Snowie Analysis of 100 Murat-Jellyfish 1 point matches
David Montgomery
I used Snowie to analyze 100 1 point matches between
Murat and Jellyfish. This is the second set of 100
games Murat played. In the first set, Jellyfish was
inadvertently set to play for money, while Murat was
playing for DMP. Murat posted the results of these
games, as well as the parameter settings and some
discussion in this article:
http://x6.dejanews.com/getdoc.xp?AN=374363109
I did not try to verify *anything* about these games,
but based on Snowie's analysis of JF's play, I see
no reason to doubt that JF was set up as Murat described,
playing 1 point matches on level 7 with a time factor
of 1000.
Murat won 53-54 of these 100 matches (see his post).
I hypothesized that Murat had had the better dice.
To test this hypothesis I ran all of the games through
Snowie. I used Snowie version 2, analyzing all the matches
at 3-ply 100% Huge.
A Snowie analysis produces a measure of how lucky each
player is -- that is, how much the dice favored one side
or the other. This figure is based on Snowie's evaluation
of each position before the dice are thrown, and on its
evaluation of the position after the dice are thrown and
the best move (in its opinion) played. It is very closely
related to variance reduction, so I refer readers who are
interested in more detail to the articles on variance reduction
in Tom Keith's Backgammon Galore archive.
One important point, however, is that the luck calculation
is unbiased. If Snowie has no clue what is a good roll and what
is a bad roll, then the luck calculation is just noise, and it
won't evaluate one side or the other as luckier on average.
In fact, Snowie has an excellent idea what is a good roll and
what is a bad roll nearly all the time, so its luck calculations
are quite accurate.
Two bugs in Snowie's luck calculation have been noted in
this newsgroup. Stuart Katz identified a problem that can
occur with matches imported from FIBS. This doesn't concern
us here. Also, Snowie doesn't calculate the luck for the
first roll correctly, failing to take into account that before
the first dice roll the evaluation should be 0. However, this
is of very little import. It is of no import unless either JF
or Murat started the game the vast majority of the time
(unlikely, but I haven't checked), and even then, it is minor.
The results of the analysis were as follows. Murat's luck
rate over the 100 1 point matches was +.473. Jellyfish's was
-.473. (They are always complementary, but I did both as a
check on my bookkeeping.) A positive luck rate corresponds
to better dice.
Without a doubt, Murat had the better dice.
Based on an independent collection of 52 1 point matches, I
estimated the standard deviation for the Snowie luck in a 1
point match as 2.16 (with mean 0 -- in my sample of 52 games
the mean was -.015).
Murat's dice rolls were well over 2 standard deviations from
the mean. This was a very unlikely event, but not as unlikely
as a 35-1 shot. It can happen.
The interpretation of the Snowie luck rate is especially simple
in 1 point matches. The luck rate is the (estimated) equity
benefit due to the dice over 100 moves. Murat's figure was
.473. Over a single move, the equity benefit was .00473. For
a single one pont match, which takes about 53 moves, the equity
benefit was .25069. This corresponds to 12 1/2%. Very roughly,
we can estimate that Murat's good rolls gave him about 12 1/2%
more wins than he would get if he played an infinite series of
games against the same opponent with fair dice.
This interpretation puts Murat about 300 FIBS rating points
lower than JF level 7[1000] in one point matches. This sounds
about right to me.
I want to thank Murat for sharing the data with me so I could
perform the analysis.
David Montgomery
mo...@cs.umd.edu
monty on FIBS and GG
Ian Shaw
wrote:
[snip]
>Without a doubt, Murat had the better dice.
So now even the 'bots are muttering together "Yeah, but he just got
lucky"!
>
>Murat's dice rolls were well over 2 standard deviations from
>the mean. This was a very unlikely event, but not as unlikely
>as a 35-1 shot. It can happen.
>
>This interpretation puts Murat about 300 FIBS rating points
>lower than JF level 7[1000] in one point matches. This sounds
>about right to me.
>
The last ranking I could find for jellyfish was Nov 1997, when it was
rated 2037.68 with 29548 experience points.
As of 7th Feb 1999, murat is rated 1836.73 with 3448 experience
points.
The difference is 200.95 points. Factors to consider are:
a) The "noise" in FIBS rating, variously estimated at 50 - 100 points,
could easily account for the disrepancy.
b) Jellyfish may not have been playing at Level 7[1000]
c) Murat was not playing of FIBS when these matches were played.
Therefore, his exposure to new opponents and styles of play have
improved his game to the point where he is only 200 points behind jf.
d) Murat has not been strictly playing-to-win all his FIBS matches. He
may therefore be better than his current rating suggests.
e) FIBS ratings are all hocus-pocus anyway. FIBS is part of the
ongoing US Government funded experiment into how people accept
external evaluations of their their worth, rather than relying on
internal self-awareness. Murat, by challenging the rankings, is
rocking the foundations of peoples' beliefs and providing valuable
experimental data. Murat Kalinyaprak is simply an anagram of "Yank up
karma trail".
>I want to thank Murat for sharing the data with me so I could
>perform the analysis.
>
Me too, and to you for analysing it.
--
Ian Shaw (ian on fibs)
JP White
> e) FIBS ratings are all hocus-pocus anyway. FIBS is part of the
> ongoing US Government funded experiment into how people accept
> external evaluations of their their worth, rather than relying on
> internal self-awareness.
Since Kit and Sally Woolsey provide the lions share of sponsorship for
FIBS, we are, therefore, to assume they are working as Government agents,
and that the US Government cleverly started FIBS in a foreign country,
then brought it over to the States in an attempt to try and divert
attention away from the fact the US Government is in control of the
experiment.
Interesting.
--
JP White
Mailto:jp.w...@nashville.com
Hugh McNeil
Ian Shaw wrote:
> e) FIBS ratings are all hocus-pocus anyway. FIBS is part of the
> ongoing US Government funded experiment into how people accept
> external evaluations of their their worth, rather than relying on
> internal self-awareness. Murat, by challenging the rankings, is
> rocking the foundations of peoples' beliefs and providing valuable
> experimental data. Murat Kalinyaprak is simply an anagram of "Yank up
> karma trail".
>
> >I want to thank Murat for sharing the data with me so I could
> >perform the analysis.
> >
> Me too, and to you for analysing it.
>
> --
> Ian Shaw (ian on fibs)
I think you miss the obvious, and that is that Snowie does not use the
cube to cash in on an opponent's weaknesses. People who are stronger cube
players will pick up ratings points through the mistakes of an opponent
that the 'bots cannot.
I'm pretty certain that the 'bots play, on the whole, better than anybody
in the world, but that you will see players with higher ratings than the
bots because equity gained through mistaken drops and takes tend to be
greater than the equity gained through close decisions, which is where the
'bots will generally excell.
Yes, i know, the 'bots play and evaluate some positions poorly.
David Montgomery
>On 6 Feb 1999 14:51:27 -0500, mo...@cs.umd.edu (David Montgomery)
>wrote:
>>lower than JF level 7[1000] in one point matches. This sounds
>>about right to me.
