On May 3, 2:31 am, help bot wrote:


-- help bot


I realized there are several "bot" handles floating in this NG, and
you "help bot", are the ignorant one.

Go back to pushing wood you Class C patzer.

RL

On May 3, 5:43 am, raylopez99 wrote:

I realized there are several "bot" handles floating in this NG, and
you "help bot", are the ignorant one.

Having posted here for some time, I have yet to
see any other "bot" than myself.


Go back to pushing wood you Class C patzer.

Speaking of patzers, your quite longwinded
blatherings about hypothetical this and that are
reminiscent of a complete patzer describing one
of his many losses. Invariably, it makes "sense"
only from the patzer's own unique perspective.

I prefer to focus on reality, on things like the
fact that in one published game I saw, the only
chess player (including both human GMs and
computers) which correctly evaluated a certain
position was Rybka, while even GM Kasparov
had mis-evaluated the key line. Things like
this lead me to believe that no crippled-Crafty
type program is going to have the wherewithal
to *correctly* rank the moves so it can then
try and rank those who played them. As they
say in the computer field, garbage in, garbage
out. But in this case it is not the moves which
amount to garbage, only the evaluations of them,
and hence, their respective rankings.

To put that game eval. into better perspective
(something you obviously are clueless about),
let me add that in another of the published
games which was supposed to show just how
much better Rybka was than its competition,
I found that, IMO, one of the games it managed
to win was on account of the opponent (a
computer) not advancing its pawns when it not
only should have, but indeed had to. In effect,
my take is that the other computer (which was
no doubt superior to crippled-Crafty) completely
bungled a good position, so Rybka was lucky
to win. In sum, even Rybka has a long way to
go before it can *accurately* rank the moves
by which we may then try and judge the world
champions relative to one another. The fact
that this game was selected for publication by
one of the writers indicates just how easily
one can be deceived when bias leaps in.

Nevertheless, so long as the chess program
is superior overall to those it is attempting to
rank, we should be able to get a general idea
as to what is going on. It's the same with
humans. A rank beginner like yourself can
still give us some clue as to the relative merits
of two even weaker players, provided we can
find any.

As for the utility of a crippled-Crafty to pinpoint
outright blunders, here we do not disagree. As
long as there are extensions for checks and
captures, 12 plys may well be sufficient for the
job at hand. But this in no way equates to any
measure of "greatness", as was the stated goal.

Greatness, as you would know if you were a
(GetClub) star like myself, eclipses the crudity
of tactics, or mundane blunderless play. It
entails a certain degree of creativity, not just
dry, flavorless technique. It is a concept well
beyond the grasp of mere patzers like you, and
of course a crippled-Crafty program cannot even
begin to sense its delicate aroma. True
greatness is best judged by those who can
understand all the subtleties of the world
champions' games, not by mere duffers like
you.

In sum, the authors of those articles I read
simply bit off more than they could chew;
they set out to achieve a vast goal but forgot
to bring along adequate tools; they set out for
parts unexplored but forgot to bring an
intellectual compass for guidance; they
became lost in a morass of ignorance and
self-delusion, much like yourself. Worst of
all, they donned the pretense of statistical
analysis, while tripping at the very first hurdle
(i.e. sample size).

-- help bot

"raylopez99" wrote in message
ups.com...

Suppose that a version of Crafty favors "defensive" players like Capa
and Kramnik, while penalizing "attacking" players like Tal and
Fischer. Call this version of Crafty "Defensive
---
Interesting post, and the para above identifies the flaw in comparisons of
player-to-player by engine analysis- that if you are evaluating players,
then even Tal himself said that anybody at all could find his own flaws
after the game, or the next day, or even next week. But he chose 'em because
very few people could find them in real time OTB.

Given enough time then Stenitz will likely always rank higher than Tal by
analytical method, but OTB, super-solid Steintitz wouldn't know which way
Tal hit him! The flaw is that this analysis is okay for games, ie picking
some best theoretical line by objective and uniform measure against all
other lines - but chess playing is not a theoretical activity - its a
real-time performance.

---
That's the way I see it, and until we get more research any rebuttal
to the contrary will simply be speculation, since the data is just not
there.

