to penalize
certain aspects while rewarding others (such as mobility,
for instance). There is no averaging-out, but rather the
semi-flaw will manifest itself again and again, ad infinitum.
A chess program is a bit like a doorbell: press it and it
makes the same sound *every time*.

Let me give a little example here. In a game against
GM Petrosian, GM Fischer singled out the move Knight
on f3 to d1 as being one which set the champion apart
from other GMs, making quite a fuss. Of course, a
chess program like crippled-Crafty might very well see
this same move -- many plys beforehand -- as a retreat
which temporarily gives up control of the vital central
squares. In other words, a program is crippled by its
inflexibility in terms of depth of search, while a human
is crippled by his inability to "see everything obvious"
all the time.


In general
sharp play is better than just pushing yourself into a passive
position, don't you think?

Sure. Since my goal is to win, I dislike dead positions,
and passive ones tend to drag things out to the point of
boredom. I don't want to win at move 123; I want to win
quickly, or at least as quickly as reasonably possible.


That's what Crafty is looking for--sharp
play. Sharp play = sharp mind bot!

Crippled-Crafty is not capable of accurately assessing
what is the sharpest move in games at that level. For
one thing, the 12-ply cutoff means that in many positions,
the program is not even in the same league as those it
is attempting to assess. However, if in addition to those
12 plys, it adds a bunch more for tactical search
extensions, that would mean it can perhaps do a good
job on just the tactical exchanges. As I understand it,
the stated goal was not to merely judge which world
champions were least prone to gross tactical blunders.
If that had been the stated goal, there would never
have been so much criticism.


I wonder just how much time, and to what depth
the moves were analyzed before scoring them. I
recall that often a player's move may be scored poorly,
but if executed and stepped forward, a program may
change its mind completely about this, suddenly
realizing it had overlooked something.

No, you're talking about "move on opponent's time" feature. The way
the study was done was to analyze each move for a fixed time,

Um, the article I read (by following the links provided here)
stated that the search was cut off at exactly 12 plys. This
is not the same as a fixed-time search at all.


so no "changing of mind",

FYI: in the famous match where world champion
Kasparov lost to Deeper Blue, in one game (at least)
the program sacrificed a pawn for whatever it thought
it saw, but then immediately changed its mind, going
into defensive mode due to the horizon effect. I am
telling you this because Deeper Blue was about a
bazillion times faster than other programs of the time,
and yet it still managed to lose due to a problem
which has plagued computers since the dawn of time.
Crippled-Crafty at 12 plys is hardly immune.


and even if so, each player had the same scoring
applied, so it doesn't really matter (over time).

What you are missing here is that only the games
of the world championships were scrutinized, so for
some, the sample size was quite small. I would
prefer a *large* sample size when making excuses
about how it all evens out in the end.


Besides, have you
noticed that _MOST_ of the time (not always) the best move found by
Fritz or Crafty in the first five seconds is also the best move found
after 60 seconds?

Yes. And this phenomenon is not limited to chess
programs. The same flaw can be found in shallow-
thinking human players, who are unable to improve
on their first guesses no matter how much time they
are given. IMO, the ability to guess well yet also to
improve on one's first guesses is the mark of a good
player. If a computer cannot do this, odds are it is
because it lacks positional understanding.


Because chess is 99% tactics, and often the tactics
are no more than 4 moves deep (most of the time).


You seem to be stuck at the beginner level, where
indeed, chess is 99% tactics, and little else matters.


RL (a 1950 Elo player, so I can speak with some authority).

The world champions have many games which were
decided by tactics, but they also have many where
strategy was the decisive battlefield, and many where
both tactics and strategy played key roles. In view of
this I think it would be wise to at the very least, make
use of the strongest program available, and give it
plenty of time to assess each position. Additionally,
I would like to see as many top-level games as
possible included, including those from tournament
play -- not just world championship matches.

Even if all this were done, it would still be a simple
matter to skewer the idea of determining "the greatest
player of all time" in this manner. I could easily
produce an example where playing a bad move was
not only intentional, but necessary in order to win.
Where a "blunder" is a thousand times more effective
than the "best" move. Where the room is filled with
gossip about a certain player having allowed his
inferior opponent a certain-lock on a draw, but
where they all have to re-figure the pairings when the
actual result is posted.

-- star bot

On Apr 29, 4:03 pm, raylopez99 wrote:

Not true at all. Crafty could easily tell you which programs
far stronger than itself played the most perfect chess.

Wrong.

This is not debatable.

Wrong again.
From your other posts bot you clearly show you are not qualified to

answer. This is over your head.

SO you are wrong.

What are you still doing here, after you already lost this
thread? :D

The idiot statisticians who had fun playing around with
math which sort of related to chess were electrocuted
long ago by their own readers! Those readers used
something called "reason" to pinpoint a few of the many
huge problems with these articles. Reason is something
beyond your grasp, but then, so too is chess. :D

-- help bot

On Apr 28, 12:51 pm, raylopez99 wrote:
On Apr 27, 2:49 pm, "Inconnux" wrote:

Then you conceed my point and indeed the point of Crafty rating chess
players.