>
>The last ranking I could find for jellyfish was Nov 1997, when it was
>rated 2037.68 with 29548 experience points.
>As of 7th Feb 1999, murat is rated 1836.73 with 3448 experience
>points.
>The difference is 200.95 points. Factors to consider are:
Here is my interpretation.
JF Level 7 with a time factor of 5-10 has many thousand experience
points with an *average* rating of 2050. This figure is probably
the most nailed down of anything here.
Level 7[1000] plays better. It is hard to say how much better --
it is a lot better in positions requiring containment play, but not
such a big deal overall from the starting position.
I believe that very strong players who play a wide mix of opponents
achieve a higher rating by playing only 1 point matches. I'm not
talking about deliberate angling by seeking out players enormously
weaker, or with low experience -- I'm just talking about playing a
wide range of opponents in the 1650-1950 range.
I estimate that Level 7[1000] playing only 1 point matches would
have an average rating of at least 2100.
With this, a 300 point rating difference puts Murat at 1800. I
think Murat is a little overrated now because he has been experimenting
by playing people grossly weaker than him -- that is, he is
probably overrated relative to players who mostly play a wide range
of opponents with ratings within about 200 points either way of their
own rating.
(My guess is that Murat is underrated relative to the people that
only play others grossly weaker than themselves... if Murat continues
with his experiment, my guess is his rating will continue to rise
slowly for quite a while.)
Despite my beliefs above, my favorite explanation by far is:
>e) FIBS ratings are all hocus-pocus anyway. FIBS is part of the
>ongoing US Government funded experiment into how people accept
>external evaluations of their their worth, rather than relying on
>internal self-awareness. Murat, by challenging the rankings, is
>rocking the foundations of peoples' beliefs and providing valuable
>experimental data. Murat Kalinyaprak is simply an anagram of "Yank up
>karma trail".
Jim Cochrane
>Murat Kalinyaprak is simply an anagram of "Yank up
>karma trail".
And I thought it was "Mark a yak until a rap"
Ian Shaw
<hbre...@interlog.com> wrote:
[snip]
>
>I think you miss the obvious, and that is that Snowie does not use the
>cube to cash in on an opponent's weaknesses. People who are stronger cube
>players will pick up ratings points through the mistakes of an opponent
>that the 'bots cannot.
>
matches, and so was jf when it played on FIBS (I think).
Besides, Murat reckons himself to be weak with the cube compared to
his checker play, having been brought up playing the cubeless version
of the game.
Hugh McNeil
Ian Shaw wrote:
Silly me, seems i missed the obvious...
Murat Kalinyaprak
David Montgomery wrote news:79i6fv$n...@krackle.cs.umd.edu...
>I used Snowie to analyze 100 1 point matches between
>Murat and Jellyfish.
Thanks very much for this effort.
> This is the second set of 100 games Murat played. In
> the first set, Jellyfish was inadvertently set to play
> for money, while Murat was playing for DMP. Murat
> posted the results of these games, as well as the
> parameter settings and some discussion in this article:
> http://x6.dejanews.com/getdoc.xp?AN=374363109
In order to understand and make the most out of what
may be to come in this thread, I think one needs to
read that article pretty carefully. It talks about
purposefully making lesser moves against JF in order
to break (imaginary?) patterns in its dice, etc.
which were subsequently discussed quite at length...
>Murat won 53-54 of these 100 matches (see his post).
>I hypothesized that Murat had had the better dice.
>To test this hypothesis I ran all of the games through
>Snowie.
I did quickly mention before that talking about luck
factor/better dice in this case is meaningless. But
I didn't want to argue it to the point of talking you
out of doing this analysis because I was very curious
to see/analyse SW's output myself... When I find the
time, I'll examine the 40 detailed match analyses and
may make some comments on them.
>The results of the analysis were as follows. Murat's
>luck rate over the 100 1 point matches was +.473.
>....
>Based on an independent collection of 52 1 point matches,
>I estimated the standard deviation for the Snowie luck
>in a 1 point match as 2.16 (with mean 0 -- in my sample
>of 52 games the mean was -.015).
>....
>For a single one pont match, which takes about 53 moves,
>the equity benefit was .25069. This corresponds to 12 1/2%.
>Very roughly, we can estimate that Murat's good rolls gave
>him about 12 1/2% more wins than he would get if he played
>an infinite series of games against the same opponent with
>fair dice.
I hope all these numbers mean something to somebody
because they are totally irrelevant/useless for me
where the dice rolls are predetermined...
>This interpretation puts Murat about 300 FIBS rating points
>lower than JF level 7[1000] in one point matches. This
>sounds about right to me.
You have been saying similar things since a while
ago when my then lower rating was perhaps even more
encouraging you to make such comments (right when
you offered to analyse my 100 games against JF it
was at around 1730's, if I remember correctly).
Apparently you have quite a faith in the accuracy
and/or meaningfulness of FIBS ratings but I don't
know to what extent. So I would like to ask you
this: would you bet on JF level 7/1000 against me
in proportion to our predicted winning chances
based on my and its FIBS rating...?
>I want to thank Murat for sharing the data with me so I
>could perform the analysis.
Again, thank you for the trouble of doing it. I
hope something useful comes out of all our times
and efforts spent...
MK
David Montgomery
>Apparently you have quite a faith in the accuracy
>and/or meaningfulness of FIBS ratings but I don't
>know to what extent. So I would like to ask you
>this: would you bet on JF level 7/1000 against me
>in proportion to our predicted winning chances
>based on my and its FIBS rating...?
Well, I wouldn't say that I have "a faith" in the
FIBS rating system. I know how it works, and it
is one piece of information I might use in estimating
how strong players are. There are many other kinds
of information. You and I have played several matches,
so I have seen you play. The analysis of your 100
matches against Jellyfish is another independent
source of information.
I would be happy to bet on Jellyfish against you or
against almost any player in the world. Against
most players, I would be happy to back Jellyfish and
give the opponent a spot.
However, it would have to be worth my time. Entering
moves and dice rolls into Jellyfish and then making
its moves isn't my idea of a fun way to gamble. I would
only give a spot that I thought was less than JF's true
advantage, so that I would have a positive expectation.
For example, if we could play face to face, I would
be happy to back Jellyfish against you in 1 point matches
giving you $55 when you win, taking $45 when Jellyfish
wins.
This isn't anything personal about you, Murat. I
would do this (with what I felt was an appropriate
spot) against any player. And I would never play
Jellyfish 7/1000 for decent money without a big spot,
because I know that it plays a lot better than I do.
Murat Kalinyaprak
>On 6 Feb 1999 14:51:27 -0500, David Montgomery wrote:
>[snip]
>>Without a doubt, Murat had the better dice.
>So now even the 'bots are muttering together "Yeah, but
>he just got lucky"!
This was, of course, one of the possible explanations
even at the time those matches were played and I like
it... :) So many lucky rolls, so little time... :(
>>This interpretation puts Murat about 300 FIBS rating points
>>lower than JF level 7[1000] in one point matches. This sounds
>>about right to me.