We rehearsed this conversation before - but there is no data of GM play
against raw chess engines. The engines are all optimised for winning, and
not for anything useful, like learning - either about its on evaluation
matrix of chess evaluation or even how people evaluate play. [because with
book+table bases=off, it wins less]

I understand the commercial need to do that, but don't understand any
academic reason to choose emulation paradigms over [exigesic] real-time
engine evaluation. I always thought Crafty in particular would be such a
base model, since it is born out of a university system, widely distributed
and adapted, and lots of people might have had a go at it. But I think
Crafty got caught up in its own early success as W CH, and continued to go
for 'win', rather than for 'learn'.

Phil Innes


RL

In article . com,
help bot wrote:
[...] But the main objection was, of course, that
since Rybka (and Hiarcs, etc.) is available, why
mess around with something vastly inferior, unless
ranking mere patzers?

Um, because Crafty is not only available, but available
in source-code form to anyone, so that anyone can instrument it,
piggle with it, and generally use it as a tool to investigate
things? If you want something reproducible, then, short of
help from the commercial companies, you have to do a fair amount
of tweaking.

Also not entirely convinced by arguments about Crafty
only being able to rank mere patzers. I'm well short of super-GM
status myself, but I think I know enough about chess to be able
to judge that [eg] Kramnik is a better player than the typical
2650-ish GM. For that matter, there are also players around who
are *much* stronger analysts [esp in specialised areas, such as
endings] than their OTB rating would suggest -- eg because of
physical/mental limitations that affect them in tournaments, or
because they can't handle the clock -- and such players [eg, top
correspondence players] may be well equipped to rank others of
much higher nominal rating.

One more criticism: in looking over a game at
one Web site, I noticed that the depth of search
DURING PLAY achieved by GM Kramnik's
opponent was around 18 plys. Now why on
earth would anyone try to rate the play of the
world champions by cutting of crippled-Crafty's
search at only 12 plys? I mean, get a REAL
computer, and a clue!

The depth achieved during play was presumably in real
time at, say, 5 hours per game? So a real-time 24-game match
occupied 5 solid days of chess, and so of computer time even
for a dedicated super-computer. Simply rating every WC game
at that speed is several months; or years for the normal PCs
that we have on our desks. The alternatives to crippled-Crafty
rating every WC game are basically to have *no* results, 'cos
it takes too long, or to have an utterly superficial analysis
of *all* the games of Tal/Kasparov/....

Once you set parameters such that you want to investigate
a reasonable corpus of games [and WC matches seems quite sensible]
using a few weeks of time on a PC, the rest somewhat falls into
place. Eg, if you want to analyse 1000 games and are willing to
wait one month for the results, then you have perforce to analyse
30+ games/day, or 48 minutes/game, or around 30s/move, on whatever
dedicated machine is available to you. Whether the resulting
investigation is worthwhile is another matter.

--
Andy Walker, School of MathSci., Univ. of Nott'm, UK.

On May 3, 1:13 am, help bot wrote:
On May 1, 5:33 am, Martin Brown
wrote:

That its shape broadly correlates with the rms error graph of the
players lends credence to the possibility that Crafty might have been
adequate for the task.

Q: How would you like it if we were to assign a player weaker
than you to "rate" your games, to assess your play and compare
it to that of your peers? I suspect your gut reaction would be to
object on the basis that a weaker player is not qualified for the
job, and indeed, it would be better to select the strongest player
available for this task.

I wouldn't like it at all. But in this case I think we are both agreed
that for blunder checking almost any decent chess engine will do
provided that you leave it to think for long enough. The difficulty is
in situations where there are small errors made by players and engines
which grow exponentially when played against a strong opponent.

The problem is that Crafty might not be able to play like a true GM,
but it might still be able to analyse games meaningfully after the
event because using the working back from the end trick it will have
its cache preloaded with an already very strong principle variation
from the human game line. I agree that a stronger engine would be
ideal.

And to be fair to the authors they did say that
others with access to the internals of stronger engines should repeat
their tests to see how they compare.