OK, now if I used Crafty for analysis of Fritz or Rybka games, it would
certainly not agree and would often call their moves 'errors' even though
they are FAR stronger than crafty.

In particular it will fail to spot moves where short to medium term
material gain is obtained at the expense of a losing long term
positional disadvantage that only shows up well beyond its terminal
node search horizon. The commercial programs with more aggressive
pruning and the largest range of positional heuristics at the moment
have the edge. And even then it is probably safer to use an ensemble
of the strongest programs guided by an independent GM to try and
analyse top players games for "perfection" meaningfully. All of the
engines have blind spots in certain positions.

To properly analyze the world champions
you would need to use a program that is atleast equal in strength to
these champions. Crippled Crafty just doesn't cut it... now if they
used Rybka for analysis I wouldn't have any problem
with the study.

J.Lohner

Not true at all. Crafty could easily tell you which programs far
stronger than itself played the most perfect chess. This is not
debatable.

Not necessarily. It might take Crafty an interminably long time to
spot that a certain capture leads to a situation where a pawn will
promote 30 or more ply into the future. Whereas an engine with more
sophisticated pruning and a heuristic for detecting "pawn can run"
patterns might see the ultimate outcome with a less than 20ply search.
I use this only as an example (I think Crafty is somewhat smarter than
this).

And I am not being rude about Crafty here. It does a lot better at
blocked pawn positions than Fritz8 which takes fully 15 minutes on top
end PC hardware to see into the classic puzzle quoted in Roger
Penroses book the "Emperors New Mind" as the sort of position
computers "will never understand". At the time he wrote the book in
1994 it was inconceivable that a program could search deeply enough or
understand that grabbing the rook and breaking up the protective pawn
barrier would lead to total disaster about 20 moves in the future.
Shredder10 now uses the position as a demo solved in 2s.

For instance, the winning program between two chess
programs playing each other by definition will produce at least one
less error than the losing program--and Crafty could, at some point,
appreciate this.

Apart from the fact that Crafty may not be able to see far enough into
the future to match the equivalent search depth of top commercial
engines or GMs there is another serious fault in your reasoning. The
only thing that you can say for certain when one program beats another
is that the losing side made a mistake that the winning program could
recognise and then exploit to its advantage. Or equivalently the
winning side saw something through selective extensions that the other
did not. Either side could have made many less than ideal moves up to
that point provided that the other was unable to extract any advantage
from it.

The only way you can get around your erroneous statement is to qualify
"properly" in "properly analyze". If you mean that it is better to
have an even stronger chess program than Crafty to better ("properly")
rate the champions, of course you're right and nobody would disagree
with you. But that doesn't mean Crafty's efforts are of no value.

Only that they are likely to be highly misleading in the situation
where the GM and/or a stronger program can see why the most obvious
strong move is not the best principle variation for gaining long term
advantage.

For detecting classic human blunders that can occur in any game any
decent chess engine will do. But if you seek to find the notional
"best" or "strongest" move in a given position you are first going to
have to define exactly what you mean by best or strongest. Some
positions may have several perfectly playable continuations that are
equivalent to within the noise on the evaluation function - even
though the program might still give them slightly different scores.

Playing against a much stronger player the continuation line most
likely to hold a draw has clear merit, whereas playing against a much
weaker player the one leading to a slightly risky quick win may be
perfectly OK.

Perhaps with a 'properly' written program you might have, in a close
tie, a switch between two players say tied for fifth place in the
pantheon of all-time champions

I am not convinced that scoring human GMs by how closely their play
resembles any particular named chess engine has merit. Perhaps ranking
them by percentage blunder rate might be meaningful though (and well
within the capability of any good chess engine). It is surprising how
effective blunder check can be even on GM level games given sufficient
time.

Regards,
Martin Brown

Martin Brown wrote:
I am not convinced that scoring human GMs by how closely their play
resembles any particular named chess engine has merit. Perhaps
ranking them by percentage blunder rate might be meaningful though
(and well within the capability of any good chess engine). It is
surprising how effective blunder check can be even on GM level games
given sufficient time.

What do you mean by `percentage blunder rate'? The proportion of the
time that the GM plays a move that the engine thinks is, say, more
than one pawn worse than the best move? How does that make a
difference?


Dave.

--
David Richerby Frozen Book (TM): it's like a romantic
www.chiark.greenend.org.uk/~davidr/ novel but it's frozen in a block
of ice!

raylopez99 wrote:
David Richerby wrote:
raylopez99 wrote:
Goodbye, duffer.
Oh. Well, if you put it like that, goodbye.
Geez, don't be so sensitive, I was only flaming you.

I wasn't being sensitive. I was realizing that I have better things
to do with my life than read your Usenet posts.


Dave.