>The last ranking I could find for jellyfish was Nov 1997, when
>it was rated 2037.68 with 29548 experience points. As of 7th
>Feb 1999, murat is rated 1836.73 with 3448 experience points.
>The difference is 200.95 points. Factors to consider are:
>a) The "noise" in FIBS rating, variously estimated at 50 - 100
>points, could easily account for the disrepancy.
I think the "noise" in FIBS ratings is much "louder"
than this but I don't believe the rating difference
talked about here is due to just this "noise".
>b) Jellyfish may not have been playing at Level 7[1000]
I thought it was playing at level 7 but with a lower
time factor (50 or less?). Can somebody confirm this?
>c) Murat was not playing of FIBS when these matches were
>played. Therefore, his exposure to new opponents and styles
>of play have improved his game to the point where he is
>only 200 points behind jf.
I have known/played the game for almost 30 years and
don't think that there is much to keep learning about
it. Any improvement probably happens in how fast and
how consistently a player can figure out better/best
moves. Since last summer was the time when I played
bg most intensively and against the most variety of
opponents as compared to any other time in my life.
So, I would find it likely that I may have "improved"
a little during this time (hopefully:) but would find
it unlikely that the amount of "improvement" would be
"visible" or considerable. Of course, I don't even
have a clue as to how many FIBS points any recent
improvement in my game would correspond to.
>d) Murat has not been strictly playing-to-win all his FIBS
>matches. He may therefore be better than his current rating
>suggests.
Keying on the term "strictly", what you say is pretty
much true. Of course, I wasn't playing to lose but my
primary concern wasn't necessarily always winning or
increasing my rating. I won't repeat them but want to
just remind that there are other factors (which I had
explained previously) that make my rating "inaccurate"
(and probably underrated at that).
>e) FIBS ratings are all hocus-pocus anyway. FIBS is part of
>the ongoing US Government funded experiment into how people
>accept external evaluations of their their worth, rather
>than relying on internal self-awareness.
This may just as well be true and there would be no
difference... :) It is accepted the FIBS formula is
arbitary (i.e. hocus-pocus), isn't it...? What is
especially impossible is implementing such a rating
system in a totally chaotic/anaymous environment like
Internet and without even any policies which could
remedy some of the problems to some extent.
The best argument that its defenders can make, is:
-If people play matches of widely enough varying lengths...
-If people play gainst a wide variety of opponents...
-If bla bla bla...
-If bla bla bla...
-Then, FIBS ratings may be reasonably accurate
to within N% error/variation, etc...
My rating is inaccurate. Ratings of "anglers" are
inaccurate. Ratings of people who keep losing and
repeatedly rejoining FIBS at 1500 under new names
are inaccurate. Some high-volume players (robots)
have rating swings as big large 250-300 points and
their ratings are inaccurate especially when they
are at the high/low extremes of those swings. Who
knows who else's ratings are inaccurate for whatever
other reasons...
Can any player still be said to have played a wide
enough variety of matches, against a wide enough
variety of opponents even after avoding all those
people with inaccurate ratings listed above...? I
doubt it and any player who plays against opponents
with inaccurate ratings will himself end up with an
inaccurate rating...
>Murat, by challenging the rankings, is rocking the
>foundations of peoples' beliefs and providing valuable
>experimental data.
Whether I do this or not, it's not intentional, as
I have nothing to gain from it either way. What is
puzzling me is people's lack of courage to express
publicly what they don't mind expressing privately,
on topics like their impressions about whether FIBS
dice is rigged or not, etc. Observing this behavior
probably matters to me more than FIBS dice.
Imagine a handful of people in a room and a "rigged"
thermometer on the wall. One asks: "Is it me, or is
it warm in here?". Another says: "I was wondering the
same thing". A third one goes looks at the thermometer
and declares: "It shows 72 degrees; it must be you",
(while he himself is sweating). And the chorus echoes:
"Oh well, we must all be running high fever then"...
I am not necessarily insisting that FIBS dice is rigged
but just trying to make a point that even the "logical"
conclusion of the "majority" may not always be proof of
what is real...
>Murat Kalinyaprak is simply an anagram of "Yank up karma trail".
One of the best I have ever seen... :)
>>I want to thank Murat for sharing the data with me so I could
>>perform the analysis.
>Me too, and to you for analysing it.
Thanks for the open-minded approach to possibilities.
I'll give some real stats related to my FIBS rating
in an another article within this thread...
MK
Murat Kalinyaprak
>In <36bebbd0....@news.force9.net> Ian Shaw writes:
>>On 6 Feb 1999 14:51:27 -0500, mo...@cs.umd.edu (David Montgomery)
>I estimate that Level 7[1000] playing only 1 point matches would
>have an average rating of at least 2100.
>With this, a 300 point rating difference puts Murat at 1800.
>I think Murat is a little overrated now because he has been
>experimenting by playing people grossly weaker than him --
>that is, he is probably overrated relative to players who
>mostly play a wide range of opponents with ratings within
>about 200 points either way of their own rating.
I think you are misunderstanding/misinterpreting a lot
of things... Let me try to clarify by giving some real
stats also. The below numbers are from a few days ago
and reflect all games I had played on FIBS by then.
Match Experience
Length Won Lost Total Points
------ ---- ---- ----- ----------
All 684 568 1252 3426
1 point 414 295 709 709
3 points 18 11 29 87
5 points 240 246 486 2430
7 points 11 15 26 182
9 points 1 1 2 18
Now let me itemize a few more facts:
1) At first I was afraid that cube players on FIBS
would eat me alive, so for the first couple of weeks
I had stuck to playing 1-pointers, which probably
numbered at least in a couple of hundreds or so.
2) Then I switched to playing 5-pointers with the
cube almost exclusively, except for the 1-pointers
I played against robots like Abbott, MonteCarlo, etc.
and some human players who refused to play anything
but 1-pointers (those 1-pointers probably numbered
in the 150-200).
3) An overwhelming majority of the above hatches
were played against relatively and progressively
higher rated players (with rare exceptions if there
were very few people logged on or when somebody
invited me persistently, etc.)
4) I only started playing against players with very
low ratings at around January 10th and pretty much
stopped doing it by January 29th (a few days after
which the above stats were compiled). I estimate the
number of 1-pointers I played during that time span
(based on dates of saved profile files and a sampling
of their contents for multiple matches against same
opponents) at about 250 or so.
So, by January 10, my rating was already above 1800
and my experience well past 3000. During that period,
my rating fluctuated, going up as far as 1860's (if
I remember correctly and perhaps even close to 1870)
and down to 1820's. Based on this and the "average"
(low) rating of my opponents, I estimate my win/lose
ratio to be about 1.5 (i.e. 100 losses for 150 wins).
And *this is why* I raised an issue with it in the
first place...!