And, to be fair, it may well be determined that crippled-
Crafty wasn't quite up to the task of ranking moves in
perfect order.

I am not even sure there is a unique perfect ordering. I reckon it may
depend on the opponent!

And I agree. I don't think Crafty is anywhere near the level to score
GM and super GM level moves reliably if given a position and asked to
think about it for even a very long time. I do however think that it
might still be able to annotate a game and find important weaknesses
in moves played despite this limitation.

I expect is is up to the simpler task of
spotting tactical blunders, provided the 12 plys cutoff
did not account for tactical search extensions for
checks and captures.

Agreed. Where did this 12 ply number come from? I can't spot it in the
summary article - and I cannot find a full copy of the entire paper.
If correct it seems way too shallow to me (even for the much stronger
engines).

I would have liked to see the rms error graph with blunders excluded.
That might have shed some more light.

According to what you say below, this would make
very little difference as the blunder rate is almost
infinitely small.

Roughly 4% of all games if my back of the envelope calculation is
right.

Capablanca maintained a blunder rate of 0.01% (1 blunder in every
10000 moves)

I don't believe that. If this were really true, then I
might go over all of his games and come up empty-
handed, having not seen enough moves to find the
one-in-ten thousand. In fact, knowing little of his
games I can easily recall a gross blunder without
any trouble, though it may or may not have
occurred in world championship play.

These numbers were taken from graphs in the paper that started this
thread.

and the worst performer was Steinitz at 0.054% (blunder
every roughly every 2000 moves).

How about defining what you mean by "blunder"?

I wish I knew. The summary article I read did not define it. I took it
to mean an unforced error significant enough that against another GM
or machine the game outcome was altered by (at least) half a point.

These are interesting numbers and
right at the limits of human error rates for purely trivial mechanical
tasks like punch key data entry.

Balderdash. Show me a data entry person who can
go 9,999 keystrokes without a blunder and I'll show
you a computerized robot/android! Back in the days
of punched cards, everything was run through twice,

I took the numbers from one of the human error rate review sites. One
that I think is pretty well reliable. That is for the very best that
can be acheived for the data entry task (typical commercial piecework
rates are much worse).

In any case, it is very difficult to have meaningful
discussion without a firm definition of what is to be
considered a blunder, and what not.

Unfortunately we can only discuss the summary article as published.
Unless someone has a link to the full version of the paper with the
complete details of the experiments or one of the authors is reading
this thread.

Regards,
Martin Brown

Even though I've been on the side that the original analysis
is interesting and of some value, your arguments are
simply not sound.

1. Your argument relies heavily on a set of
invented numbers "ringing true". That's not an
argument. In fact, the original article gives us
info indicating the level of error. It finds errors
when analyzing much stronger computers. (0.06-0.09)
This error is essentially the noise in the results.
It's not big enough to completely invalidate
the value of the assessments, but it certainly
throws into question the detailed rankings.
(Differences between humans are of the
order of 0.005, much smaller than the noise)

The authors did not do any of the obvious
things that they should have done to motivate
their choice of ply depth (12 + quiescence).


2. Even if the experiment were performed with
a perfect tool, the authors have not provided
any evidence that the measure chosen (average error,
corrected for position type) is the one that correlates
with winning play. We simply do not know whether
a string of 10 moves each with error 0.1 predicts
the same winning chances as 9 perfect moves
followed by a single error of 1.0. In fact, the only data they
gave supports the idea that the selected measure is
not that great: the correlation between the difference
in error and the outcome of the game was only 0.89.
Moreover, they give no details of that analysis.

What should have been done was to examine
games with larger differences and then experimented
with the measure to find the best corrrelation with
results. Once some general relationship was determined,
it could be applied to the world champions. Instead
they picked a (plausible) measure out of thin air and
made no effort to establish that it was meaningful.


"raylopez99" wrote in message
ups.com...
Let me close this thread by making some observations that will shed
some light on the two warring camps in this thread.

The first camp argues that a computer has to be (1) very, very strong
and (2) play like a GM, before it can be used to rate champions.