--
David Richerby Crystal Perforated Boss (TM): it's
www.chiark.greenend.org.uk/~davidr/ like a middle manager but it's full
of holes and completely transparent!

On Apr 30, 11:52 am, David Richerby
wrote:
Martin Brown wrote:
I am not convinced that scoring human GMs by how closely their play
resembles any particular named chess engine has merit. Perhaps
ranking them by percentage blunder rate might be meaningful though
(and well within the capability of any good chess engine). It is
surprising how effective blunder check can be even on GM level games
given sufficient time.

What do you mean by `percentage blunder rate'? The proportion of the
time that the GM plays a move that the engine thinks is, say, more
than one pawn worse than the best move?

That would probably do as a rough working definition. The search depth
or time might also need to be specified.

If a move is sufficiently far off the mark then the engine is probably
right to fault it. I reckon 100cp ought to be a wide enough window to
avoid too many false positives.

How does that make a
difference?

Unforced tactical errors play their part in the outcome of games. And
these are precisely the sorts of thing that computer chess engines are
very good at spotting. Subtle long term structural games are much
harder for them to score.

Regards,
Martin Brown

Martin Brown wrote:
David Richerby wrote:
What do you mean by `percentage blunder rate'? The proportion of
the time that the GM plays a move that the engine thinks is, say,
more than one pawn worse than the best move?

That would probably do as a rough working definition. The search
depth or time might also need to be specified.

Sure.


How does that make a difference?

Unforced tactical errors play their part in the outcome of games.
And these are precisely the sorts of thing that computer chess
engines are very good at spotting. Subtle long term structural games
are much harder for them to score.

So you're suggesting that ``Player X makes a one-pawn blunder in n% of
games'' is a better measure than ``Player X, on average scores n cp
lower per move.'' That does sound like a reasonable statement, though
I do worry that sacrifices of pawns are relatively common and might
still be mis-evaluated quite often. Kasparov used to sacrifice a pawn
for long-term initiative faster than you can say, ``My computer thinks
that's a pretty dodgy move.'' :-)

Do you have any guess (or, shock!, data) on how often errors occur in
WC games that an engine (given reasonable time) would score down by
say 100cp?


Dave.

--
David Richerby Swiss Old-Fashioned Atom Bomb (TM):
www.chiark.greenend.org.uk/~davidr/ it's like a weapon of mass destruction
but it's perfect for your grandparents
and made in Switzerland!

On Apr 29, 5:39 pm, (Chris Mattern) wrote:
In article .com,



raylopez99 wrote:
On Apr 27, 2:49 pm, "Inconnux" wrote:


To properly analyze the world champions
you would need to use a program that is atleast equal in strength to
these champions. Crippled Crafty just doesn't cut it... now if they
used Rybka for analysis I wouldn't have any problem
with the study.

J.Lohner

Not true at all. Crafty could easily tell you which programs far
stronger than itself played the most perfect chess. This is not
debatable.

Not only is it debatable, it's not true.

No it is true.


For instance, the winning program between two chess
programs playing each other by definition will produce at least one
less error than the losing program--and Crafty could, at some point,
appreciate this.

Er, how? If Crafty is less able than the losing program, how
can it reliably see the error the losing program couldn't?


Easy. The evaluation function of Crafty will indicate that the losing
program, which we've said is much stronger than Crafty, scored, over
the length of the game, worse than the winning program.

To give a simple example: two programs, A and B, both much stronger
than Crafty, play a slugfest game that extends over 100 moves. Play
is evenly matched, and Crafty scores both programs about the same up
to this point. However, at the 101st move, program A sees a winning
10 move combination--that happens to be a mating net-- that is just
outside the 8 move horizon of program B. Program A enters into the
combination and after say the 5th move, Crafty, with a mere five move
chess horizon, also "sees" the winning combination. Of course program
B also has seen this combination wins after the second move but let's
say is programmed with a contempt factor not to resign but to play to
the end. Program A checkmates program B after the 10 move
combination. Crafty will reward Program A and penalize Program B for
this play, even though it is much weaker than either program A or B.


--
Christopher Mattern


Do you really work for Sun? What a disaster that stock has been.
Back to work for you.

RL

David Richerby wrote:

Do you have any guess (or, shock!, data) on how often errors occur in
WC games that an engine (given reasonable time) would score down by
say 100cp?

I will say, that in my own practical experience, running through games.
That in the same, and not unusual positions, that Fritz 8,9,and 10
have evaluated positions over 100cp different than Rybka 2.3.1 And
that different moves have been suggested.

That alone should provide enough of a question as to the results here.
The fact is that we don't know when the engines will be strong enough to
represent the "truth".

I will say that I do not use Crafty for day-to-day analysis so I don't
have an opinion other than that you need to remember in ELO that the
difference between 2500 and 2800 is vast, and the difference between
2800 and ~ 3100 is as vast. It is not 10% better, it is closer to think
of it as TWICE as good. Or more likely to win MOST of the time. It is
a HUGE difference.