There I was, playing the match legth that supposedly
would favor me, against the weakest players, and yet
after a big surge during the first couple of days
of this experiment, I was barely doing as well (in
fact, even slightly worse) as I had done all along
against players stronger than myself. Being underrated
in my checker skills, I was the perfect condidate who
should be able to "angle" and bump my rate up quickly,
at least to the extent of my being underrated in
1-pointers (i.e. checker play). But it didn't happen.
Why do I think I'm so underrated in checker play? Look
at the ratio of multi-pointers I won/lost. My net gain
from all 1-pointers played prior to January 10 would
at best add up to about 160 points (based on about 70
surplus of wins at 2.25 points per game; most likely
the real number is quite less than that). That leaves
at least another 170 (and most likely more than that)
points I need to account for and by looking for their
source I see something very interesting. In the number
of multi-point matches I'm a net loser with 270 wins
against 273 losses (almost flat even). Obviously those
points must have come from my playing against players
rated much higher than myself. For a "cube rookie" to
accomplish this, he must have extraordinary checker
skills (and thus grossly underrated in that area). Of
course, there is an alternative, which is that maybe
the "cube skill" is just not what it's bragged up to
be... :)
Oh, btw, based on my own experience, I'm confused
about what may be the best way to "angle". Maybe I
should stick to "preying" on highest rated players
instead of the lowest rated ones. And others may want
to give it a try just for the heck of it also... :)
>(My guess is that Murat is underrated relative to the people
>that only play others grossly weaker than themselves...
If I'm reading what I think I'm reading here, this
sounds terrible in terms of credibility/accuracy of
FIBS rating system (and I don't mind it at all:)...
>if Murat continues with his experiment, my guess is his
>rating will continue to rise slowly for quite a while.)
The 250 or so games I played in a span of about 20
days almost points in the opposite direction. But
I agree that my expectations were in the direction
that my rating would go up by doing this also and
I may experiment with it again...
If I could bump up my rating by 200 points overnight,
I would. And the only reason I would want to do that
is beacuse I often wondered whether (under the FIBS
rating scheme) it wouldn't be easier to maintain a
given rating than to attain that rating after starting
at 1500. In other words, I wonder where would I be
now (after 3500 experience) had I set out to find my
"true rating" going downwards from 2100 instead of
going upwards from 1500...
MK
Daniel Murphy
<mu...@compuplus.net> wrote:
>I have known/played the game for almost 30 years and
>don't think that there is much to keep learning about
>it.
I appreciate few of Murat Kalinyaprak's voluminous contributions to r.g.bg
as much as this one, for here is a succinct statement of the most
formidable obstacle to becoming a better player. For if this is one's
opinion of one's ability, and unless one's assessment is accurate, these
are nothing but words to lose by.
In "Dynamic Cube Strategy," Gaby Horowitz, the infamous backgammon cheat
(but damn good backgammon teacher) wrote: "One of the first steps in
learning anything is to realize that there is something that you don't know
about the particular subject you are about to study." In backgammon,
ignorance of one's own ignorance makes improvement exceedingly difficult.
Ignorance in combination with a blinding obsession with the luck of the
dice makes improvement impossible.
That's not the end of the world, of course. The game may still be fun. The
dice may still be an endless source of amusement. But refusing to
acknowledge the possibility that one has not yet learned everything there
is to know does condemn one to continue repeating the same mistakes,
attributing one's well-deserved victories to good play and blaming one's
unhappy losses on bad dice.
In "Vision Laughs At Counting," Danny Kleinman wrote:
"You have read all the best books on backgammon. You have spent leisurely
hours analyzing positions. You have come up with the correct solutions to
all the problems in newspaper and magazine columns. But you are not yet
ready to be a winner in a tough game at the club. All you have learned is
not enough. You need experience playing in tough games, and lots of it.
"Recognize how difficult a game backgammon really is. It requires alertness
and judgment. THERE IS SO MUCH IN AN ORDINARY POSITION TO LOOK AT AND THINK
ABOUT. And you cannot really ever take enough time to do it all."
This winter I watched and recorded a 17-point match between two strong
players, both with over 10 years' experience and some tournament successes
to boast of. The final score was 17-13. When it was over, Mr. Unlucky's
only comment about the match was: "I'm *sure* she had a drop when I doubled
in the 11th game!"
Bad luck again, no doubt. So a few days later I'm feeding the match to
JellyFish and observing Mr. Unlucky's play, and what do you suppose I see?
Game 1, Move 3: fifth best move. Move 4: 2nd best. Move 6: missed a
double/take. Move 8: doubled in a no double/take position, and followed up
with a game-losing 2nd best move.
And so it continued. Missed doubling opportunities. Taking drops. Dropping
takes. And suboptimal checker play throughout the match. Mr. Unlucky's
double in the 11th game turned out to be a good take; sadly, his own take
in game 5 was a clear drop.
I saw Mr. Unlucky some weeks later, and since he had received a copy of the
match but had not yet reviewed it, he asked me what I had thought of his
cube action. Ever the diplomat, I replied that the match had presented both
players with numerous difficult decisions.
Difficult decisions -- arising from ordinary backgammon positions. Some
time ago in rec.games.backgammon, David Montgomery suggested a formula for
improvement. I believe he said:
"Identify your mistakes. Correct your mistakes. Stop making those
mistakes."
But first, admit even the merest possibility that you are making mistakes.
Not only minor errors, but game losing blunders as well!
I've not meant this as a personal chastisement for Murat Kalinyaprak. I've
known so many backgammon players who are incapable of improving. They know
they are not experts. They admit to making occasional errors. What they
find incomprehensible is why they keep losing to opponents they "know" play
worse than themselves.
But Murat Kalinyaprak's preoccupation with dice and his incessant search
for explanations wholly outside of himself (rigged servers, cheating dice)
for his unlucky streaks indicate to me that he may not be seeing the errors
in his own play.
"I don't think that there is much to keep learning" strikes me as a baldly
delusional self-assessment for anyone but a master player.
And when Murat Kalinyaprak says "From what I see thus far, cube only
effects how the score is kept and nothing more.... The cube is a gambling
tool that can be used in any competition/game at the expense of killing its
character/spirit and serves no other use more in my opinion..." -- what
does this indicate but a complete lack of appreciation for the difficulty
and subtleties of backgammon with the doubling cube?
There is no shame in playing less well than one desires. Any backgammon
match, even between experts, is replete with errors in both checker and
cube play. But there is something unfortunate in refusing to examine what
-- for most backgammon players -- is the main reason we lose: our own
mistakes.
I've never taken up Murat's invitation to watch him play and await those
uncannily predictable horror rolls he's claimed to have suffered, but I
have played him once on FIBS, in a 5-point match which he won in 4 games,
5-0. The cube was turned once. In 4 games, Murat had 61 unforced moves to
make. In the first 3 games, Murat had 47 cube decisions to make (47, for as
every player should know, *every* roll presents a cube decision).
So if Murat Kalinyaprak unshakingly feels that none of my comments have the
least application to him, I would invite him to estimate:
How many of his 61 moves did JellyFish rate as best? How many were second
best? How many were third best? How many were fourth best? And how many
were twentieth best?