The second camp, myself included, argues that (1) any consistently
applied ("normalized") computer can be used, even a 'weak' computer
like Crafty, since the engines of computers

are largely the same (based on the MinMax and A/Beta algorithm, with
pruning and the like, and a decent evaluation function for position
evaluation of candidate moves), and (2) the

requirement that the program 'play like a GM' is not necessary (though
it is sufficient).

In truth, neither side has all the facts to make their case, but on
balance I believe the second camp is more persuasive.

A brief hypothetical will illustrate my point better.

Suppose that a version of Crafty favors "defensive" players like Capa
and Kramnik, while penalizing "attacking" players like Tal and
Fischer. Call this version of Crafty "Defensive

Crafty".

(Note the final results seem to give some credence to this view that
Crafty favors defensive or conservative players: from the Chessbase
article, Capa scored first at 0.1008 error rate,

followed by Kramnik at 0.1058, Karpov (another 'defensive' or
'positional' specialist) at 0.1275, Kasparov at 0.1292 (perhaps from
an Elo point of view the best player ever, but

remember the paper only looked at Championship match games), then
Spasky (a very well rounded player) at 0.1334, Petrosian (positional
God!) at 0.1343, Lasker at 0.1370 (best

player over the length of a career ever, one study found; perhaps
coincidentally he was known for finding great resources in defense),
Fischer at 0.1383 (an attacking player),

Alkehine at 0.1402 (another 'attacking' player), Smyslov at 0.1403,
Botvinnik at 0.1581 (a 'weak' World Champion some have argued, and
this study bears it out), Euwe at 0.177 (a

very underrated player), and Steinitz, from an earlier era where the
competition was weak and you could win more games with 'cheapos' so no
need to play flawless chess, at 0.23

(some have argued Steinitz was in fact, for his time, the best player
ever--same argument has been advanced for Morphy. THe point being, as
an aside, that if your competition is

weak, why bother playing perfect chess? It's like being the great
athlete Jim Thorpe or Jesse Owen--you have no competition so your
competition is yourself, which ultimately means

you fail to reach your highest potential).

Anyway, back to my hypothetical. Assume Defensive Crafty favors the
defense, not the offense. Let's throw in a few more players and
hypothetical championships: Janowski

(http://en.wikipedia.org/wiki/David_Janowski) and Tartakover (http://
en.wikipedia.org/wiki/Ksawery_Tartakower) for the offense, and Ulf
Andersson

(http://en.wikipedia.org/wiki/Ulf_Andersson) and Carl Schlecter
(http://en.wikipedia.org/wiki/Karl_Schlecter) for the defense, the
last two players known for their defensive skill.

Let's throw in four more hypothetical players: Class C Defensive Putz
(CCDP) and Class C Offensive Putz (CCOP) which we deem to be ELO 1500
players that are good at defense and

offense respectively. Also same distinction but with Master Defensive
Player (MDP) and Master Offensive Player (MOP) (strong national
masters)

Let's do a ranking of these players, using centipawns (cpawns), and
assume that Camp 1 is correct, in that Defensive Crafty is penalizing
offensive players (or to be more precise, let's

assume Crafty rewards players that play like it, and for purposes of
our discussion let's assume Crafty favors defensive players). We'll
assume that a 100 cp difference is a 'big deal'

(however you want to define that, perhaps like ELO 200 is a 'big deal'
as a difference in performance.

OK, we run the simulations, and if Camp 1 is correct, they'll look
something like this (best to worse players ranked by Defensive Crafty,
given our above assumptions):

[For brevity I'll leave out some of the above names, but you'll get
the idea]

1. Capa +200 cp
2. Kramnik +180 cp
3. Karpov +170 cp
4. Petrosian +160 cp
5. Lasker +155 cp
5.1 Andersson +150 cp
5.2 Schlecter +150 cp
6. Kasparov +75 cp (note the big drop! because he's an attacking
player)
7. Fischer +60 cp (" ")
8. Euwe +55 cp (Euwe ranked ahead of Alekhine since Euwe is
defensive!)
9. Alekhine +50 cp
9.1 Tartakover +33 cp
9.2 Janowski +31 cp
10. MOD +30 cp
11. MOP +10 cp (big difference between masters since MOD plays more
like our defensive minded hypothetical Crafty)
12. CCDP +9 cp
13. CCOP -25 cp (note CCDP nearly as good as a MOP, who is three or
four classes in ELO performance strength higher, simply because the
CCDP defense putz plays closer to our

hypothetical Crafty, meanwhile the CCOP offensive putz is heavily
penalized, even though both CCDP and CCOP are in the same ELO class
C.)