That will account for all 61 moves (for Murat does move the checkers fairly
well -- just not as well as someone who doesn't have much to learn about
the game).
How many times in the first 3 games should he have doubled me into a take
but did not? How many times should he have doubled me into a drop but did
not? How many times did he double in a "no double/take" position?
The guesses may be illuminating. I know I learned something from reviewing
our match.
_______________________________________________
Daniel Murphy http://www.cityraccoon.com
Humlebæk BG Klub http://www.hbgk.dk
Raccoon on FIBS http://www.fibs.com
"Anybody who likes backgammon should join FIBS right now and stop playing
against computer programs." - Murat Kalinyaprak, Sept. 17, 1998
kap...@banet.net
plays are 2nd best third best or even 10th best. Or snowie for that matter!
Robert-Jan Veldhuizen
Murphy) wrote:
[about improving on the cruelest game...]
>_______________________________________________
>Daniel Murphy http://www.cityraccoon.com
>Humlebæk BG Klub http://www.hbgk.dk
>Raccoon on FIBS http://www.fibs.com
>
>"Anybody who likes backgammon should join FIBS right now and stop playing
>against computer programs." - Murat Kalinyaprak, Sept. 17, 1998
Great quote, Daniel, you're indeed the diplomat you claim to be ;-)
Mind if I steal your tagline? :-)
--
Robert-Jan/Zorba
Daniel Murphy
>Did it ever occur to you that jellyfish may be wrong?
Yes. In the Backgammon Galore rec.games.backgammon archive at
www.bkgm.com/rgb/rgb.cgi?view+454, you'll find an article I wrote in
January 1998 about some positions where JellyFish is wrong.
And if you're ever in Denmark, there are a couple of propositions I would
be happy to play -- taking in positions that JellyFish says are "too good"
to double. The Fish isn't always right.
>How many of jellyfish's plays are 2nd best third best or even 10th best. Or snowie for that matter!
If I said "few enough that backing JellyFish or Snowie against almost any
human player is a real good bet," would that be precise enough? Or suppose
we took 100 positions at random where JellyFish's preferred play is
different, say, from yours! How confident would you be that you had made
the correct choices? Or suppose we found 100 positions from matches between
world-class human players where JellyFish preferred a different play. Who
would you put your money on?
Or suppose you have a 66 to play in this position:
0-0, Match to 5
X to play 66
+24-23-22-21-20-19-+---+18-17-16-15-14-13-+
| X O | | O X |
| O | | O X |
| O | | 0 X |
| 0 | | 0 X |
| | | |
| | 0 | | 64
| | | |
| | | 0 |
| X | | 0 |
| X X | | 0 |
| X X | | X 0 |
| 0 X X | | X X 0 |
+-1--2--3--4--5--6-+---+-7--8--9-10-11-12-+
Please choose between playing 13/1*(2) or 20/8 13/7(2). Guess which play
was JellyFish's first choice and which was the human player's. Guess who
was wrong. And do you think it made much difference?
Chuck Bower
>> <mu...@compuplus.net> wrote:
>>
>> >I have known/played the game for almost 30 years and
>> >don't think that there is much to keep learning about
>> >it.
>Daniel Murphy wrote:
>
(snip)
>> "I don't think that there is much to keep learning" strikes me as a baldly
>> delusional self-assessment for anyone but a master player.
IMO, Daniel's post was excellent. The only thing I disagree with
is the last phrase above ("...but a master player.") I would remove this
completely.
(Daniel continued:)
(snip)
>> I've never taken up Murat's invitation to watch him play and await those
>> uncannily predictable horror rolls he's claimed to have suffered, but I
>> have played him once on FIBS, in a 5-point match which he won in 4 games,
>> 5-0. The cube was turned once. In 4 games, Murat had 61 unforced moves to
>> make. In the first 3 games, Murat had 47 cube decisions to make (47, for as
>> every player should know, *every* roll presents a cube decision).
>>
>> So if Murat Kalinyaprak unshakingly feels that none of my comments have the
>> least application to him, I would invite him to estimate:
>>
>> How many of his 61 moves did JellyFish rate as best? How many were second
>> best? How many were third best? How many were fourth best? And how many
>> were twentieth best?
>>
>> That will account for all 61 moves (for Murat does move the checkers fairly
>> well -- just not as well as someone who doesn't have much to learn about
>> the game).
>>
>> How many times in the first 3 games should he have doubled me into a take
>> but did not? How many times should he have doubled me into a drop but did
>> not? How many times did he double in a "no double/take" position?
>>
>> The guesses may be illuminating. I know I learned something from reviewing
>> our match.
In article <36C6FEE1...@banet.net>, <kap...@banet.net> wrote:
>Did it ever occur to you that jellyfish may be wrong? How many of jellyfish's
>plays are 2nd best third best or even 10th best. Or snowie for that matter!
I had an experience over the weekend that might shed some light on
this subject, including kapla24's question. I attended the (always
enjoyable) Pittsburgh Backgammon Championship directed by Steve Hast and
his capable crew of assistants. About 12:30 AM Saturday after a couple long
doubles matches I headed down the hallway to snare a nightcap when I ran
into (not literally ;) Neil Kazaross. After a bit of smalltalk we landed
on the subject of Snowie and wound up discussing bots and some positions
for about an hour.
For those of you who aren't familiar with the backgammon tournament
scene, Neil is one of the world's top players. (This can be substantiated
by his tournament record as well as his high ranking in the "Giant 32 of
backgammon" survey.) He is an interesting person (but who among us isn't?)
--a securites trader by profession. Neil has an outgoing personality, a
GREAT enthusiasm for all parts of the game, and is happy to share his
ideas and opinions with others.
Neil mentioned that he had taken some time off from work recently and
had been alternately playing matches on GamesGrid and plugging the match
logs into Snowie for its match analysis. When discussing the strengths
of the two bots (SW and JF) and their human counterparts, he said that
he felt that at match play, SW was a bit stronger than JF. (I'm not sure
which versions--probably JF3.0 and SW2.0, but he did mean at their
respective highest levels.)
Neil then went a bit further. He said that he would back Snowie at
match play against any human player. And, unlike the hollow claims we so
often see on this newsgroup, he said at $1000 per match. Does Neil think
SW and JF make mistakes? Yes. Just not as often (or with as high a
magnitude) as the top humans.
I didn't ask Neil if he is still learning backgammon. I wouldn't
waste his (or my) time with a question whose answer is so obvious.
Finally, take a guess who won the Open Division at this weekend's Pittsburgh
Championships. [No, not me. :( But thanks for the vote.]
Chuck
bo...@bigbang.astro.indiana.edu
c_ray on FIBS
kap...@banet.net
player. The point is people rely too much on these bots and see them as infallible
when they are from from it. People don't seem to be able to think about a
backgammon position for themselves and have become way too dependent on these
bots.