Does the above ring true? (Try and conceptualize what I'm getting at,
without referencing the actual Crafty in the article, since I'm making
a point based on hypotheticals. In fact

that's why we call the above "Defensive Crafty").

The above does NOT ring true, if you've ever played with different
chess engines. You know that playing with the parameters will not give
such a large change in state for the same

group of players. For example, there is no way Lasker is going to
come before Kasparov, I don't care how defensive minded he is, or that
our CCDP is going to nearly equal the MOP.

Nor that Andersson and Schlecter rank before Kasparov and Fischer. Nor
that the strong attacking GMs Janowski and Tartakover are only
slightly better than a mere national master

who plays good defense.

Before you squack about my making up the numbers, bear with me since
I'm not at the punch line yet.

We now assume that a version of Crafty --Offensive Crafty--favors the
offensive players.

Our hypothetical rankings:

1. Kasparov +500 cp
2. Fischer +450 cp
3. Janowski +200 cp
4. Tartakover +150 cp
5. Capa +50 cp
6. MOP +49 cp
7. Kramnik +45 cp
8. Andersson +10 cp
8. Schlecter +10 cp
9. CCOP +9 cp
10. MOD +5 cp
11. CCDP - 500 cp

Do you see where I'm going with this? If so, pat yourself on the
back. If not, keep reading. Obviously this list by our hypothetical
"Offensive Crafty" is as ridiculous as the first ranking

list, by "Defensive Crafty". No way, and no how, is this feasible, if
you know anything about chess and chess playing engines.

Now let's continue with a third and final hypothetical. Assume that
instead of the above scores, you got the following rating list
rankings, using EITHER the same hypothetical

"Defensive Crafty" OR "Offensive Crafty", but call it "Reasonable
Crafty", since it's the same program as the above, but it's a 'real
world' program IMO that will yield something that

resonates close to the truth, as we intuitively know it:

1. Capa +200 cp
2. Kramnik +200 cp
3. Karpov +199 cp
4. Petrosian +199 cp
5. Lasker +198 cp
6. Kasparov +196 cp (note the 'real good' players, that we know are
good, are bunched together)
7. Fischer +196 cp (" ")
8. Alekhine +188 cp
9. Euwe +130 cp
9.1 Schlecter +130 cp
9.2 Tartakover +100 cp
9.3 Janowski +100 cp
9.4 Andersson +100 cp

10. MOD +30 cp
11. MOP +29 cp
12. CCDP -25 cp
13. CCOP -25 cp

Now, does this list 'ring true'? Yes. You can quibble about the
ranking of the 'closely bunched up rated players' but the list rings
true overall--and if you know anything about chess,

you know it rings true. Master ranks ahead of patzer, no matter how
the program is 'tweaked', whether defensive or offensive.

Which gets to the punchline: if in fact Crafty produces an ordering
that rings true (our "Reasonable Crafty" above), based on what we
know, from historical ELO, and how humans play

chess, and the fact you know a good player when you analyze their
games, then you can make the claim that this Craft is not clearly
wrong. The program may be wrong in how it ranks

the players when the players are closely bunched together (i.e. the
first 7 players of the third list above), butthe program is not
clearly wrong overall.

Yet--here's the punchline--if Camp 1 above is to be believed, then
programs such as "Defensive Crafty" and "Offensive Crafty" exist, and
they will produce rankings as ridiculous and

clearly wrong as our first two examples. But in fact, not only does
that defy logic (if you've ever played with the parameters of a chess
program, making it more or less aggressive, and

noted that the best moves found are often not that different from
other parameter settings), but it also defies the *actual* rankings of
the Crafty in the original article reported in

ChessBase. As per the original article, the rankings were 'reasonably
bunched' as per our third hypothetical above and as I reported above
in the paragraph "Note the final results

seem to give some credence to this view that Crafty favors defensive
or conservative players".