Chuck Bower wrote:
> >> On Fri, 12 Feb 1999 13:19:19 -0700, "Murat Kalinyaprak"
> >> <mu...@compuplus.net> wrote:
> >>
> >> >I have known/played the game for almost 30 years and
> >> >don't think that there is much to keep learning about
> >> >it.
>
> >Daniel Murphy wrote:
> >
> (snip)
> >> "I don't think that there is much to keep learning" strikes me as a baldly
> >> delusional self-assessment for anyone but a master player.
>
> IMO, Daniel's post was excellent. The only thing I disagree with
> is the last phrase above ("...but a master player.") I would remove this
> completely.
>
> (Daniel continued:)
> (snip)
> >> I've never taken up Murat's invitation to watch him play and await those
> >> uncannily predictable horror rolls he's claimed to have suffered, but I
> >> have played him once on FIBS, in a 5-point match which he won in 4 games,
> >> 5-0. The cube was turned once. In 4 games, Murat had 61 unforced moves to
> >> make. In the first 3 games, Murat had 47 cube decisions to make (47, for as
> >> every player should know, *every* roll presents a cube decision).
> >>
> >> So if Murat Kalinyaprak unshakingly feels that none of my comments have the
> >> least application to him, I would invite him to estimate:
> >>
> >> How many of his 61 moves did JellyFish rate as best? How many were second
> >> best? How many were third best? How many were fourth best? And how many
> >> were twentieth best?
> >>
> >> That will account for all 61 moves (for Murat does move the checkers fairly
> >> well -- just not as well as someone who doesn't have much to learn about
> >> the game).
> >>
> >> How many times in the first 3 games should he have doubled me into a take
> >> but did not? How many times should he have doubled me into a drop but did
> >> not? How many times did he double in a "no double/take" position?
> >>
> >> The guesses may be illuminating. I know I learned something from reviewing
> >> our match.
>
> In article <36C6FEE1...@banet.net>, <kap...@banet.net> wrote:
>
> >Did it ever occur to you that jellyfish may be wrong? How many of jellyfish's
> >plays are 2nd best third best or even 10th best. Or snowie for that matter!
>
Murat Kalinyaprak
>In <79rku3$59...@taisp3.in-tch.com> Murat Kalinyaprak writes:
>>Apparently you have quite a faith in the accuracy
>>and/or meaningfulness of FIBS ratings but I don't
>>know to what extent. So I would like to ask you
>>this: would you bet on JF level 7/1000 against me
>>in proportion to our predicted winning chances
>>based on my and its FIBS rating...?
>Well, I wouldn't say that I have "a faith" in the
>FIBS rating system. I know how it works, and it
>is one piece of information I might use in estimating
>how strong players are.
I hope we would agree that the "FIBS rating *system*"
is more than the formula it uses. I can't even bring
myself to use the word "works" when talking about an
arbitrary formula. When even the usefulness of such a
formula in a physical/finite/controlled environment
would be questionable, it's hard for me to understand
how anybody could think/argue that it could work in a
virtual/infinite/uncontrolled environment (as FIBS on
cyberspace) without some "faith" (for lack of a better
term)...
>There are many other kinds of information. You and I
>have played several matches, so I have seen you play.
I think we had played some unrated games in the past,
but I found no record of them in your BBGT profile.
Other than that, we played 6 cubeful 5-point matches
a few weeks ago, with you winning 5 to 1. What stands
out in my mind from those matches is that there were
a few watchers, you had commented yourself that dice
had favored you for (at least) the fist few matches
(as best as I can recall), I had made decisions based
on my belief that I was "scheduled" to lose some games
anyway. I wouldn't blame anybody for thinking of such
acts as dumb but I remember clearly that in one match
I had resigned 4 points (and thus the whole match with
it) when I didn't even have to resign 2 points... If
you recall anything more and/or different about those
matches, I would be happy to hear them (as well as any
observations/conclusions you had derived from them).
Based on the moves/decisions I make, most people would
probably just wonder how in the world I ever reached a
rating above 1800 (without angling or cheating) but if
you can somehow relate my rating to what you observed
playing against me, well, I guess that's great... :)
>The analysis of your 100 matches against Jellyfish is
>another independent source of information.
Yes and it looks like it presents for some people a
problem that they don't quite know how to handle...
Some guy (me:) had talked about "breaking patterns
in JF's dice or throwing off JF by making unexpected
lesser moves", about "the way to beating JF being to
play not like JF but unlike JF", etc...
At that point, if you had called him some names and put
him in your kill-file, you had "solved":) your problem.
But if you kept listening, he then said he played 100
1-pointers against JF with such and such seed/counter
settings for each match. If we let JF1 play against JF2
100 1-pointers with the same seed/counter values, JF2
loses by a couple of points.
This is the best a "near perfect" player can do with
the dice given to it against another "near perfect"
player. Then this guys comes, he deviates from what
would be the "perfect/near perfect" moves and does
a blunder here and there (as he said he was going to
do knowingly at least part of the time) and ends up
winning aginst JF1 by a couple of points...
I suppose if we substitute "Jerry-Fish", Snowie or
another world-class player for JF2, we would obtain
similar results. So, how do we explain this...? Even
my "wild" speculations about patterns in the dice
rolls, etc. sound better than your proposing that I
had gotten "lucky". Snowie's analysis confirming it
is even worse because it does effectively prove that
an intentional blunder turns a not-so-lucky sequence
of subsequent dice rolls into a lucky sequence...(?)
>For example, if we could play face to face, I would
>be happy to back Jellyfish against you in 1 point matches
>giving you $55 when you win, taking $45 when Jellyfish
>wins. This isn't anything personal about you, Murat. I
>would do this (with what I felt was an appropriate
>spot) against any player.
I'm not taking it personally either. I just wanted to
know where you would draw the line with money involved.
A 55/45 ratio would translate to about 175(?) FIBS points
difference between my/JF's ratings, which is quite less
than the 300+ (with wmy rating at around 1730's at the
time you had first offered to put those 100 MK/JF matches
through Snowie). Now my rating is at around 1830's but
based on JF's rating of about 2050, I guess I can look
forward to reaching my "true" FIBS rating at 1875's... :)
MK
Murat Kalinyaprak
Daniel Murphy wrote news:36c6a10c...@news.inet.tele.dk...
>Murat Kalinyaprak wrote:
>>I have known/played the game for almost 30 years and
>>don't think that there is much to keep learning about
>>it.
>I appreciate few of Murat Kalinyaprak's voluminous
>contributions to r.g.bg as much as this one, for here
>is a succinct statement of the most formidable obstacle
>to becoming a better player. For if this is one's
>opinion of one's ability, and unless one's assessment
>is accurate, these are nothing but words to lose by.
Daniel, I enjoyed this article of yours but I don't
understand why you would think that I wouldn't want
to become a better player. I think there is a little
miscommunication here. I'll try to clarify some and
hopefully "contribute" a few more useful words...
>Ignorance in combination with a blinding obsession with
>the luck of the dice makes improvement impossible.
Some of us are more observant than others and also
some of us have been trained to, became experienced
in and sometime even get paid to:) "observe/analyse".