Camp #1 wants you to believe that there are huge differences between
chess playing engines, that will give radically different rankings.
Camp #2 says the opposite--that this is not

true.

Yet here is a paradox: Camp #2 does not deny that the choice of who
is champion can be influenced by the chess engine, but ONLY IF THE
RANKINGS ARE CLOSE. In other words, if

you are using a "Defensive Crafty" to score games, and Defensive
Specialist #1, a very strong player, edges out an equally strong
(based on ELO) Offensive player #2 by 0.001 cp, then

yes, Camp #2 will acknowledge this ranking perhaps was due to the
"defensive bias" of the chess program. But the difference between the
players has to be minute and subtle.

Back to the actual article then: are the differences subtle?
Honestly, as a Camp #2 member, I don't know. But neither does Camp
#1. You have to play with numerous engines and see

if the rankings change. But just eyeballing the ranking, (see the
above paragraph "Note the final results seem to..."), I see certain
"breakpoints" in the rankings that indicate there are

clear demarcations in the list, and the list is not subtle.

Some breakpoints:

I. Capa and Kramnik. Clearly, their error rates are in the 'tens',
while the next group is in the 'twelves'--a clear class by
themselves. Perhaps Kramnik is better than Capa, depending

on the program used, but again that's a nit.

II. Karpov and Kasparov. Again, with "twelves" (0.1275 and 0.1292)
they are equal to one another (as evidenced by their nearly equal
match record) and in a class by themselves

III. Spassky, Petrosian, Lasker and Fischer (all about equal in
scores)

IV. Alekhine and Smyslov (though you can probably lump them with
group III)

V. Botvinnik

VI. Everybody else.

That's the way I see it, and until we get more research any rebuttal
to the contrary will simply be speculation, since the data is just not
there.

RL

On May 3, 8:14 am, "David Kane" wrote:
Even though I've been on the side that the original analysis
is interesting and of some value, your arguments are
simply not sound.

1. Your argument relies heavily on a set of
invented numbers "ringing true". That's not an
argument.

I think my post helps crystalize the two competing viewpoints, that I
laid out as Camp #1 and #2. Each camp has a good argument, but I have
a feeling my viewpoint will prevail.

In fact, the original article gives us
info indicating the level of error. It finds errors
when analyzing much stronger computers. (0.06-0.09)
This error is essentially the noise in the results.
It's not big enough to completely invalidate
the value of the assessments, but it certainly
throws into question the detailed rankings.
(Differences between humans are of the
order of 0.005, much smaller than the noise)

Good point. If this error is RMS error rather than net error, whether
this error or noise cancels out so to give a less than 0.005 true net
error remains a topic for discussion.



The authors did not do any of the obvious
things that they should have done to motivate
their choice of ply depth (12 + quiescence).

I wonder: why is that? Perhaps they did not have the hardware? 12
ply+ quiescence seems like it would take a long time. I think I read
once that 10 ply is about max for personal computers.


2. Even if the experiment were performed with
a perfect tool, the authors have not provided
any evidence that the measure chosen (average error,
corrected for position type) is the one that correlates
with winning play. We simply do not know whether
a string of 10 moves each with error 0.1 predicts
the same winning chances as 9 perfect moves
followed by a single error of 1.0.

As chess is a game of errors, and the serial Markov chain probability
of winning is probably weak in any given sequence of moves (that is,
from move-to-move, as a book by Australian chess master Purdy once
pointed out, and as is well known in the chess maxim that 'every board
position has to be looked at with a fresh pair of eyes, de novo,
without regard to what was played before'), I would imagine that the
former dominates the latter, but I agree this needs to be
investigated.


In fact, the only data they
gave supports the idea that the selected measure is
not that great: the correlation between the difference
in error and the outcome of the game was only 0.89.
Moreover, they give no details of that analysis.

0.89 correlation is very high, no? correlation is a real number 0 X
1.0, no? 89% is very high correlation.