If you have been following this newgroup for the past
year, you probably know that there has been a few
cases where I noticed/questioned/discovered some things
about bg bots, etc. that had escaped the masses for
long times. Of course, this doesn't mean that what I
claim to observe about some bg servers dice rolls is
necessarily true but I would like to clarify that my
"obsession" with the dice that you may be referring to
never involved manual dice. And when talking about a
bot's or server's dice generator, I have never used
the term "luck" either...
>"You have read all the best books on backgammon. You have
>spent leisurely hours analyzing positions. You have come
>up with the correct solutions to all the problems in
>newspaper and magazine columns. But you are not yet ready
>to be a winner in a tough game at the club. All you have
>learned is not enough. You need experience playing in
>tough games, and lots of it.
In my previous article, I had expressed that I had
"hopefully" *improved* since I have been playing bg
so intensly during the past 6 months. But I don't
equate "improving" with "learning" as I don't even
equate "knowing the game" with "being good at it"...
For me, knowing bg involves the rules, the object of
the game and the fundemental strategies. The number
of positions and dice combinations seem to be in the
millions if not billions. Figuring out the "correct"
move in each case is not quite learning, reading
about some specific cases in books in not learning
either, as the same case won't happen again within
the next 200 years and even if it did, you wouldn't
remember/recognize it anyway. (Of course, if I were
peddling my bg books, on-line magazine subcriptions,
etc. I would be talking differently:)...
>"Recognize how difficult a game backgammon really is. It
>requires alertness and judgment.
Unfortunately these can't be learned... :(
>THERE IS SO MUCH IN AN ORDINARY POSITION TO LOOK AT AND
>THINK ABOUT. And you cannot really ever take enough time
>to do it all."
Yes, I'm sure there are some position/dice combinations
that we had faced numerous times over but at their next
occurrence we pause and think again, instead of making
the move roboticly (as a result of having learned them).
Notice that we talk about "skill" in bg quite often and
perhaps much more often than "knowledge". As you quoted
above, experience can "develop" and intensive playing
can "sharpen" that skill...
>... They admit to making occasional errors. What they
>find incomprehensible is why they keep losing to
>opponents they "know" play worse than themselves.
>But Murat Kalinyaprak's preoccupation with dice and his
>incessant search for explanations wholly outside of
>himself (rigged servers, cheating dice) for his unlucky
>streaks indicate to me that he may not be seeing the
>errors in his own play.
The problem with your argument here is that I don't
search for explanations outside of myself only for
my unlucky streaks but also for my lucky streaks...
Remember what wild explanations I had come up with
for my beating JF in a set of 100 1-pointers...?
David Montgomery fed those matches through Snowie's
analyser and determined that the explanation was
much simpler (yet still outside of myself though:)
that I had simply gotten lucky... :)
I recently replayed those 100 games and a new set
of 10 matches of 25 points against JF level 7/100.
I'll report the results and post some comments on
them soon...
>And when Murat Kalinyaprak says "From what I see thus far,
>cube only effects how the score is kept and nothing more....
>The cube is a gambling tool that can be used in any
>competition/game at the expense of killing its character/spirit
>and serves no other use more in my opinion..." -- what does
>this indicate but a complete lack of appreciation for the
>difficulty and subtleties of backgammon with the doubling cube?
I thought we had gone through this extensively enough
but if you want to rehash it, fine with me. Nobody
has yet responded to my asking how they play bg any
differently with the cube than without the cube in a
way that doesn't involve a decision based on score...
I don't deny the skill needed to use the cube but I'll
stick to my argument that it's external to backgammon
as it could be used in any competition for gambling
purposes (i.e. basketball, tennis, chess, checkers,
you name it) and it would be external to those games
also...
Initially I had kept from making any comparative
judgement between checker play skills and cube skills
but somebody with zero cube knowledge/experience like
me reaching a rating above 1800 by playing mostly 5
point cubeful matches against ever higher rated
opponents should be telling something...
>I've never taken up Murat's invitation to watch him play
>and await those uncannily predictable horror rolls he's
>claimed to have suffered, but I have played him once on
>FIBS, in a 5-point match which he won in 4 games, 5-0.
>The cube was turned once. In 4 games, Murat had 61 unforced
>moves to make. In the first 3 games, Murat had 47 cube
>decisions to make (47, for as every player should know,
>*every* roll presents a cube decision).
>So if Murat Kalinyaprak unshakingly feels that none of my
>comments have the least application to him, I would invite
>him to estimate:
>How many of his 61 moves did JellyFish rate as best? How
>many were second best? How many were third best? How many
>were fourth best? And how many were twentieth best?
I don't understand the part about your comments not
having application to me but I would be gald to try
answering the above questions if you had recorded
the match and can send me a copy of it in text or
JF .mat format. If you would later compare then to
what JF actually says, it would be very interesting
for me to see it also.
>That will account for all 61 moves (for Murat does move
>the checkers fairly well -- just not as well as someone
>who doesn't have much to learn about the game).
With my differentiating between "learning more" and
"improving" clarified above, I have no claims to be
such a good player. However, I'm not a robot and my
making a certain move may not necessarily indicate
my not knowing a better/best move. If you warn me
ahead of time that you will feed my moves to a bot's
analyser to measure my skills, I think I would play
a little differently and probably do quite well too.
(except for cube handling:( of course).
>How many times in the first 3 games should he have doubled
>me into a take but did not? How many times should he have
>doubled me into a drop but did not? How many times did he
>double in a "no double/take" position?
What is the ultimate goal...? Winning, right? So,
why does it matter how did I win enough of all of
those 5-point cubefule matches to reach a 1800+
rating on FIBS...? Personally, because of some
comments about the cube made to me (including the
private ones by email), I'm deriving even a slight
pleasure out of trying to and succeeding to some
extent in marginalizing the cube by checker play...
>The guesses may be illuminating. I know I learned
>something from reviewing our match.
If you would like to post the match here and make
comments on it publicly, please feel free to do
so. I would appreciate the opportunity to "learn":)
something from it as much as anybody else would...
MK
David Montgomery
>David Montgomery wrote news:79shkv$f...@krackle.cs.umd.edu...
>>be happy to back Jellyfish against you in 1 point matches
>>giving you $55 when you win, taking $45 when Jellyfish
>>wins.
>I just wanted to
>know where you would draw the line with money involved.
>A 55/45 ratio would translate to about 175(?) FIBS points
>difference between my/JF's ratings, which is quite less
> than the 300+
As I mentioned, this sounds boring to me, so I want to
make some money at it. It is an opportunity for you to
make money, too, if you can continue to beat or play even
with JF.
But I have decided that the above spot was too much - not
because JF wouldn't beat you enough, but because my time
is worth more than that. So I hereby rescind that offer.
Should you ever really want to play, send me email or call
me and we can negotiate.
David
David Montgomery
>David Montgomery fed those matches through Snowie's
>analyser and determined that the explanation was
>much simpler (yet still outside of myself though:)
>that I had simply gotten lucky... :)
I'm pretty sure that I didn't say that you got lucky.