What should have been done was to examine
games with larger differences and then experimented
with the measure to find the best corrrelation with
results. Once some general relationship was determined,
it could be applied to the world champions. Instead
they picked a (plausible) measure out of thin air and
made no effort to establish that it was meaningful.

They did make an effort, just that it raised more questions, as I
indicated in my "two Camps" post.

Thanks for your insight.

RL

On May 3, 4:59 am, "Chess One" wrote:
"raylopez99" wrote in message

ups.com...

Suppose that a version of Crafty favors "defensive" players like Capa
and Kramnik, while penalizing "attacking" players like Tal and
Fischer. Call this version of Crafty "Defensive

---
Interesting post, and the para above identifies the flaw in comparisons of
player-to-player by engine analysis- that if you are evaluating players,
then even Tal himself said that anybody at all could find his own flaws
after the game, or the next day, or even next week. But he chose 'em because
very few people could find them in real time OTB.

Yes, but this problem is also present in computer chess! Even if the
program is 'backsolving' or anotating a completed game, it has to look
in the chess tree, which means that it faces time constraints similar
to an OTB player vs a 'time is not of the essence' correspondence
player.



Given enough time then Stenitz will likely always rank higher than Tal by
analytical method, but OTB, super-solid Steintitz wouldn't know which way
Tal hit him! The flaw is that this analysis is okay for games, ie picking
some best theoretical line by objective and uniform measure against all
other lines - but chess playing is not a theoretical activity - its a
real-time performance.


Yes, and again, it's also true for chess playing computers (since no
program is given infinite time to analyse a game).

---

That's the way I see it, and until we get more research any rebuttal
to the contrary will simply be speculation, since the data is just not
there.

We rehearsed this conversation before - but there is no data of GM play
against raw chess engines. The engines are all optimised for winning, and
not for anything useful, like learning - either about its on evaluation
matrix of chess evaluation or even how people evaluate play. [because with
book+table bases=off, it wins less]

I disagree. I think winning is very closely correlated to 'learning
which move was the best'. And I don't think you therefore need a
database of GM play versus raw chess engines either, since winning is
immaterial as to how you win (whether thinking like a GM or thinking
like a machine using algorithms), though the SSDF site does just that
(rates PCs using humans).


I understand the commercial need to do that, but don't understand any
academic reason to choose emulation paradigms over [exigesic] real-time
engine evaluation. I always thought Crafty in particular would be such a
base model, since it is born out of a university system, widely distributed
and adapted, and lots of people might have had a go at it. But I think
Crafty got caught up in its own early success as W CH, and continued to go
for 'win', rather than for 'learn'.

Phil Innes


Crafty was once world champion? I didn't know that or it must have
escaped me.

Thanks,

Ray

On May 3, 5:44 am, (Dr A. N. Walker) wrote:
In article . com,
help bot wrote:

--
Andy Walker, School of MathSci., Univ. of Nott'm, UK.


You bothered to reply to help bot? help bot is pretty clueless, I
think he's essentially a troll.

Thanks for your insights though.

RL

In article . com,
raylopez99 wrote:

The program may be wrong in how it ranks

the players when the players are closely bunched together (i.e. the
first 7 players of the third list above), butthe program is not
clearly wrong overall.

But you've just set up a straw man.

Because the only thing we're interested in here, is the precise ranking
of the top players.

As the thread title says, Capa, then Kramnik, then Karpov, then
Kasparov. In that order.

If all the computer program is telling you is that Capablanca was
stronger than Janowski, well, shoot, you haven't really added anything
to the debate. We knew that already.

The whole POINT of the exercise is to take those people who are bunched
up at the top, and put them up in the correct order.

Now, what you've done, in your hypothetical example, is create two
programs which are hideously wrong. It thinks a class C player is
better than a master. Nobody's claiming that Crafty-at-12-ply is making
mistakes of that magnitude (that's a 600-rating-point error).

And, it seems to me, that you've basically conceded the point. The
biases of the program will affect it's judgement. Just because it's not
making 600-rating-point errors doesn't mean it isn't making
50-rating-point errors.

And, when talking about whether Kasparov is stronger than Petrosian, a
50-rating-point error could well give you the wrong result.

-Ron