I said you had better dice.
There are a lot of ways that these two statements
could be different. You have claimed that you often
have foreknowledge of the rolls that are coming up,
due to your ability to intuit the devious workings
of the dice generators. If you know the dice coming
up, you can move so that you get the better dice --
but because it's based on some sort of knowledge or
skill or intuition, or some other factor, it isn't lucky.
So you *may* have gotten lucky. You *definitely*
had the better dice.
Robert-Jan Veldhuizen
<mu...@compuplus.net> wrote:
>Even
>my "wild" speculations about patterns in the dice
>rolls, etc. sound better than your proposing that I
>had gotten "lucky".
Err, no.
--
Robert-Jan/Zorba
Robert-Jan Veldhuizen
<mu...@compuplus.net> wrote:
>Some of us are more observant than others
Daniel seems one of the more observant types I think...
> and also
>some of us have been trained to, became experienced
>in and sometime even get paid to:) "observe/analyse".
>
>If you have been following this newgroup for the past
>year, you probably know that there has been a few
>cases where I noticed/questioned/discovered some things
> about bg bots, etc. that had escaped the masses for
>long times.
So, did that improve your bg skills a lot? Was it interesting at all?
>The number
>of positions and dice combinations seem to be in the
>millions if not billions.
Add a few more zeros there.
>Remember what wild explanations I had come up with
>for my beating JF in a set of 100 1-pointers...?
They were sure wild, yes. :-)
For with your rating it isn't that surprising that you beat JF in a 100
1pt games contest; if I remember correctly you have at least a 5% chance
if rated 300pts lower (for 1ptrs, 200 might be more correct) by just
outlucking the bot; David's analysis supports that view.
>I don't deny the skill needed to use the cube but I'll
>stick to my argument that it's external to backgammon
It isn't; it's been proven; so your argument doesn't exist.
In some positions, the best move is different depending on matchscore,
value of the cube and ownership of the cube.
--
Robert-Jan/Zorba
Murat Kalinyaprak
David Montgomery wrote news:7brocp$b...@krackle.cs.umd.edu...
>In <7br86u$d5...@taisp3.in-tch.com> Murat Kalinyaprak writes:
>>David Montgomery fed those matches through Snowie's
>>analyser and determined that the explanation was
>>much simpler (yet still outside of myself though:)
>>that I had simply gotten lucky... :)
>I'm pretty sure that I didn't say that you got lucky.
>I said you had better dice.
Let me quote a section from an article you wrote on
12/20/1998 under the thread "Re: JF rolls are not random"
"If an analysis showed that you were only equally lucky,
"less lucky, or not lucky enough to explain the discrepancy
"then I would be very interested to look into the matter
"further. Without this data, my belief is simply that you
"got lucky.
However, it's true that in the articles you posted
after having done the analysis, you explicitly used
the expression "having better dice". But, let's just
say that I/you/we/they had to use the term "lucky"
because SW calculates "luck rate" not "better dice
rate" and move on. I think we have bigger problems
than terminology here...
>There are a lot of ways that these two statements
>could be different. You have claimed that you often
>have foreknowledge of the rolls that are coming up,
>due to your ability to intuit the devious workings
>of the dice generators. If you know the dice coming
>up, you can move so that you get the better dice --
>but because it's based on some sort of knowledge or
>skill or intuition, or some other factor, it isn't lucky.
I'm really pleased to hear comments like these from
other people. I myself haven't made an issue out of
getting lucky/unlucky, especially regarding robots.
For me the issue regarding dice generators of some
robots or bg servers is with whether they are rigged
ot not rigged...
Yes, in the past, I had expressed that I could often
guess some dice rolls to come and make lesser (thus
unexpected by JF) moves to break patterns, etc. Your
emphasizing this using words like "knowledge" makes
me suspect that you may be insinuating that I was
peeking ahead into JF dice rolls...? This is pretty
easy to do of course, and there would certainly be no
less reason for my doing something like that than for
someone else rigging JF's, SW's, FIBS', etc. dice. :)
I suppose such a cheating could be detected rather
easily. For example, you could look for blunders I had
made that would be unexpected from the level of skill
you attribute (based on whatever means) to me and see
if they paid off immediately. If some did and others
didn't pay off immediately, all you can conclude is
that somebody succeded in fooling himself into making
some random hocus-pocus moves. It would take hours to
construct even one fake match in which such moves are
inserted in a way that they would start paying off
after a delay of a few turns. They would have to be
constructed vs. taking 100 matches that I won from a
batch of 500 because they are all consecutive starting
at a given seed/counter setting. There is still the
possibility that I might have spent weeks constructing
such matches as part of conducting a psychological
experiment in rgb, of course...:)
I'm not accusing you of accusing me or anything of the
sort but just elaborating on the topic because I had
myself wondered whether thos hocus-pocus moves had
actually helped or harmed me. I was going to look at
the 40 matches you emailed back after analyzing them
with SW but I haven't had time to do yet. However, in
the meantime there were other things talked about here
which made me opt for spending the time for re-playing
those 100 matches instead. This time, I wasn't trying
to make any hocus-pocus moves on purpose. I might have
made some without noticing though but excepting those,
any other errors or blunders would be just because I
couldn't come up with a better move at the time. I'll
still post a separate article about it because I want
to solicit advice from others on some specific things
but I guess I can at least tell you now that I won 61
this time. And believe me, it troubles my mind just as
much as it may other people's. Unfortunately, we may
not be able to come up with any better explanation for
it than "luck rate" (in SW's words:)...
MK
Murat Kalinyaprak
David Montgomery wrote news:7brnus$b...@krackle.cs.umd.edu...
>In <7bqpvn$d5...@taisp3.in-tch.com> Murat Kalinyaprak writes:
>>I just wanted to
>>know where you would draw the line with money involved.
>>A 55/45 ratio would translate to about 175(?) FIBS points
>>difference between my/JF's ratings, which is quite less
>> than the 300+
>As I mentioned, this sounds boring to me, so I want to
>make some money at it. It is an opportunity for you to
>make money, too, if you can continue to beat or play even
>with JF.
I have to admit that even the idea of it (i.e. gambling)
makes me nervous but nevertheless the idea has crossed my
mind quite a few times lately...
>But I have decided that the above spot was too much - not
>because JF wouldn't beat you enough, but because my time
>is worth more than that. So I hereby rescind that offer.
My time is precious too, yet the way to make it worthwhile
would be to not adjust the odds but to raise the stakes...
>Should you ever really want to play, send me email or call
>me and we can negotiate.
I'm not sure if I ever would want to play bg for money, so
I can't make promises but one never knows either...
MK
Donald Kahn
I have been reading this gentleman's postings for a number of months
now, and it puts me in mind of something attributed to Turing, to wit:
If you are unable to determine whether you are communicating with a
machine or with a human, then...(and I forget the conclusion - can
someone supply it?)
Only in KMK's case, it is absolutely impossible to tell whether or not
he is serious. Bravo for a great sustained performance!!!
dk